KR102479755B1 - heavy electric equipment preventive maintenance recommendation system based on real-time sensing data and method therefor - Google Patents

Abstract

The present invention relates to a heavy electric equipment preventive maintenance recommendation system based on real-time sensing data and, more specifically, to a heavy electric equipment preventive maintenance recommendation system based on real-time sensing data, which can solve a problem in which a diagnosis cycle lasts several months or years due to the characteristics of conventional heavy electric equipment, and there is a relatively small amount of learning data due to a relatively small amount of diagnostic data, which causing low accuracy, and a problem in which the interval between the diagnostic cycles is too great, which causing overfitting in trend analysis and deterioration prediction. The system learns a deterioration simulation model using diagnostic history data for the plurality of heavy electric equipment included in the same heavy electric equipment group as prediction target heavy electric equipment, and learns the deterioration simulation model by calculating a deterioration weight index based on sensor measurement result data of the plurality of heavy electric equipment included in the same heavy electric equipment group as the prediction target heavy electric equipment, thereby Providing higher degradation prediction timing accuracy.

Description

Heavy electric equipment preventive maintenance recommendation system based on real-time sensing data and method therefor}

The present invention relates to a technology for recommending preventive maintenance of heavy electrical equipment based on real-time sensing data. A plurality of heavy electric machine groups are created by grouping information, and data group by grouping a plurality of heavy electric machines having a relationship by receiving sensor measurement result data from at least one sensor and identification information of heavy electric machine for a plurality of heavy electric machines periodically. and calculates an equation having the form of a linear equation by performing regression analysis through a pre-stored regression model using diagnostic history data of a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted from the database. Based on the sensor measurement result data of a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted, according to the error trend with the optimal value set for each diagnostic item of the heavy electric machine to be predicted Preventive maintenance recommendation for heavy electrical equipment based on real-time sensing data that calculates the deterioration weight index (γ) of heavy electrical equipment to be predicted, reflects the degradation weight index to the degradation simulation model, generates preventive maintenance recommendation information through degradation simulation, and provides it to users Its purpose is to provide a system and its method.

Heavy electric machines such as high-voltage rotors and transformers are key components constituting power plants, and cause serious problems such as power generation stoppage in the event of a failure. In the event of insulation breakdown, not only restoration costs for the faulty motor but also loss of profit due to power supply disruption occurs, and in the event of a motor failure, not only the relevant heavy electric equipment but also peripheral facilities are lost.

Therefore, before the failure of such heavy electrical equipment, it is necessary to diagnose the abnormality in advance and predict the remaining life of the machine. A method of identifying and predicting the remaining life based on this is used, but this method has a problem in that it is difficult to accurately predict the remaining life.

With the conventional method, it is difficult to predict accurately, so the possibility of failure is not low. In case of failure, not only does it cause serious problems such as output depletion, stoppage of power generation, or damage to facilities, but it also takes about 20 days to rewind in case of motor failure. In the case of insulation breakdown in generator windings, the cost of operation loss during the recovery period is enormous enough to take about 6 months, such as new manufacturing and installation of windings. It is a trend.

In addition, as a result of performing the insulation judgment of heavy electric equipment through MTS (Mahalanobis Taguchi System) analysis using only the diagnosis history data determined as the normal state of the heavy electric equipment identical to the heavy electric equipment to be predicted among the diagnostic history data of the conventional heavy electric equipment Characteristically, the diagnosis cycle is several months or years, so only a relatively small number of diagnosis data exists, and the interval between diagnosis cycles is too wide, resulting in overfitting problems in trend analysis and degradation prediction, resulting in low accuracy. existed.

In the present invention, as a result of performing insulation determination of heavy electric equipment using conventional diagnostic history data, only a relatively small number of diagnostic data exists due to the characteristics of heavy electric equipment, as the diagnostic cycle is several months or years, and the interval between diagnostic cycles is In order to solve the difficulty that accuracy is low due to over-fitting problem in trend analysis and degradation prediction because it is too wide, diagnosis history data for a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted are used from the database. Higher degradation prediction timing accuracy by learning the degradation simulation model, calculating the degradation weight index based on the sensor measurement result data of a plurality of heavy electrical equipment included in the same heavy electrical equipment group as the heavy electrical equipment to be predicted, and learning the degradation simulation model Its purpose is to provide preventive maintenance recommendation technology for heavy electrical equipment based on real-time sensing data that generates preventive maintenance recommendation information that can be provided.

According to an embodiment of the present invention, in the heavy electric machine preventive maintenance recommendation system based on real-time sensing data, diagnostic history data and heavy electric machine identification information for a plurality of heavy electric machines generated by chronologically aligning diagnostic information for each heavy electric machine a database for storing sensor measurement data collected by at least one sensor; And a management server for generating preventive maintenance recommendation information by analyzing diagnosis history data, heavy electric machine identification information, and sensor measurement data for a plurality of heavy electric machines from the received database, wherein the database includes a data record file. Off-line data management unit for managing diagnosis history data and heavy electric machine identification information for a plurality of heavy electric machines by using or providing a data input form to receive direct input from a user; An online data management unit that receives and manages identification information of a plurality of heavy electric machines and sensor measurement result data from at least one or more sensors and periodically uploads them; And based on the heavy electric machine identification information, a data group is created by grouping a plurality of heavy electric machines having a correlation with respect to at least one item of manufacturer, manufacturing period, and motor capacity, and diagnosis is made for each heavy electric machine included in the data group. Further comprising a related data group generating unit for managing history data and sensor measurement result data, wherein the management server,

From the database, a plurality of data sets are generated by grouping a plurality of data included in the diagnostic history data of a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted, and a plurality of data sets are generated. A deterioration simulation model learning unit for learning a deterioration simulation model to calculate an equation having a form of a linear equation by performing a regression analysis through a pre-stored regression model for each data set using the included history data as learning data; Based on the sensor measurement result data of a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted from the database, deterioration of the heavy electric machine to be predicted according to the error trend with the optimal value set for each diagnostic item of the heavy electric machine to be predicted a degradation weight reflection unit that calculates a weight index (γ) and reflects the degradation weight index to a degradation simulation model; and performing linear regression analysis for each data set according to the degradation simulation model, generating degradation prediction information based on the performed results, and generating preventive maintenance recommendation information according to a degradation prediction time included in the degradation prediction information. A preventive maintenance recommendation information generation unit may be further included.

According to an embodiment of the present invention, the degradation simulation model learning unit generates a plurality of data sets by grouping a plurality of data included in the diagnosis history data, and the history data included in each generated data set. The deterioration simulation model may be trained to calculate an equation having a form of a linear equation by performing regression analysis through a pre-stored regression model for each data set using as learning data.

According to an embodiment of the present invention, the deterioration simulation model learning unit compares the diagnosis history data with the analysis result value through the equation calculated by the pre-stored regression model, so that the error between the diagnosis history data and the diagnosis history data is relatively small. A degradation simulation model composed of a plurality of regression analysis models may be designed by selecting a regression model having an analysis result value as a regression model of the corresponding data set.

According to an embodiment of the present invention, the deterioration weight reflection unit analyzes the sensor measurement result data to calculate an error with the optimal value set for each diagnosis item of the heavy electric machine to be predicted, and the larger the size of the error, the higher the value A weight exponent γ may be calculated and a weight may be applied to the equation as shown in Equation 2 and reflected in the degradation simulation model.

According to an embodiment of the present invention, the management server receives diagnostic history data pre-diagnosed in normal and abnormal states among diagnostic history data for a plurality of heavy electric machines as learning data, and constructs an insulation state determination model and a failure determination model. Further comprising a health analysis information generation unit for learning and generating health analysis information including failure prediction time information of the heavy electrical equipment to be predicted using the learned insulation state determination model and failure determination model, wherein the preventive maintenance recommendation information generation unit , the weight δ at the time of prediction of degradation included in the degradation prediction information, and the weight ε at the time of prediction of failure included in the health analysis information, and the preventive maintenance recommendation information using the weighted degradation prediction information and health analysis information can create

According to an embodiment of the present invention, the soundness analysis information generation unit is diagnosed in advance as a normal state and an abnormal state among diagnosis history data for a plurality of heavy electric devices included in the same heavy electric device group as the heavy electric machine to be predicted from the database. An insulation state determination model capable of determining the insulation state through distance average by receiving history data as learning data, setting a unit space, and calculating a distance average within the unit space for determining a normal state and an abnormal state. an insulation state determination model learning unit for learning; Among the diagnosis history data for a plurality of heavy electrical devices included in the same heavy electrical device group as the heavy electrical equipment to be predicted from the database, diagnosis history data pre-diagnosed in normal and abnormal states and insulation state information for each time point of the plurality of heavy electrical devices It is entered into the failure determination model to divide each heavy electrical device into system layers, and based on the diagnosis history data pre-diagnosed in normal and abnormal states, for each heavy electrical device, each time point for important measurement items of the function level required for each system layer. a failure determination model learning unit that derives a state value and learns a failure determination model for determining a failure prediction time point according to whether or not the state value at each time point of a functional level required for each derived system layer is satisfied; For each heavy electric machine derived from the failure determination model, a state value at each point in time of a plurality of functional levels required for each system layer is selected as an output variable, and a plurality of heavy electric machines included in the same heavy electric machine group as the received heavy electric machine to be predicted A model that analyzes diagnostic history data and output variables for , calculates a probability distribution, sets the calculated probability distribution as a prior distribution, and continuously updates the prior distribution for the output variables to a posterior distribution using a Bayesian approach. data learning execution unit; and diagnostic history data and heavy electric machine identification information of the heavy electrical equipment to be predicted are input to the insulation state determination model to receive insulation state information of the heavy electric equipment to be predicted at each time point, and insulation state information at each time point of the heavy electric equipment to be predicted. And it may further include a failure prediction time information generation unit for generating failure prediction time information by inputting the diagnosis history data of the heavy electrical equipment to be predicted into a failure determination model.

According to an embodiment of the present invention, the insulation state determination model learning unit inputs a plurality of major variable measurement data of diagnostic history data pre-diagnosed in a normal state and an abnormal state among the received diagnosis history data of heavy electrical equipment to an insulation determination model. Thus, a normal group is created based on the main variable measurement data of a plurality of diagnostic history data pre-diagnosed in a normal state among the input diagnostic history data of heavy electrical equipment, and normalization is performed for the diagnostic history data included in the normal group A probability distribution distance from the center point of the normal group to a plurality of diagnostic history data included in the normal group is measured as a unit distance, and a space in which the unit distance average of the measured unit distances is 1.0 is selected as the unit space. can do.

According to an embodiment of the present invention, the insulation state determination model learning unit measures the main variable measurement data of each diagnosis history data included in the normal group for the main variable measurement data of each diagnosis history data included in the normal group. A plurality of normalized vectors for each diagnosis history data may be calculated by performing normalization by dividing the value obtained by subtracting the average value by the standard deviation of the measurement data of the main variables of each diagnosis history data included in the normal group.

According to an embodiment of the present invention, the insulation state determination model learning unit applies an inverse matrix of a correlation matrix composed of correlation coefficients to the normalized vector to obtain a probability distribution distance with a plurality of diagnostic history data included in the normal group. can measure

According to an embodiment of the present invention, the insulation state determination model learning unit inputs a plurality of major variable measurement data of diagnostic history data pre-diagnosed in an abnormal state among the received diagnostic history data of heavy electrical equipment to an insulation determination model, and inputs Among the diagnostic history data of the received heavy electrical equipment, an abnormal group is created based on the main variable measurement data of a plurality of diagnostic history data previously diagnosed as abnormal, and the main variable measurement data of the plurality of diagnostic history data included in the abnormal group Normalize the abnormal group by dividing the average value of the main variable measurement data of each diagnosis history data included in the normal group by the standard deviation of the main variable measurement data of each diagnosis history data included in the normal group. A probability distribution distance with a plurality of diagnostic history data included in an abnormal group may be measured as a unit distance by calculating a vector and applying an inverse matrix of a correlation matrix composed of correlation coefficients to the plurality of normalized vectors.

According to an embodiment of the present invention, in the insulation state determination model learning unit, a distribution of a unit distance of a plurality of diagnosis history data included in the normal group and a unit distance of a plurality of diagnosis history data included in the abnormal group is set in advance. If the difference does not exceed the reference level, it is determined that the learning of the insulation status determination model is not completed, and the learning of the insulation status determination model may be repeated until the difference exceeds a predetermined reference level.

According to an embodiment of the present invention, the insulation state determination model learning unit calculates a difference in distribution between a unit distance of a plurality of diagnostic history data included in the normal group and a unit distance of a plurality of diagnostic history data included in the abnormal group. In order to increase the size, a plurality of predictive useful variables are selected among the main variables, and the insulation state determination model may be re-learned to re-select the unit space and unit distance of the normal group based on the selected predictive useful variables.

According to an embodiment of the present invention, the insulation state determination model learning unit arranges variables and signal factors in an inner array and an outer array using a two-level orthogonal array table to select the plurality of predictive useful variables, and each It is possible to calculate the unit distance for the main variable and derive the signal-to-noise ratio of the calculated value.

According to an embodiment of the present invention, the insulation state determination model learning unit determines whether each variable included in the plurality of predictive useful variables has predictive capability, and the signal-to-noise ratio of a unit distance calculated for each variable. The signal-to-noise ratio gain can be calculated using the difference in average.

According to an embodiment of the present invention, the insulation state determination model learning unit determines that the corresponding variable has no predictive ability when the SNR gain has a negative value,

If the SNR gain has a positive value, it is determined that there is predictive ability;

It can be determined that the higher the positive value, the higher the predictive ability.

According to an embodiment of the present invention, the heavy electric machine preventive maintenance recommendation method based on real-time sensing data includes diagnostic history data and heavy electric machine identification information for a plurality of heavy electric machines generated by chronologically aligning diagnostic information for each heavy electric machine, and at least one Storing the sensor measurement data collected by the sensor of the; And generating preventive maintenance recommendation information by analyzing diagnostic history data, heavy electric machine identification information, and sensor measurement data for a plurality of heavy electric machines from the input database, wherein the storing of the data includes data recording. Managing diagnosis history data and heavy electric machine identification information for a plurality of heavy electric machines by receiving direct input from a user by using a file or providing a data input form; Periodically uploading and managing sensor measurement result data from at least one sensor and identification information of a heavy electric machine for a plurality of heavy electric machines; Based on the heavy electric machine identification information, a data group is created by grouping a plurality of heavy electric machines having a correlation with respect to at least one item of manufacturer, manufacturing period, and electric machine capacity, and diagnosis history for each heavy electric machine included in the data group Further comprising managing data and sensor measurement result data, and generating the preventive maintenance recommendation information, the diagnosis history data of a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted from the database. A plurality of data sets are created by grouping a plurality of included data, and regression analysis is performed through a pre-stored regression model for each data set using the history data included in each generated data set as learning data. training the degradation simulation model to calculate an equation having a form of a first-order equation; Based on the sensor measurement result data of a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted from the database, deterioration of the heavy electric machine to be predicted according to the error trend with the optimal value set for each diagnostic item of the heavy electric machine to be predicted Calculating a weight index (γ) and reflecting the degradation weight index to a degradation simulation model; and

Performing linear regression analysis for each data set according to the degradation simulation model, generating degradation prediction information based on the performed result, and generating preventive maintenance recommendation information according to a degradation prediction time included in the degradation prediction information may further include.

According to an embodiment of the present invention, in the step of learning the degradation simulation model, a plurality of data sets are generated by grouping a plurality of data included in the diagnosis history data, and included in each generated data set. The deterioration simulation model may be trained to calculate an equation having a form of a linear equation by performing regression analysis through a pre-stored regression model for each data set using the historical data as learning data.

According to an embodiment of the present invention, the step of learning the deterioration simulation model may include comparing an analysis result value through the equation calculated by the pre-stored regression model and diagnosis history data so that an error between the diagnosis history data and the diagnosis history data is relatively A degradation simulation model composed of a plurality of regression analysis models may be designed by selecting a regression model having the least occurring analysis result value as a regression model of the corresponding data set.

According to an embodiment of the present invention, in the step of reflecting the deterioration simulation model, the sensor measurement result data is analyzed to calculate an error with an optimal value set for each diagnosis item of the heavy electric machine to be predicted, and the size of the error is A weighting exponent γ having a larger value may be calculated, and a weighted value may be applied to the equation as shown in Equation 2 and reflected in the deterioration simulation model.

In the generating of the preventive maintenance recommendation information, the diagnostic history data pre-diagnosed in normal and abnormal states among the diagnostic history data for a plurality of heavy electric machines is input as learning data, and an insulation state determination model and a failure determination model are learned, Further comprising the step of generating soundness analysis information including information about the time of prediction of failure of the heavy electrical equipment to be predicted using the learned insulation state determination model and failure determination model, wherein the generating of the preventive maintenance recommendation information comprises the step of generating the preventive maintenance recommendation information. Calculate the weight δ at the time of prediction of degradation included in the degradation prediction information and the weight ε at the time of prediction of failure included in the health analysis information, and generate the preventive maintenance recommendation information using the weighted degradation prediction information and health analysis information can do.

According to an embodiment of the present invention, the generating of the health analysis information may include, from the database, a normal state and an abnormal state of diagnostic history data for a plurality of heavy electric devices included in the same heavy electric device group as the heavy electric machine to be predicted from the database. Insulation state that can determine the insulation state through the distance average by receiving the diagnosed diagnosis history data as learning data, setting the unit space, and calculating the distance average within the unit space to determine the normal state and the abnormal state training a judgment model; Among the diagnosis history data for a plurality of heavy electrical devices included in the same heavy electrical device group as the heavy electrical equipment to be predicted from the database, diagnosis history data pre-diagnosed in normal and abnormal states and insulation state information for each time point of the plurality of heavy electrical devices It is entered into the failure determination model to divide each heavy electrical device into system layers, and based on the diagnosis history data pre-diagnosed in normal and abnormal states, for each heavy electrical device, each time point for important measurement items of the function level required for each system layer. deriving a state value and learning a failure determination model for determining a failure prediction time point according to whether or not the state value for each time point of a function level required for each derived system layer is satisfied; For each heavy electric machine derived from the failure determination model, a state value at each point in time of a plurality of functional levels required for each system layer is selected as an output variable, and a plurality of heavy electric machines included in the same heavy electric machine group as the received heavy electric machine to be predicted Analyzing diagnostic history data and output variables for , calculating a probability distribution, setting the calculated probability distribution as a prior distribution, and continuously updating the prior distribution for the output variable as a posterior distribution using a Bayesian approach. ; and diagnostic history data and heavy electric machine identification information of the heavy electrical equipment to be predicted are input to the insulation state determination model to receive insulation state information of the heavy electric equipment to be predicted at each time point, and insulation state information at each time point of the heavy electric equipment to be predicted. and generating failure prediction time information by inputting diagnostic history data of the heavy electrical equipment to be predicted into a failure determination model.

According to an embodiment of the present invention, the step of learning the insulation state determination model is to determine the insulation of a plurality of major variable measurement data of diagnostic history data pre-diagnosed in normal and abnormal states among the input diagnostic history data of heavy electrical equipment. By inputting into the model,

A normal group is created based on the main variable measurement data of a plurality of diagnostic history data pre-diagnosed in a normal state among the input diagnostic history data of heavy electrical equipment, and normalization is performed for the diagnostic history data included in the normal group. A probability distribution distance from the center point of the normal group to a plurality of diagnostic history data included in the normal group may be measured as a unit distance, and a space in which the unit distance average of the measured unit distances is 1.0 may be selected as the unit space. there is..

According to an embodiment of the present invention, the step of learning the insulation state determination model may include measuring the main variable of each diagnostic history data included in the normal group for the main variable measurement data of each diagnosis history data included in the normal group. A value obtained by subtracting the average value of the measured data is divided by the standard deviation of the main variable measured data of each diagnostic history data included in the normal group to perform normalization, thereby calculating a plurality of normalized vectors for each diagnostic history data.

According to an embodiment of the present invention, in the step of learning the insulation state determination model, a plurality of diagnostic history data included in the normal group is obtained by applying an inverse matrix of a correlation matrix composed of correlation coefficients to the plurality of normalized vectors. Probability distribution distance can be measured.

According to an embodiment of the present invention, the step of learning the insulation state determination model includes inputting a plurality of major variable measurement data of diagnostic history data pre-diagnosed in an abnormal state among the received diagnostic history data of heavy electrical equipment to the insulation determination model. Therefore, an abnormal group is created based on the main variable measurement data of a plurality of diagnostic history data pre-diagnosed in an abnormal state among the input diagnostic history data of heavy electrical equipment, and the main variables of the plurality of diagnostic history data included in the abnormal group The abnormal group is normalized by dividing the value obtained by subtracting the average value of the main variable measurement data of each diagnosis history data included in the normal group from the measured data by the standard deviation of the main variable measurement data of each diagnosis history data included in the normal group. Calculate a normalized vector of , and apply the inverse matrix of the correlation matrix composed of correlation coefficients to the plurality of normalized vectors to measure the probability distribution distance with a plurality of diagnostic history data included in the abnormal group as a unit distance. can

According to an embodiment of the present invention, the step of learning the insulation state determination model may include distribution of unit distances of a plurality of diagnostic history data included in the normal group and unit distances of a plurality of diagnostic history data included in the abnormal group. If there is no difference greater than a preset criterion, it is determined that learning of the insulation state determination model is not completed, and learning of the insulation state determination model may be repeated until a difference greater than a preset criterion is obtained.

According to an embodiment of the present invention, the step of learning the insulation state determination model may include distribution of unit distances of a plurality of diagnostic history data included in the normal group and unit distances of a plurality of diagnostic history data included in the abnormal group. In order to increase the difference in , a plurality of predictive useful variables are selected among the main variables, and the insulation state judgment model can be re-learned to re-select the unit space and unit distance of the normal group based on the selected predictive useful variables. .

According to an embodiment of the present invention, in the step of learning the insulation state determination model, variables and signal factors are arranged in an inner array and an outer array using a two-level orthogonal array table in order to select the plurality of predictive useful variables. , calculate the unit distance for each major variable, and derive the signal-to-noise ratio of the calculated value.

According to an embodiment of the present invention, the step of learning the insulation state determination model is a unit distance calculated for each variable in order to determine whether each variable included in the plurality of predictive useful variables has predictive capability. The signal-noise ratio gain can be calculated using the difference between the average signal-noise ratio of .

According to an embodiment of the present invention, in the step of learning the insulation state determination model, when the SNR gain has a negative value, it is determined that the corresponding variable has no predictive ability, and the SNR gain is a positive value. If it has, it is determined that it has predictive ability, and it can be determined that the predictive ability is high as the positive value increases.

The real-time sensing data-based preventive maintenance recommendation system for heavy electrical equipment implemented according to the embodiment of the present invention has a diagnosis cycle of several months or years due to the characteristics of conventional heavy electrical equipment, so there is only a relatively small number of diagnostic data, so learning data It overcomes the disadvantages of over-fitting problems in trend analysis and deterioration prediction because the accuracy is low due to relatively small existence and the interval between diagnosis cycles is too wide. By learning a deterioration simulation model using the diagnostic history data for the predicted target heavy electric machine and learning a deterioration simulation model by calculating a deterioration weight index based on sensor measurement result data of a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted. It is possible to provide an effect capable of providing high degradation prediction timing accuracy.

1 is a block diagram of a heavy electrical equipment preventive maintenance recommendation system based on real-time sensing data implemented according to an embodiment of the present invention.
2 is a diagram showing detailed configurations of a database and a management server included in a system for recommending preventive maintenance of heavy electrical equipment based on real-time sensing data according to a first embodiment of the present invention.
3 is a diagram showing detailed configurations of a database and a management server included in a system for recommending preventive maintenance of heavy electrical equipment based on real-time sensing data according to a second embodiment of the present invention.
4 is a detailed configuration diagram of a soundness analysis information generation unit shown in FIG. 2 .
5 is a diagram illustrating a method of performing an update using a Bayesian approach in the model data learning performer disclosed in FIG. 4 .
6 is a diagram illustrating a flow of data for generating degradation prediction information generated using a regression model selected for each data set using a degradation simulation model according to an embodiment of the present invention.
8 is a diagram illustrating a process of calculating an average distance within a unit space for determining a normal state and an abnormal state in order to train an insulation state determination model according to an embodiment of the present invention.
8 is a flowchart of a method for recommending preventive maintenance of heavy electrical equipment based on real-time sensing data according to an embodiment of the present invention.

Hereinafter, with reference to the accompanying drawings, embodiments of the present invention will be described in detail so that those skilled in the art can easily carry out the present invention. However, the present invention may be implemented in many different forms and is not limited to the embodiments described herein.

Terms used in the present invention are only used to describe specific embodiments, and are not intended to limit the present invention. Singular expressions include plural expressions unless the context clearly dictates otherwise.

In the present invention, terms such as "comprise" or "having" are intended to designate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, but one or more other features It should be understood that the presence or addition of numbers, steps, operations, components, parts, or combinations thereof is not precluded.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which the present invention belongs.

Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related art, and unless explicitly defined in the present invention, they should not be interpreted in an ideal or excessively formal meaning. don't

It will also be understood that combinations of each block of the drawings and flowchart drawings can be performed by computer program instructions, and these computer program instructions can be loaded into a processor of a general-purpose computer, special-purpose computer, or other programmable data processing equipment. Thus, those instructions executed by a processor of a computer or other programmable data processing equipment create means for performing the functions described in the flowchart block(s).

These computer program instructions may also be stored in a computer usable or computer readable memory that can be directed to a computer or other programmable data processing equipment to implement functionality in a particular way, such that the computer usable or computer readable memory The instructions stored in are also capable of producing an article of manufacture containing instruction means that perform the functions described in the flowchart block(s).

The computer program instructions can also be loaded on a computer or other programmable data processing equipment, so that a series of operational steps are performed on the computer or other programmable data processing equipment to create a computer-executed process to generate computer or other programmable data processing equipment. Instructions for performing processing equipment may also provide steps for performing the functions described in the flowchart block(s).

Additionally, each block may represent a module, segment, or portion of code that includes one or more executable instructions for executing specified logical function(s).

And it should be noted that in some alternative embodiments it is also possible for the functions mentioned in the blocks to occur out of order. For example, two blocks shown in succession may in fact be executed substantially concurrently, or the blocks may sometimes be executed in reverse order depending on their function.

At this time, the term '~unit' used in this embodiment means software or a hardware component such as a field-programmable gate array (FPGA) or application specific integrated circuit (ASIC), and what role does '~unit' have? perform them

However, '~ part' is not limited to software or hardware. '~bu' may be configured to be in an addressable storage medium and may be configured to reproduce one or more processors.

Therefore, as an example, '~unit' refers to components such as software components, object-oriented software components, class components, and task components, processes, functions, properties, and procedures. , subroutines, segments of program code, drivers, firmware, microcode, circuitry, data, databases, data structures, tables, arrays, and variables. Functions provided within components and '~units' may be combined into smaller numbers of components and '~units' or further separated into additional components and '~units'. In addition, components and '~units' may be implemented to play one or more CPUs in a device or a secure multimedia card.

In describing the embodiments of the present invention in detail, an example of a specific system will be the main target, but the main subject matter to be claimed in this specification extends the scope disclosed herein to other communication systems and services having a similar technical background. It can be applied within a range that does not deviate greatly, and this will be possible with the judgment of those skilled in the art.

Hereinafter, a heavy electrical equipment preventive maintenance recommendation system and method based on real-time sensing data according to an embodiment of the present invention will be described with reference to the drawings.

1 is a block diagram of a heavy electrical equipment preventive maintenance recommendation system based on real-time sensing data implemented according to an embodiment of the present invention.

Referring to FIG. 1, a system for recommending preventive maintenance of heavy electrical equipment based on real-time sensing data is shown, and the system for recommending preventive maintenance of heavy electric equipment based on real-time sensing data may include a database 100 and a management server 2000.

According to an embodiment of the present invention, a heavy electrical equipment preventive maintenance recommendation system based on real-time sensing data may be driven using at least one processor.

The database 100 may store diagnosis history data and heavy electric machine identification information and sensor measurement data collected by at least one sensor for a plurality of heavy electric machines generated by chronologically arranging diagnostic information for each heavy electric machine.

Here, the heavy electric machine may mean large-capacity industrial equipment or industrial machinery used in industrial sites and power plants.

According to an embodiment of the present invention, diagnosis history data may refer to information generated as one piece of information by arranging diagnosis information including diagnosis results performed at a plurality of specific points in time according to the lapse of time.

According to one embodiment of the present invention, the heavy electric machine identification information may mean a unique serial number possessed by the heavy electric machine for identifying the heavy electric machine or an ID number randomly designated on the database, regardless of the format, to identify the heavy electric machine. Anything that can be used can be used without limitation.

In addition, the heavy electric machine identification information may include information specific to the electric machine, such as the manufacturer of the heavy electric machine, the manufacturing period, and the capacity of the electric machine, by matching the serial number or ID number of the heavy electric machine.

The management server 2000 may generate preventive maintenance recommendation information by analyzing diagnosis history data, heavy electric machine identification information, and sensor measurement data for a plurality of heavy electric machines from the input database 1000 .

The database 1000 and the management server 2000 will be described in more detail with reference to FIGS. 2 and 3 .

2 is a diagram showing detailed configurations of a database and a management server included in a system for recommending preventive maintenance of heavy electrical equipment based on real-time sensing data according to a first embodiment of the present invention.

Referring to FIG. 2 , the database 1000 included in the system for recommending preventive maintenance of heavy electrical equipment based on real-time sensing data according to the first embodiment of the present invention includes an offline data management unit 1100, an online data management unit 1200, and a related data group. A generator 1300 may be further included.

The offline data management unit 1100 may receive direct input from a user by using a data record file or by providing a data input form to manage diagnosis history data and heavy electric machine identification information for a plurality of heavy electric machines.

According to an embodiment of the present invention, the offline data management unit 1100 may receive diagnosis history data and heavy electric machine identification information for a plurality of heavy electric machines directly from a user or in a file format having a certain form.

The online data management unit 1200 may receive and manage identification information of a plurality of heavy electric machines and sensor measurement result data periodically uploaded from at least one sensor.

According to an embodiment of the present invention, the online data management unit 1200 collects measurement data of a plurality of measurement items for a plurality of heavy electric machines from a plurality of sensors connected to the heavy electric machine preventive maintenance recommendation system according to a preset cycle, and creates a database can be uploaded to

According to an embodiment of the present invention, a plurality of sensors and the database 1000 may be connected using a wireless network environment, but may use a wired network environment.

According to an embodiment of the present invention, sensor measurement result data for each inspection item for the heavy electric machine may be generated by receiving the measurement values of the heavy electric machine inspection items from the sensor installed in the heavy electric machine and aligning them in time series.

The related data group creation unit 1300 creates a data group by grouping a plurality of heavy electric devices having a correlation with respect to at least one item of manufacturer, manufacturing period, and motor capacity based on heavy electric machine identification information, and generates a data group, Diagnosis history data and sensor measurement result data can be managed for each included heavy electric device.

According to the above embodiment, in performing grouping, it is possible to group a plurality of electric devices that achieve the conditions of the same manufacturer, manufacturing time within a certain period, and electric device capacity within a certain period into one group, but are not limited thereto, and manufacturers and manufacturers are not limited thereto. Any criterion for creating a group using the above conditions, such as weighting according to the number of conditions satisfied when a plurality of conditions are satisfied among time and capacity of the motor, can be used without limitation.

In addition, the management server 2000 included in the system for recommending preventive maintenance of heavy electrical equipment based on real-time sensing data according to the first embodiment of the present invention includes a deterioration simulation model learning unit 2100, a deterioration weight reflection unit 2200, and a preventive maintenance recommendation. An information generator 2300 may be included.

The degradation simulation model learning unit 2100 generates a plurality of data sets by performing grouping on a plurality of data included in diagnosis history data of a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted from the database. And, using the history data included in each generated data set as training data, regression analysis is performed through a pre-stored regression model for each data set to learn a degradation simulation model to calculate an equation having the form of a linear equation. can

According to an embodiment of the present invention, the degradation simulation model learning unit 2100 generates a plurality of data sets by grouping a plurality of data included in the diagnosis history data, and records the history included in each generated data set. The degradation simulation model may be trained to calculate an equation having a form of a linear equation by performing regression analysis through a pre-stored regression model for each data set using the data as learning data.

According to an embodiment of the present invention, an equation representing a plurality of data included in the diagnosis history data may be expressed in the form of a linear equation (y = ax + b), and the linear equation calculated in this form is used. Thus, a solution of an equation representing a plurality of data can be calculated.

According to an embodiment of the present invention, the degradation simulation model learning unit 2100 generates a plurality of data sets by grouping a plurality of data included in the diagnosis history data, and records the history included in each generated data set. Linear regression analysis can be performed through pre-stored regression models for each data set by using the data as learning data.

According to an embodiment of the present invention, the deterioration simulation model learning unit 2100 constructs a polynomial for the data included in the input diagnosis history data of the heavy electric machine to be predicted, performs parameter estimation, and predicts the tendency of the parameter-estimated data. According to , an equation expressed as a straight line can be calculated.

According to an embodiment of the present invention, the polynomial formed for the data included in the diagnosis history data of heavy electrical equipment may follow the Dx3+Cx2+Bx+A format, but is not limited thereto, and the polynomial format in which parameter estimation can be performed If so, it can be used without limitation.

According to an embodiment of the present invention, random sampling may be performed on the polynomial in performing parameter estimation by constructing a polynomial targeting the data included in the received diagnosis history data of the heavy electric machine to be predicted. Instead of performing random sampling, a method of estimating parameters immediately and estimating data after a specific period of time after the final measurement may be used.

According to an embodiment of the present invention, when using a method of estimating data during the course of a specific institution after parameter estimation and final measurement, data may be expressed in the form of a quadratic function, and in fact, including the first measurement result and the last measurement result In the case of using parameters estimated from data, verification using past data showed high accuracy in estimating deterioration in the next two years.

According to an embodiment of the present invention, in performing parameter estimation by constructing a polynomial targeting the data included in the input diagnosis history data of the heavy electric machine to be predicted, after randomly selecting and modeling the data based on the colony, the suitability It is possible to relatively improve the accuracy of estimation by repeating the process of performing random selection after going through the estimation process to select the most suitable model.

The deterioration simulation model learning unit 2100 applies a weight reflecting at least one or more information of the frequency of stop events of the heavy electric machine, the driving pattern of the heavy electric machine, and the installation environment to the equation, and predicts the deterioration time of the heavy electric machine using the equation. Deterioration prediction information may be generated by inputting a plurality of data included in the diagnosis history data to the degradation simulation model.

According to an embodiment of the present invention, the degradation simulation model learning unit 2100 calculates in the degradation simulation model learning unit 2100 based on at least one or more information of the frequency of occurrence of stop events of heavy electric machines, driving patterns of heavy electric machines, and installation environments Weights can be applied to an equation.

According to an embodiment of the present invention, the equation calculated by the degradation simulation model learning unit 2100 may be expressed in the form of a linear equation (y = ax + b), and according to this embodiment, the stop event of the heavy electric machine occurs The weight index β can be calculated based on at least one information of frequency, heavy electric equipment operation pattern, and installation environment, and the calculated β can be reflected in the linear equation as shown in Equation 1 below to apply a weight to the equation.

According to an embodiment of the present invention, the weight index β may be calculated according to one condition among the frequency of stop events of heavy electric equipment, the driving pattern of heavy electric equipment, and the installation environment, but is not limited thereto and may be calculated according to a plurality of conditions. When calculated according to each condition, a weight is applied according to each condition, and the absolute value can be calculated in a form that increases or decreases.

According to one embodiment of the present invention, a plurality of data included in the diagnosis history data is input into a deterioration simulation model that predicts the deterioration point of heavy electrical equipment including an equation with weights applied thereto, and according to the trend trend indicated by the graph according to the equation Deterioration time prediction or the degradation degree of the heavy electrical equipment may be calculated according to the predicted degradation time point, and the degradation prediction information may include degradation time prediction information and/or deterioration degree information of the heavy electrical equipment.

According to an embodiment of the present invention, the degradation simulation model learning unit 2100 performs parameter estimation on the data included in the diagnosis history data of the heavy electric machine to be predicted in the degradation simulation model learning unit 2100 to perform parameter estimation data. When an equation expressed as a straight line is calculated according to the tendency of Deterioration time of heavy electrical equipment can be predicted by inputting parameter estimated data to the degradation simulation model that predicts the time of deterioration.

According to the above embodiment, the deterioration time of the heavy electric machine may be predicted by inputting the parameter-estimated data to the deterioration simulation model for predicting the deterioration time of the heavy electric machine including the weighted equation.

According to an embodiment of the present invention, when receiving diagnosis history data of a heavy electric machine classified into the same group as a heavy electric machine to be predicted, the degradation simulation model learning unit 2100 includes the diagnosis history data of the heavy electric machine to be predicted It is possible to perform parameter estimation by constructing a polynomial targeting the data included in the plurality of data and the diagnostic history data of the heavy electric machine classified into the same group received, and at this time, the degradation simulation model learning unit 2100 is the heavy electric machine A weight that reflects at least one information of stop event occurrence frequency, heavy electrical equipment operation pattern, and installation environment is applied to the equation, and parameter-estimated data is input into a degradation simulation model that predicts the deterioration point of heavy electrical equipment using the equation The deterioration point of the device can be predicted.

According to an embodiment of the present invention, the degradation simulation model learning unit 2100 compares the diagnosis history data with the analysis result value through the equation calculated by the pre-stored regression model so that the error between the diagnosis history data and the diagnosis history data is relatively minimized. A degradation simulation model composed of a plurality of regression analysis models may be designed by selecting a regression model having generated analysis result values as a regression model of the corresponding data set.

According to an embodiment of the present invention, the degradation simulation model learning unit 2100 generates a plurality of data sets by grouping a plurality of data included in the diagnosis history data, and records the history included in each generated data set. When performing regression analysis through a pre-stored regression model for each data set using data as training data, a model with the smallest error is selected by comparing the analysis result value and the historical data among the pre-stored regression models. A degradation simulation model composed of a plurality of regression analysis models may be generated by selecting a set of regression models, linear regression analysis is performed for each of the plurality of data sets according to the degradation simulation model, and degradation is predicted based on the performed results. information can be generated.

According to an embodiment of the present invention, the deterioration simulation model learning unit 2100 compares the analysis result value and diagnosis history data among pre-stored regression models, and selects a model with a relatively least error as a regression model of the corresponding data set. It is possible to generate a degradation simulation model composed of a plurality of regression analysis models.

According to an embodiment of the present invention, in order to select a regression model for each data set, a pre-stored regression model is applied to each data set to derive an analysis result value, and data on the derived analysis result value and actual diagnosis history data A regression model that derives an analysis result value with the smallest error by comparing the values may be selected as the regression model of the corresponding data set.

According to an embodiment of the present invention, the degradation simulation model may be formed of regression models selected for each data set, and diagnostic history data is input to the formed degradation simulation model, and the degradation simulation model is generated by comparing output values and diagnostic history data. can be verified

According to an embodiment of the present invention, the degradation simulation model learning unit 2100 uses a bagging technique in performing linear regression analysis to estimate an equation representing a plurality of data included in diagnostic history data, resulting in measurement error and noise. In order to relatively reduce the influence of , a degradation simulation model may be generated by selecting a regression analysis model with the smallest error for each generated data set.

According to an embodiment of the present invention, the degradation simulation model learning unit 2100 may perform linear regression analysis for each data set according to the degradation simulation model and generate degradation prediction information based on the performed result.

According to an embodiment of the present invention, the degradation simulation model learning unit 2100 may generate degradation prediction information according to the degradation simulation model, and when a failure occurs because degradation proceeds before the degradation prediction information according to the degradation simulation model Depending on the type and cause of the failure, the deterioration prediction criteria of the deterioration simulation model can be automatically reflected and applied to the prediction of deterioration of heavy electrical equipment of the same type.

The deterioration weight reflection unit 2200 is based on the sensor measurement result data of a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted from the database, based on the error trend with the optimal value set for each diagnosis item of the heavy electric machine to be predicted. Accordingly, the deterioration weight index (γ) of the heavy electric machine to be predicted may be calculated, and the deterioration weight index may be reflected in the deterioration simulation model.

According to an embodiment of the present invention, the deterioration weight reflection unit 2200 is configured for each diagnostic item of the heavy electric machine to be predicted by analyzing sensor measurement result data of a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted. It is possible to calculate the error between the optimal value and the measurement value for each diagnosis item of the heavy electrical equipment to be predicted included in the actually measured sensor measurement result data, and calculate the weight index γ using the calculated error value.

According to an embodiment of the present invention, as a method of calculating an optimal value set for each diagnosis item of a heavy electric machine to be predicted, the degree of similarity with the heavy electric machine to be predicted is calculated for each of a plurality of heavy electric machines according to a predetermined criterion, and each heavy electric machine is diagnosed. A method of calculating the average value after summing the measured values included in the sensor measurement result data for each item with the similarity applied as a weighted value can be used, but it is not limited to this, and a method of presetting the optimal value for each diagnostic item of the heavy electrical equipment to be predicted available.

According to an embodiment of the present invention, as a method of calculating the weight index γ, an average value of error values calculated for each diagnostic item may be obtained, and the average value may be calculated based on the average value, but is not limited thereto. A method of calculating the average value after summing the values to which the weights determined according to the result are applied may be used.

According to an embodiment of the present invention, the deterioration weight reflection unit 2200 calculates Equation 2 using the weight index γ calculated by analyzing sensor measurement result data of a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted. Deterioration prediction information can be generated by giving a high weight (γ) to the equations included in the degradation simulation.

The preventive maintenance recommendation information generating unit 2300 performs linear regression analysis for each data set according to the degradation simulation model, generates degradation prediction information based on the performed result, and determines the degradation prediction time included in the degradation prediction information. Accordingly, preventive maintenance recommendation information can be generated.

3 is a diagram showing detailed configurations of a database and a management server included in a system for recommending preventive maintenance of heavy electrical equipment based on real-time sensing data according to a second embodiment of the present invention.

Referring to FIG. 3, detailed configurations of a database and a management server included in the real-time sensing data-based preventive maintenance recommendation system for heavy electrical equipment according to a second embodiment of the present invention are shown, and the management server 2010 is a degradation simulation model learning unit. 2100, a deterioration weight reflection unit 2200, a preventive maintenance recommendation information generator 2300, and a soundness analysis information generator 2400 may be further included.

The soundness analysis information generation unit 2400 receives diagnostic history data pre-diagnosed in normal and abnormal states among diagnostic history data for a plurality of heavy electric machines as learning data, learns an insulation state determination model and a failure determination model, and learns It is possible to generate soundness analysis information including failure prediction time point information of the heavy electrical equipment to be predicted using the insulation state determination model and the failure determination model.

The soundness analysis information generator 2400 will be described in detail with reference to FIG. 4 .

According to the second embodiment of the present invention, the preventive maintenance recommendation information generation unit 2300 calculates a weight δ at the time of prediction of deterioration included in the deterioration prediction information and a weight ε at the time of prediction of failure included in the health analysis information, and the weight It is possible to generate preventive maintenance recommendation information by using deterioration prediction information and soundness analysis information applied.

According to an embodiment of the present invention, the preventive maintenance recommendation information generating unit 2300 calculates a weight δ at the time of prediction of deterioration included in the deterioration prediction information and a weight ε at the time of prediction of failure included in the soundness analysis information, and for each time point Preventive maintenance recommendation information recommending preventive maintenance may be generated based on the time point at which the average value of the weighted values is obtained.

According to an embodiment of the present invention, the weight δ and the weight ε can be initially assigned the same as 1, and it is determined whether the time information output by inputting the diagnosis history data matches the failure history data that occurred in the past. The weights can be calculated by performing an update by adjusting the ratio of the weight δ and the weight ε, respectively, in a direction in which accuracy increases relatively.

4 is a detailed configuration diagram of a soundness analysis information generation unit shown in FIG. 2 .

Referring to FIG. 4, according to an embodiment of the present invention, the health analysis information generating unit 2400 includes an insulation state determination model learning unit 2410, a failure determination model learning unit 2420, a model data learning execution unit 2430, A failure prediction time information generation unit 2440 may be included.

The insulation state determination model learning unit 2410 converts diagnosis history data pre-diagnosed in normal and abnormal states among diagnosis history data for a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted from the database as learning data. It is possible to learn an insulation state determination model capable of determining an insulation state through a distance average by setting a unit space after receiving an input and calculating a distance average within the unit space for determining a normal state and an abnormal state.

According to an embodiment of the present invention, a unit distance determination method may be used as a multivariate data mining technique for a plurality of data included in diagnosis history data.

Here, the unit distance determination method may refer to a technique of determining a state by setting a unit space based on the average value of a group and measuring the unit distance how far a new observation value is from the unit space.

According to an embodiment of the present invention, according to an embodiment using the unit distance determination method, after setting the unit space of the normal group and expressing each individual data in the unit space based on the center point of the normal group, how far is it from the center point of the normal group? Whether or not they are apart is calculated as a unit distance, and the greater the unit distance, the higher the probability that the group is not a normal group. If it exceeds a preset threshold value, it can be determined as an abnormal group.

According to an embodiment of the present invention, the insulation state determination model learning unit 2410 receives diagnostic history data for a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted received from the user from the database 100. can receive

According to an embodiment of the present invention, the insulation state determination model learning unit 2410 is a heavy electric machine that has been diagnosed as normal according to criteria previously diagnosed as normal or abnormal among the received diagnosis history data for a plurality of heavy electric machines. A distance average within a unit space can be calculated using diagnostic history data for .

According to an embodiment of the present invention, the insulation state determination model learning unit 2410 is based on the average value of a plurality of major variable measurement data of previously diagnosed diagnostic history data in a normal state among the received diagnostic history data of heavy electrical equipment as a unit. space can be set.

According to an embodiment of the present invention, a normal group is set based on diagnostic history data pre-diagnosed in a normal state among input diagnostic history data of heavy electrical equipment, and a plurality of major variables of the diagnostic history data pre-diagnosed in a normal state A unit space based on the average value of the measured data can be derived.

According to an embodiment of the present invention, the main variable data included in the diagnosis history data for a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted is expressed in the reference space, and each individual data expressed and the reference space The unit distance can be measured by measuring the probability distribution distance from the center point of the normal group.

According to an embodiment of the present invention, the insulation state judgment model learning unit 2410 converts a plurality of major variable measurement data of diagnostic history data pre-diagnosed into normal and abnormal states among the received diagnosis history data of heavy electrical equipment into an insulation determination model. It is possible to create a normal group based on the main variable measurement data of a plurality of diagnostic history data pre-diagnosed in a normal state among the input diagnostic history data of heavy electrical equipment.

According to the above embodiment, the insulation state determination model learning unit 2410 normalizes diagnosis history data included in the normal group, and compares the diagnosis history data included in the normal group with a plurality of diagnosis history data included in the normal group from the center point of the normal group. The probability distribution distance is measured as a unit distance, and a space in which the unit distance average of the measured unit distances is 1.0 may be selected as the unit space.

According to an embodiment of the present invention, the insulation state determination model learning unit 2410 measures the main variable measurement data of each diagnosis history data included in the normal group for the main variable measurement data of each diagnosis history data included in the normal group. A plurality of normalized vectors for each diagnosis history data may be calculated by performing normalization by dividing the value obtained by subtracting the average value by the standard deviation of the measurement data of the main variables of each diagnosis history data included in the normal group.

According to an embodiment of the present invention, the insulation state determination model learning unit 2410 is diagnostic history data pre-diagnosed in a normal state in order to derive a unit space having an average distance of 1.0 from the center point of a normal group to be selected as a unit space. Normalization may be performed on the measurement data of a plurality of major variables included in .

According to the above embodiment, as shown in Equation 3, normalization is the standard of the main variable measurement data of each diagnosis history data included in the normal group by subtracting the average value of the main variable measurement data of each diagnosis history data included in the normal group. This can be done by dividing by the deviation.

는 i번째 변수의 j번째 관측치에서의 값,

는 i번째 변수의 평균,

는 i번째 변수의 표준편차를 의미할 수 있다.where i is the number of variables (i=1, 2, …, k), j is the number of observations (j=1, 2, …, n),

is the value at the jth observation of the ith variable,

is the mean of the ith variable,

May mean the standard deviation of the i-th variable.

According to an embodiment of the present invention, the insulation state determination model learning unit 2410 applies an inverse matrix of a correlation matrix composed of correlation coefficients to a plurality of normalized vectors, thereby determining a plurality of diagnosis history data included in a normal group. Probability distribution distance can be measured.

According to an embodiment of the present invention, the insulation state determination model learning unit 2410 inputs a plurality of major variable measurement data of diagnostic history data pre-diagnosed as an abnormal state among the received diagnostic history data of heavy electrical equipment to an insulation determination model. can

According to an embodiment of the present invention, the insulation state determination model learning unit 2410 is a heavy electric machine that has been diagnosed as abnormal according to criteria previously diagnosed as normal or abnormal among the received diagnostic history data for a plurality of heavy electric machines. A distance average within a unit space can be calculated using diagnostic history data for .

According to an embodiment of the present invention, a normal group is set based on diagnostic history data pre-diagnosed in a normal state among input diagnostic history data of heavy electrical equipment, and a plurality of major variables of the diagnostic history data pre-diagnosed in an abnormal state. A unit space based on the average value of the measured data can be derived.

According to an embodiment of the present invention, the main variable data included in the diagnosis history data for a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted is expressed in the reference space, and each individual data expressed and the reference space The unit distance can be measured by measuring the probability distribution distance from the center point of the abnormal group.

According to the above embodiment, the insulation state determination model learning unit 2410 generates an abnormal group based on the main variable measurement data of a plurality of diagnostic history data pre-diagnosed as abnormal among the received diagnostic history data of heavy electrical equipment, The value obtained by subtracting the average value of the main variable measurement data of each diagnosis history data included in the normal group from the main variable measurement data of multiple diagnosis history data included in the abnormal group. The main variable measurement data of each diagnosis history data included in the normal group. A plurality of normalized vectors can be calculated by normalizing the abnormal group by dividing by the standard deviation of .

According to the above embodiment, the insulation state determination model learning unit 2410 applies an inverse matrix of a correlation matrix composed of correlation coefficients to a plurality of normalized vectors to obtain a probability distribution with a plurality of diagnosis history data included in the abnormal group. Distance can be measured in units of distance.

According to an embodiment of the present invention, the normalized vector (Z) calculates the unit distance of the normal group using the inverse matrix of the correlation matrix (C) composed of correlation coefficients, and the formula for this is as shown in Equation 4.

는

의 정규화 된 벡터,

는

의 전치행렬,

는 상관행렬의 역행렬을 의미할 수 있다.where k is the number of variables,

Is

The normalized vector of ,

Is

transpose matrix of ,

may mean an inverse matrix of the correlation matrix.

According to an embodiment of the present invention, in the insulation state determination model learning unit 2410, the distribution of unit distances of a plurality of diagnostic history data included in the normal group and unit distances of a plurality of diagnostic history data included in the abnormal group is set in advance. If the difference does not exceed the reference level, it is determined that the learning of the insulation condition determination model is not completed, and the learning of the insulation condition determination model may be repeated until the difference exceeds a preset reference level.

According to an embodiment of the present invention, the insulation state determination model learning unit 2410 compares the distribution of unit distances of a plurality of diagnostic history data included in the normal group and unit distances of a plurality of diagnostic history data included in the abnormal group When there is no difference by more than a preset criterion When it is judged that the insulation state judgment model has not been learned with predictive useful variables with high predictive ability and the insulation state judgment model is not completed, and there is a difference by more than a preset criterion Learning of the insulation state determination model may be repeated by changing the type of variable until

According to an embodiment of the present invention, a predictive useful variable may refer to a variable having a high predictive influence that helps clarify the distinction between groups by making the difference between the normal group and the abnormal group larger than other variables.

According to an embodiment of the present invention, the insulation state determination model learning unit 2410 calculates a difference in distribution between a unit distance of a plurality of diagnostic history data included in the normal group and a unit distance of a plurality of diagnostic history data included in the abnormal group. In order to increase the size, a plurality of predictive useful variables may be selected among the main variables, and the insulation state determination model may be re-learned to reselect the unit space and unit distance of the normal group based on the selected predictive useful variables.

According to an embodiment of the present invention, the insulation state determination model learning unit 2410 arranges variables and signal factors in the inner and outer arrays using a two-level orthogonal array table in order to select a plurality of predictive useful variables, A unit distance can be calculated for each major variable, and a signal-to-noise ratio of the calculated value can be derived.

According to an embodiment of the present invention, an orthogonal array table may mean a table made to make an experiment plan that can reduce the number of experiments at the expense of information on high-order interactions, and a two-level orthogonal array table Can be expressed as in Equation 5.

According to an embodiment of the present invention, in order to select useful predictive variables and increase the accuracy of insulation state information for each time point of heavy electrical equipment to be predicted, by using the difference in signal-noise ratio averages of unit distances calculated for each of a plurality of variables, The signal-to-noise ratio gain can be calculated.

According to an embodiment of the present invention, in order to select predictive useful variables using a two-level orthogonal array table, a unit distance is calculated by arranging variables and signal factors in the inner and outer arrays of the two-level orthogonal array, and the calculation The signal-to-noise ratio can be derived using the unit distance.

According to an embodiment of the present invention, the signal-noise ratio may vary according to the characteristics of the variable value, and in the case of a variable meaning that the higher the variable value, the better the characteristic, the lower the signal-noise ratio of the lager-the-better Signal-to-noise ratio of smallr-the-better characteristics in the case of variables that mean better characteristics, and nominal-the-better characteristics in the case of variables that mean better characteristics as the value of the variable approaches a specific target value If there is a signal-to-noise ratio of , and there is a known signal factor, the dynamic signal-to-noise ratio calculation formula can be used.

According to an embodiment of the present invention, a dynamic signal-noise ratio may be calculated as in Equation 6 in order to calculate a signal-noise ratio.

은 평균 제곱합(Sum of squares due to mean),

는 오차 분산(Error variance)이다.where n is the number of samples in the abnormal group,

is the Sum of squares due to mean,

is the error variance.

According to an embodiment of the present invention, in order to determine the magnitude of the prediction influence for each variable using the signal-noise ratio, a SNR gain may be calculated and the magnitude of the prediction influence for each variable may be determined according to the calculated value.

According to the above embodiment, the difference between the average of signal-noise ratios calculated by including the corresponding variable as one of a plurality of variables for each variable as shown in Equation 7 and the average of the signal-noise ratios calculated without the corresponding variable among the plurality of variables The gain of the signal-to-noise ratio of the variable can be calculated by the obtaining method.

은 i번째 변수가 사용된 신호 잡음비의 평균이고,

은 i번째 변수가 사용되지 않은 신호 잡음비의 평균이다.where i is the number of the variable,

is the average of the signal-to-noise ratios using the ith variable,

is the average of signal-to-noise ratios in which the ith variable is not used.

According to an embodiment of the present invention, the insulation state determination model learning unit 2410 determines that the corresponding variable has no predictive ability when the SNR gain has a negative value, and predicts it when the SNR gain has a positive value. It is determined that there is an ability, and the higher the positive value, the higher the predictive ability.

According to an embodiment of the present invention, insulation resistance 1 minute, polarization index determination, polarization index, dielectric loss tangent determination, dielectric loss tangent, alternating current distribution, alternating current, partial discharge high voltage are preset as predictive useful variable candidates. In addition, the signal-noise-ratio gain is calculated for each variable included in the predictive useful variable candidates, and only variables with a positive signal-noise ratio gain greater than a preset threshold value are selected as useful predictive variables to train the insulation state determination model. there is.

The failure determination model learning unit 2420 includes diagnosis history data and a plurality of heavy electric machines pre-diagnosed in normal and abnormal states among the diagnostic history data for a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted from the database. Insulation state information for each point of time is entered into the failure determination model to divide each heavy electrical device into system layers, and based on the diagnostic history data pre-diagnosed in normal and abnormal states, the importance of the function level required for each system layer for each heavy electrical device It is possible to learn a failure judgment model that derives state values for measurement items at each time point and determines a failure prediction time point according to whether or not the state values at each time point of the functional level required for each derived system layer are satisfied.

According to an embodiment of the present invention, the failure determination model learning unit 2420 is a system in the form of representing each heavy electric machine as an upper layer, which is a device included in an upper layer having a device included in a plurality of lower layers according to a preset criterion. The hierarchies are divided, and as the higher hierarchies go, the upper devices are higher, and a plurality of parts or components constituting the upper devices may be expressed as lower hierarchies of the upper hierarchies.

According to an embodiment of the present invention, the failure determination model derives state values at each time point for important measurement items of functional levels required for each system layer for each system layer, and diagnoses history data of a plurality of heavy electrical equipment and insulation state at each time point. By analyzing the information and calculating the failure probability for each time point according to the probability that the state value of each time point for the important measurement items derived for each system layer does not meet the preset threshold, the failure prediction time point can be determined.

According to an embodiment of the present invention, by dividing heavy electrical equipment into system layers, deriving state values for each system layer for important measurement items of functional levels required for each layer, and predicting the point of failure for the entire system of heavy electrical equipment. It is possible to solve a problem in which accurate prediction is impossible due to life or probability failure issues of various parts or components and provide higher accuracy.

The model data learning execution unit 2430 selects, as an output variable, state values at each point in time of a plurality of functional levels required for each system layer for each heavy electric device derived from the failure determination model, and the same heavy electric device as the received heavy electric machine to be predicted. A probability distribution is calculated by analyzing diagnostic history data and output variables for a plurality of heavy electric machines included in the group, the calculated probability distribution is set as a prior distribution, and the prior distribution for the output variables is a posterior distribution using a Bayesian approach. can be continuously updated.

According to an embodiment of the present invention, in order for the model data learning execution unit 2430 to update the failure determination model for higher accuracy, a plurality of functional levels required for each system layer for each heavy electric machine derived from the failure determination model. Select the state value for each point of time as an output variable, enter diagnosis history data and output variables for a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted, calculate a probability distribution according to the degree of uncertainty, and convert it to a prior distribution. can be set

)에 가능도 함수를 적용하여 사후 분포(

)를 산출하여, 산출된 사후 분포가 고장 판단 모델을 이용하여 산출된 출력 변수를 통해 사전 분포로 산출될 수 있도록 베이시안 접근법을 이용하여 고장 판단 모델을 업데이트 할 수 있다.According to the above embodiment, as shown in Equation 8, the prior distribution (

) by applying the likelihood function to the posterior distribution (

), the failure determination model can be updated using the Bayesian approach so that the calculated posterior distribution can be calculated as a prior distribution through the output variable calculated using the failure determination model.

The failure prediction time information generation unit 2440 inputs diagnosis history data and heavy electric device identification information of the heavy electric device to be predicted to the insulation state determination model, receives insulation state information for each time point of the prediction target heavy electric device, and performs the prediction Failure prediction time information may be generated by inputting insulation state information of each time point of the target heavy electric device and diagnostic history data of the heavy electric device to be predicted into a failure determination model.

Here, the failure prediction time information may refer to information including information about how much a specific heavy electric device is currently degraded or when a failure time is predicted.

According to an embodiment of the present invention, the failure prediction time information generation unit 2440 may receive the prediction target failure time point using the learned insulation state determination model and the failure determination model.

According to an embodiment of the present invention, the learned insulation state determination model receives diagnosis history data and heavy electric device identification information of the heavy electric equipment to be predicted, outputs insulation state information for each time point of the heavy electric equipment to be predicted, and determines failure. The model can receive diagnosis history data of the heavy electrical equipment to be predicted and information on the insulation status of the heavy electric equipment to be predicted at each time point, and receive failure prediction time information indicating the failure probability of the heavy electrical equipment to be predicted at each time point.

5 is a diagram illustrating a method of performing an update using a Bayesian approach in the model data learning performer disclosed in FIG. 4 .

)로 설정할 수 있으며, 사전 분포(

)에 가능도 함수를 적용하여 사후 분포(

)를 산출하여 산출된 사후 분포가 고장 판단 모델을 이용하여 산출된 출력 변수를 통해 사전 분포로 산출될 수 있도록 베이시안 접근법을 이용하여 고장 판단 모델을 업데이트 할 수 있다.Referring to FIG. 2, a method of performing an update using a Bayesian approach in the model data learning performing unit 2430 is shown, and the output variable selected in the model data learning performing unit 2430 and the same heavy electric device as the heavy electric machine to be predicted By analyzing the diagnostic history data for a plurality of heavy electric machines included in the group, a probability distribution is calculated according to the degree of uncertainty, and the prior distribution (

), and the prior distribution (

) by applying the likelihood function to the posterior distribution (

) can be calculated and the failure determination model can be updated using the Bayesian approach so that the calculated posterior distribution can be calculated as a prior distribution through the output variable calculated using the failure determination model.

6 is a diagram illustrating a flow of data for generating degradation prediction information generated using a regression model selected for each data set using a degradation simulation model according to an embodiment of the present invention.

Referring to FIG. 6, it is shown that degradation prediction information is generated using a regression model selected for each data set using a degradation simulation model according to an embodiment of the present invention, and a plurality of data included in diagnosis history data A plurality of data sets can be created by performing grouping by target.

In addition, among the regression models stored in advance for multiple data sets, the model with the smallest error can be selected as the regression model of the data set by comparing the analysis result value and the diagnosis history data. A built-in degradation simulation model can be created.

7 is a diagram illustrating a process of calculating an average distance within a unit space for determining a normal state and an abnormal state in order to train an insulation state determination model according to an embodiment of the present invention.

Referring to FIG. 7 , a process of calculating an average distance within a unit space for determining a normal state and an abnormal state in order to learn an insulation state determination model according to an embodiment of the present invention is shown, and from the center point of a normal group A space in which the unit distance average of unit distances is 1.0 is selected as the unit space, and a space in which the unit distance average of unit distances from the center point of the abnormal group is 1.0 is selected as the unit space. It is determined whether the distribution of unit distances of the plurality of diagnostic history data respectively located in is different than a preset standard, and if there is no difference, it is determined that the distribution of unit distances is an insulation state model requiring re-learning, and learning can be repeated.

8 is a flowchart of a method for recommending preventive maintenance of heavy electrical equipment based on real-time sensing data according to an embodiment of the present invention.

Diagnostic history data and heavy electric machine identification information for a plurality of heavy electric machines generated by chronologically aligning diagnostic information for each heavy electric machine and sensor measurement data collected by at least one sensor are stored (S10).

Diagnostic history data and heavy electric machine identification information and sensor measurement data collected by at least one sensor for a plurality of heavy electric machines generated by time-sequentially aligning diagnostic information for each heavy electric machine may be stored.

Here, the heavy electric machine may mean large-capacity industrial equipment or industrial machinery used in industrial sites and power plants.

According to an embodiment of the present invention, diagnosis history data may refer to information generated as one piece of information by arranging diagnosis information including diagnosis results performed at a plurality of specific points in time according to the lapse of time.

According to one embodiment of the present invention, the heavy electric machine identification information may mean a unique serial number possessed by the heavy electric machine for identifying the heavy electric machine or an ID number randomly designated on the database, regardless of the format, to identify the heavy electric machine. Anything that can be used can be used without limitation.

According to an embodiment of the present invention, the heavy electric machine identification information may include information specific to the electric machine, such as the manufacturer of the heavy electric machine, the manufacturing period, and the capacity of the electric machine, by matching the serial number or ID number of the heavy electric machine.

It is possible to manage diagnostic history data and heavy electric machine identification information for a plurality of heavy electric machines by receiving direct input from a user by using a data record file or providing a data input form.

According to an embodiment of the present invention, the offline data management unit 1100 may receive diagnosis history data and heavy electric machine identification information for a plurality of heavy electric machines directly from a user or in a file format having a certain form.

According to an embodiment of the present invention, sensor measurement result data can be periodically uploaded and managed from identification information of a plurality of heavy electric machines and at least one sensor of heavy electric machines.

According to an embodiment of the present invention, measurement data of a plurality of measurement items for a plurality of heavy electric machines from a plurality of sensors connected to the heavy electric machine preventive maintenance recommendation system may be collected according to a preset cycle and uploaded to a database.

According to an embodiment of the present invention, a plurality of sensors and a database can be connected using a wireless network environment, but a wired network environment can be used.

According to an embodiment of the present invention, sensor measurement result data for each inspection item for the heavy electric machine may be generated by receiving the measurement values of the heavy electric machine inspection items from the sensor installed in the heavy electric machine and aligning them in time series.

A data group is created by grouping a plurality of heavy electric machines having a relationship based on heavy electric machine identification information, and diagnostic history data and sensor measurement result data are managed for each heavy electric machine included in the data group (S20).

According to an embodiment of the present invention, a data group is created by grouping a plurality of heavy electric machines having a correlation with respect to at least one item of manufacturer, manufacturing period, and motor capacity based on heavy electric machine identification information, and the data group is included in the data group. Diagnosis history data and sensor measurement result data can be managed for each heavy electric device.

According to the above embodiment, in performing grouping, it is possible to group a plurality of electric devices that achieve the conditions of the same manufacturer, manufacturing time within a certain period, and electric device capacity within a certain period into one group, but are not limited thereto, and manufacturers and manufacturers are not limited thereto. Any criterion for creating a group using the above conditions, such as weighting according to the number of conditions satisfied when a plurality of conditions are satisfied among time and capacity of the motor, can be used without limitation.

A plurality of data sets are created by grouping a plurality of data included in the diagnosis history data, and a regression model previously stored for each data set is used by using the history data included in each generated data set as learning data. The deterioration simulation model is learned to calculate an equation having the form of a linear equation by performing regression analysis (S30).

According to an embodiment of the present invention, a plurality of data sets are generated by grouping a plurality of data included in the diagnosis history data of a plurality of heavy electric machines included in the same heavy electric device group as the heavy electric machine to be predicted, and The degradation simulation model may be learned to calculate an equation having a form of a linear equation by performing regression analysis through a pre-stored regression model for each data set using historical data included in each data set as training data.

According to an embodiment of the present invention, a plurality of data sets are created by grouping a plurality of data included in diagnosis history data, and each data set is used as learning data by using the history data included in each generated data set as learning data. The degradation simulation model may be trained to calculate an equation having a form of a linear equation by performing regression analysis through a regression model stored in advance for each set.

According to an embodiment of the present invention, an equation representing a plurality of data included in the diagnosis history data may be expressed in the form of a linear equation (y = ax + b), and the linear equation calculated in this form is used. Thus, a solution of an equation representing a plurality of data can be calculated.

According to an embodiment of the present invention, a plurality of data sets are created by grouping a plurality of data included in diagnosis history data, and each data set is used as learning data by using the history data included in each generated data set as learning data. Linear regression analysis can be performed through pre-stored regression models for each set.

According to an embodiment of the present invention, a polynomial is constructed for the data included in the diagnosis history data of the heavy electric machine to be predicted, and a parameter is estimated, and an equation expressed as a straight line is calculated according to the tendency of the parameter-estimated data. can do.

According to an embodiment of the present invention, the polynomial formed for the data included in the diagnosis history data of heavy electrical equipment may follow the Dx3+Cx2+Bx+A format, but is not limited thereto, and the polynomial format in which parameter estimation can be performed If so, it can be used without limitation.

According to an embodiment of the present invention, random sampling may be performed on the polynomial in performing parameter estimation by constructing a polynomial targeting the data included in the received diagnosis history data of the heavy electric machine to be predicted. Instead of performing random sampling, a method of estimating parameters immediately and estimating data after a specific period of time after the final measurement may be used.

According to an embodiment of the present invention, when using a method of estimating data during the course of a specific institution after parameter estimation and final measurement, data may be expressed in the form of a quadratic function, and in fact, including the first measurement result and the last measurement result In the case of using parameters estimated from data, verification using past data showed high accuracy in estimating deterioration in the next two years.

According to an embodiment of the present invention, in performing parameter estimation by constructing a polynomial targeting the data included in the input diagnosis history data of the heavy electric machine to be predicted, after randomly selecting and modeling the data based on the colony, the suitability It is possible to relatively improve the accuracy of estimation by repeating the process of performing random selection after going through the estimation process to select the most suitable model.

According to an embodiment of the present invention, the deterioration point of the heavy electric machine is predicted by applying a weight reflecting at least one information of the frequency of stop events of the heavy electric machine, the driving pattern of the heavy electric machine, and the installation environment to an equation, and using the equation. Deterioration prediction information may be generated by inputting a plurality of data included in diagnosis history data to the simulation model.

According to an embodiment of the present invention, a weight may be applied to an equation calculated on the basis of at least one information of the occurrence frequency of stop events of heavy electric machines, driving patterns of heavy electric machines, and installation environments.

According to an embodiment of the present invention, the calculated equation may be expressed in the form of a linear equation (y = ax + b), and according to this embodiment, the frequency of stop events of heavy electric machines, driving patterns of heavy electric machines, and installation environment A weight exponent β can be calculated based on at least one of the information, and a weight can be applied to the equation by reflecting the calculated β to the linear equation as shown in Equation 1.

According to an embodiment of the present invention, the weight index β may be calculated according to one condition among the frequency of stop events of heavy electric equipment, the driving pattern of heavy electric equipment, and the installation environment, but is not limited thereto and may be calculated according to a plurality of conditions. When calculated according to each condition, a weight is applied according to each condition, and the absolute value can be calculated in a form that increases or decreases.

According to one embodiment of the present invention, a plurality of data included in the diagnosis history data is input into a deterioration simulation model that predicts the deterioration point of heavy electrical equipment including an equation with weights applied thereto, and according to the trend trend indicated by the graph according to the equation Deterioration time prediction or the degradation degree of the heavy electrical equipment may be calculated according to the predicted degradation time point, and the degradation prediction information may include degradation time prediction information and/or deterioration degree information of the heavy electrical equipment.

According to an embodiment of the present invention, when parameter estimation is performed on the data included in the diagnosis history data of the heavy electric machine to be predicted and an equation expressed by a straight line is calculated according to the tendency of the parameter-estimated data, By applying the weight reflecting at least one information of stop event occurrence frequency, heavy electrical equipment operation pattern, and installation environment to the equation, and inputting parameter estimated data to the deterioration simulation model that predicts the deterioration point of heavy electrical equipment using the equation, degradation time can be predicted.

According to the above embodiment, the deterioration time of the heavy electric machine may be predicted by inputting the parameter-estimated data to the deterioration simulation model for predicting the deterioration time of the heavy electric machine including the weighted equation.

According to an embodiment of the present invention, when receiving diagnosis history data of a heavy electric machine classified into the same group as a heavy electric machine to be predicted, the degradation simulation model learning unit 2100 includes the diagnosis history data of the heavy electric machine to be predicted Parameter estimation can be performed by constructing a polynomial targeting multiple data and the data included in the diagnostic history data of heavy electric machines classified into the same group received. At this time, the frequency of stop events of heavy electric machines, A weight reflecting at least one or more information of the installation environment is applied to the equation, and the deterioration time of the heavy electric machine can be predicted by inputting the parameter-estimated data to the deterioration simulation model that predicts the deterioration time of the heavy electric machine using the equation.

According to an embodiment of the present invention, a regression model having an analysis result value having a relatively minimal error with the diagnosis history data by comparing the analysis result value through the above equation calculated by the pre-stored regression model and the diagnosis history data. can be selected as the regression model of the data set to design a degradation simulation model composed of multiple regression analysis models.

According to an embodiment of the present invention, a plurality of data sets are created by grouping a plurality of data included in diagnosis history data, and each data set is used as learning data by using the history data included in each generated data set as learning data. When regression analysis is performed through a regression model stored in advance for each set, a model with the smallest error is selected as the regression model of the data set by comparing the analysis result value and the historical data among the previously stored regression models. A degradation simulation model composed of a regression analysis model may be generated, linear regression analysis may be performed for each of the plurality of data sets according to the degradation simulation model, and degradation prediction information may be generated based on the performed result.

According to an embodiment of the present invention, among pre-stored regression models, a model with a relatively least error is selected as a regression model of the data set by comparing the analysis result value and the diagnosis history data, and deterioration composed of a plurality of regression analysis models. A simulation model can be created.

According to an embodiment of the present invention, in order to select a regression model for each data set, a pre-stored regression model is applied to each data set to derive an analysis result value, and data on the derived analysis result value and actual diagnosis history data A regression model that derives an analysis result value with the smallest error by comparing the values may be selected as the regression model of the corresponding data set.

According to an embodiment of the present invention, the degradation simulation model may be formed of regression models selected for each data set, and diagnostic history data is input to the formed degradation simulation model, and the degradation simulation model is generated by comparing output values and diagnostic history data. can be verified

According to an embodiment of the present invention, in performing linear regression analysis, in estimating an equation representing a plurality of data included in diagnostic history data using a bagging technique, in order to relatively reduce the influence of measurement error and noise A degradation simulation model may be generated by selecting a regression analysis model having the smallest error for each generated data set.

According to an embodiment of the present invention, linear regression analysis may be performed for each data set according to the degradation simulation model, and degradation prediction information may be generated based on the performed result.

According to an embodiment of the present invention, degradation prediction information may be generated according to the degradation simulation model, and when a failure occurs due to degradation occurring before the degradation prediction information according to the degradation simulation model, the degradation simulation model is determined according to the type and cause of the failure. It can be applied to the prediction of deterioration of heavy electrical equipment of the same type by automatically reflecting the degradation prediction criteria of

Based on the sensor measurement result data, the deterioration weight index γ of the heavy electric machine to be predicted is calculated according to the error trend with the optimal value set for each diagnosis item of the heavy electric machine to be predicted, and the deterioration weight index is reflected in the deterioration simulation model (S40) .

According to an embodiment of the present invention, based on sensor measurement result data of a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted from the database, the error trend with the optimal value set for each diagnosis item of the heavy electric machine to be predicted Accordingly, the deterioration weight index (γ) of the heavy electric machine to be predicted may be calculated, and the deterioration weight index may be reflected in the deterioration simulation model.

According to an embodiment of the present invention, the optimal value set for each diagnostic item of the heavy electric machine to be predicted and the actually measured sensor measurement calculated by analyzing sensor measurement result data of a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted It is possible to calculate the error with the measured value for each diagnosis item of the heavy electrical equipment to be predicted included in the result data, and calculate the weight index γ using the value of the calculated error.

According to an embodiment of the present invention, as a method of calculating an optimal value set for each diagnosis item of a heavy electric machine to be predicted, the degree of similarity with the heavy electric machine to be predicted is calculated for each of a plurality of heavy electric machines according to a predetermined criterion, and each heavy electric machine is diagnosed. A method of calculating the average value after summing the measured values included in the sensor measurement result data for each item with the similarity applied as a weighted value can be used, but it is not limited to this, and a method of presetting the optimal value for each diagnostic item of the heavy electrical equipment to be predicted available.

According to an embodiment of the present invention, as a method of calculating the weight index γ, an average value of error values calculated for each diagnostic item may be obtained, and the average value may be calculated based on the average value, but is not limited thereto. A method of calculating the average value after summing the values to which the weights determined according to the result are applied may be used.

According to an embodiment of the present invention, the equation included in the deterioration simulation as shown in Equation 2 using the weight index γ calculated by analyzing sensor measurement result data of a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted. It is possible to generate degradation prediction information by assigning a high weight (γ) to .

Linear regression analysis is performed for each data set according to the degradation simulation model, degradation prediction information is generated based on the performed results, and preventive maintenance recommendation information is generated according to the degradation prediction time included in the degradation prediction information (S50). .

According to an embodiment of the present invention, linear regression analysis is performed for each data set according to the degradation simulation model, degradation prediction information is generated based on the performed result, and prevention is performed according to the degradation prediction time included in the degradation prediction information Maintenance recommendation information can be created.

According to an embodiment of the present invention, among the diagnostic history data for a plurality of heavy electric machines, diagnostic history data pre-diagnosed in normal and abnormal states is received as learning data, an insulation state determination model and a failure determination model are learned, and the learned Soundness analysis information including failure prediction time information of the heavy electrical equipment to be predicted may be generated using the insulation state determination model and the failure determination model.

According to an embodiment of the present invention, a weight δ at the time of prediction of degradation included in the degradation prediction information and a weight ε at the time of prediction of failure included in the health analysis information are calculated, and the weighted degradation prediction information and health analysis information are used. Therefore, preventive maintenance recommendation information can be generated.

According to an embodiment of the present invention, the weight δ is calculated at the time of prediction of degradation included in the degradation prediction information and the weight ε is calculated at the time of prediction of failure included in the health analysis information. Preventive maintenance recommendation information that recommends preventive maintenance based on a point in time can be created.

According to an embodiment of the present invention, the weight δ and the weight ε can be initially assigned the same as 1, and it is determined whether the time information output by inputting the diagnosis history data matches the failure history data that occurred in the past. The weights can be calculated by performing an update by adjusting the ratio of the weight δ and the weight ε, respectively, in a direction in which accuracy increases relatively.

According to an embodiment of the present invention, diagnosis history data pre-diagnosed in a normal state and an abnormal state among diagnosis history data for a plurality of heavy electric machines included in the same heavy electric device group as a heavy electric machine to be predicted from a database is input as learning data An insulation state determination model capable of determining an insulation state through a distance average may be learned by setting a unit space and calculating a distance average within the unit space for determining a normal state and an abnormal state.

According to an embodiment of the present invention, a unit distance determination method may be used as a multivariate data mining technique for a plurality of data included in diagnosis history data.

According to an embodiment of the present invention, according to an embodiment using the unit distance determination method, after setting the unit space of the normal group and expressing each individual data in the unit space based on the center point of the normal group, how far is it from the center point of the normal group? Whether or not they are apart is calculated as a unit distance, and the greater the unit distance, the higher the probability that the group is not a normal group. If it exceeds a preset threshold value, it can be determined as an abnormal group.

According to an embodiment of the present invention, diagnosis history data for a plurality of heavy electric machines included in the same heavy electric machine group as a prediction target heavy electric machine input from a user may be received.

According to an embodiment of the present invention, among the received diagnostic history data for a plurality of heavy electric machines, diagnosis history data for heavy electric machines diagnosed as normal according to criteria previously diagnosed as normal or abnormal are used in unit space. The average of the distances within can be calculated.

According to an embodiment of the present invention, a unit space may be set based on an average value of a plurality of major variable measurement data of diagnostic history data pre-diagnosed in a normal state among input diagnostic history data of heavy electrical equipment.

According to an embodiment of the present invention, a normal group is set based on diagnostic history data pre-diagnosed in a normal state among input diagnostic history data of heavy electrical equipment, and a plurality of major variables of the diagnostic history data pre-diagnosed in a normal state A unit space based on the average value of the measured data can be derived.

According to an embodiment of the present invention, the main variable data included in the diagnosis history data for a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted is expressed in the reference space, and each individual data expressed and the reference space The unit distance can be measured by measuring the probability distribution distance from the center point of the normal group.

According to an embodiment of the present invention, a plurality of major variable measurement data of diagnosis history data pre-diagnosed in normal and abnormal states among the input diagnosis history data of heavy electrical equipment are input to an insulation judgment model to diagnose the input heavy electrical equipment. A normal group may be created based on measurement data of major variables of a plurality of diagnostic history data pre-diagnosed as normal among history data.

According to the above embodiment, normalization is performed on diagnosis history data included in the normal group, and a probability distribution distance between a center point of the normal group and a plurality of diagnosis history data included in the normal group is measured as a unit distance; A space in which the unit distance average of measured unit distances is 1.0 may be selected as the unit space.

According to an embodiment of the present invention, a value obtained by subtracting the average value of the main variable measurement data of each diagnosis history data included in the normal group from the main variable measurement data of each diagnosis history data included in the normal group is included in the normal group. It is possible to calculate a plurality of normalized vectors for each diagnostic history data by performing normalization by dividing the main variable measurement data by the standard deviation of each diagnostic history data.

According to an embodiment of the present invention, in order to derive a unit space in which the distance average is 1.0 from the center point of the normal group in order to select it as a unit space, a plurality of key variable measurement data included in diagnosis history data diagnosed in advance in a normal state Normalization can be performed on the target.

According to the above embodiment, as shown in Equation 3, normalization is the standard of the main variable measurement data of each diagnosis history data included in the normal group by subtracting the average value of the main variable measurement data of each diagnosis history data included in the normal group. This can be done by dividing by the deviation.

According to an embodiment of the present invention, it is possible to measure a probability distribution distance with a plurality of diagnostic history data included in a normal group by applying an inverse matrix of a correlation matrix composed of correlation coefficients to a plurality of normalized vectors.

According to an embodiment of the present invention, it is possible to input a plurality of major variable measurement data of diagnostic history data pre-diagnosed as an abnormal state among received diagnostic history data of heavy electrical equipment to an insulation determination model.

According to an embodiment of the present invention, among the received diagnostic history data on a plurality of heavy electric machines, diagnostic history data on heavy electric machines diagnosed as abnormal according to criteria pre-diagnosed as normal or abnormal are used in unit space. The average of the distances within can be calculated.

According to an embodiment of the present invention, a normal group is set based on diagnostic history data pre-diagnosed in a normal state among input diagnostic history data of heavy electrical equipment, and a plurality of major variables of the diagnostic history data pre-diagnosed in an abnormal state. A unit space based on the average value of the measured data can be derived.

According to an embodiment of the present invention, the main variable data included in the diagnosis history data for a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted is expressed in the reference space, and each individual data expressed and the reference space The unit distance can be measured by measuring the probability distribution distance from the center point of the abnormal group.

According to the above embodiment, an abnormal group is created based on the main variable measurement data of a plurality of diagnostic history data pre-diagnosed in an abnormal state among the received diagnostic history data of heavy electrical equipment, and a plurality of diagnostic history data included in the abnormal group. Normalize the abnormal group by dividing the average value of the main variable measurement data of each diagnosis history data included in the normal group from the main variable measurement data of the normal group by the standard deviation of the main variable measurement data of each diagnosis history data included in the normal group A plurality of normalized vectors can be calculated.

According to the above embodiment, a probability distribution distance with a plurality of diagnostic history data included in an abnormal group can be measured as a unit distance by applying an inverse matrix of a correlation matrix composed of correlation coefficients to a plurality of normalized vectors.

According to an embodiment of the present invention, the normalized vector (Z) calculates the unit distance of the normal group using the inverse matrix of the correlation matrix (C) composed of correlation coefficients, and the formula for this is as shown in Equation 4.

According to an embodiment of the present invention, if the distribution of the unit distance of a plurality of diagnosis history data included in the normal group and the unit distance of the plurality of diagnosis history data included in the abnormal group do not differ by more than a preset standard, the insulation is isolated. It is determined that learning of the state determination model is not completed, and learning of the insulation state determination model may be repeated until a difference greater than or equal to a preset standard is obtained.

According to an embodiment of the present invention, a distribution of unit distances of a plurality of diagnosis history data included in the normal group and unit distances of a plurality of diagnosis history data included in the abnormal group is compared and predicted when the difference is not greater than a preset standard. It is judged that the insulation state judgment model has not been learned with the predictive useful variables with high ability, and the insulation state judgment model is judged to have not been completed. The training of the model can be repeated.

According to an embodiment of the present invention, a predictive useful variable may refer to a variable having a high predictive influence that helps clarify the distinction between groups by making the difference between the normal group and the abnormal group larger than other variables.

According to an embodiment of the present invention, in order to increase the difference between the distribution of the unit distance of a plurality of diagnostic history data included in the normal group and the unit distance of the plurality of diagnostic history data included in the abnormal group, a plurality of predictions among major variables are useful. Variables can be selected, and the insulation state determination model can be re-learned to re-select the unit space and unit distance of the normal group based on the selected predictive useful variable.

According to an embodiment of the present invention, in order to select a plurality of predictive useful variables, a two-level orthogonal array table is used to arrange variables and signal factors in the inner and outer arrays, and to calculate unit distances for each major variable. and the signal-to-noise ratio of the calculated value can be derived.

According to an embodiment of the present invention, an orthogonal array table may mean a table made to make an experiment plan that can reduce the number of experiments at the expense of information on high-order interactions, and a two-level orthogonal array table Can be expressed as in Equation 5.

According to an embodiment of the present invention, in order to select useful predictive variables and increase the accuracy of insulation state information for each time point of heavy electrical equipment to be predicted, by using the difference in signal-noise ratio averages of unit distances calculated for each of a plurality of variables, The signal-to-noise ratio gain can be calculated.

According to an embodiment of the present invention, in order to select predictive useful variables using a two-level orthogonal array table, a unit distance is calculated by arranging variables and signal factors in the inner and outer arrays of the two-level orthogonal array, and the calculation The signal-to-noise ratio can be derived using the unit distance.

According to an embodiment of the present invention, the signal-noise ratio may vary according to the characteristics of the variable value, and in the case of a variable meaning that the higher the variable value, the better the characteristic, the lower the signal-noise ratio of the lager-the-better Signal-to-noise ratio of smallr-the-better characteristics in the case of variables that mean better characteristics, and nominal-the-better characteristics in the case of variables that mean better characteristics as the value of the variable approaches a specific target value If there is a signal-to-noise ratio of , and there is a known signal factor, the dynamic signal-to-noise ratio calculation formula can be used.

According to an embodiment of the present invention, a dynamic signal-to-noise ratio may be calculated as shown in Equation 6 in order to calculate a signal-to-noise ratio.

According to an embodiment of the present invention, in order to determine the magnitude of the prediction influence for each variable using the signal-noise ratio, a SNR gain may be calculated and the magnitude of the prediction influence for each variable may be determined according to the calculated value.

According to the above embodiment, the difference between the average of signal-noise ratios calculated by including the corresponding variable as one of a plurality of variables for each variable as shown in Equation 7 and the average of the signal-noise ratios calculated without the corresponding variable among the plurality of variables The gain of the signal-to-noise ratio of the variable can be calculated by the obtaining method.

According to an embodiment of the present invention, when the SNR gain has a negative value, it is determined that the corresponding variable has no predictive ability, and when the SNR gain has a positive value, it is determined that the variable has predictive ability, and a positive value It can be determined that the larger this is, the higher the predictive ability.

According to an embodiment of the present invention, insulation resistance 1 minute, polarization index determination, polarization index, dielectric loss tangent determination, dielectric loss tangent, alternating current distribution, alternating current, partial discharge high voltage are preset as predictive useful variable candidates. In addition, the signal-noise-ratio gain is calculated for each variable included in the predictive useful variable candidates, and only variables with a positive signal-noise ratio gain greater than a preset threshold value are selected as useful predictive variables to train the insulation state determination model. there is.

According to an embodiment of the present invention, among the diagnostic history data for a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted from the database, diagnosis history data pre-diagnosed in normal and abnormal states and a plurality of heavy electric machines Insulation state information at each point in time is input into the failure determination model to divide each heavy electrical device into system layers, and to measure the important function level required for each system layer for each heavy electrical device based on the diagnosis history data pre-diagnosed in normal and abnormal states It is possible to learn a failure determination model that derives state values for each item for each time point and determines a failure prediction time point according to whether or not the state values for each time point of the functional level required for each derived system layer are satisfied.

According to an embodiment of the present invention, each heavy electric machine is divided into a system layer in a form represented by an upper layer, which is a device included in an upper layer having a device included in a plurality of lower layers according to a preset criterion, and the higher the upper layer goes It is a device, and a plurality of parts or components constituting an upper device may be expressed as a lower layer of a corresponding upper layer.

According to an embodiment of the present invention, the failure determination model derives state values at each time point for important measurement items of functional levels required for each system layer for each system layer, and diagnoses history data of a plurality of heavy electrical equipment and insulation state at each time point. By analyzing the information and calculating the failure probability for each time point according to the probability that the state value of each time point for the important measurement items derived for each system layer does not meet the preset threshold, the failure prediction time point can be determined.

According to an embodiment of the present invention, by dividing heavy electrical equipment into system layers, deriving state values for each system layer for important measurement items of functional levels required for each layer, and predicting the point of failure for the entire system of heavy electrical equipment. It is possible to solve a problem in which accurate prediction is impossible due to life or probability failure issues of various parts or components and provide higher accuracy.

According to an embodiment of the present invention, for each heavy electrical device derived from the failure determination model, a state value at each point in time of a plurality of functional levels required for each system layer is selected as an output variable, and the same heavy electrical device group as the received heavy electrical device to be predicted. By analyzing the diagnostic history data and output variables for a plurality of heavy electric machines included in, a probability distribution is calculated, the calculated probability distribution is set as a prior distribution, and the prior distribution for the output variables is converted into a posterior distribution using a Bayesian approach. Updates can be made continuously.

According to an embodiment of the present invention, in order to update the failure determination model for higher accuracy, state values at each point in time of a plurality of functional levels required for each system layer are selected as output variables for each heavy electric machine derived from the failure determination model. And, by inputting diagnostic history data and output variables for a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted, a probability distribution can be calculated according to the degree of uncertainty and set as a prior distribution.

)에 가능도 함수를 적용하여 사후 분포(

)를 산출하여, 산출된 사후 분포가 고장 판단 모델을 이용하여 산출된 출력 변수를 통해 사전 분포로 산출될 수 있도록 베이시안 접근법을 이용하여 고장 판단 모델을 업데이트 할 수 있다.According to the above embodiment, as shown in Equation 8, the prior distribution (

) by applying the likelihood function to the posterior distribution (

), the failure determination model can be updated using the Bayesian approach so that the calculated posterior distribution can be calculated as a prior distribution through the output variable calculated using the failure determination model.

According to an embodiment of the present invention, diagnosis history data and heavy electric machine identification information of a heavy electric machine to be predicted are input to the insulation state determination model, insulation state information is provided for each time point of the heavy electric machine to be predicted, and Failure prediction time information may be generated by inputting insulation state information at each point in time and diagnosis history data of heavy electrical equipment to be predicted to a failure determination model.

According to an embodiment of the present invention, it is possible to receive a failure prediction time of the prediction target heavy electric machine using the learned insulation state determination model and the failure determination model.

According to an embodiment of the present invention, the learned insulation state determination model receives diagnosis history data and heavy electric device identification information of the heavy electric equipment to be predicted, outputs insulation state information for each time point of the heavy electric equipment to be predicted, and determines failure. The model can receive diagnosis history data of the heavy electrical equipment to be predicted and information on the insulation status of the heavy electric equipment to be predicted at each time point, and receive failure prediction time information indicating the failure probability of the heavy electrical equipment to be predicted at each time point.

Embodiments of the present invention are not implemented only through the devices and / or methods described above, and the embodiments of the present invention have been described in detail above, but the scope of the present invention is not limited thereto, and the following claims Various modifications and improvements of those skilled in the art using the basic concept of the present invention defined in , also belong to the scope of the present invention.

Claims (30)

In the heavy electric equipment preventive maintenance recommendation system,
A database for storing diagnostic history data and heavy electric machine identification information and sensor measurement data collected by at least one sensor for a plurality of heavy electric machines generated by chronologically aligning diagnostic information for each heavy electric machine; and
A management server for generating preventive maintenance recommendation information by analyzing diagnosis history data, heavy electric machine identification information, and sensor measurement data for a plurality of heavy electric machines from the input database,
The database is
Off-line data management unit for managing diagnosis history data and heavy electric machine identification information for a plurality of heavy electric machines by receiving direct input from a user by using a data record file or providing a data input form;
An online data management unit that receives and manages identification information of a plurality of heavy electric machines and sensor measurement result data from at least one or more sensors and periodically uploads them; and
Based on the heavy electric machine identification information, a data group is created by grouping a plurality of heavy electric machines having a correlation with respect to at least one item of manufacturer, manufacturing period, and electric machine capacity, and diagnosis history for each heavy electric machine included in the data group Further comprising a related data group generating unit for managing data and sensor measurement result data;
The management server,
From the database, a plurality of data sets are generated by grouping a plurality of data included in the diagnostic history data of a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted, and a plurality of data sets are generated. A deterioration simulation model learning unit for learning a deterioration simulation model to calculate an equation having a form of a linear equation by performing a regression analysis through a pre-stored regression model for each data set using the included history data as learning data;
Based on the sensor measurement result data of a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted from the database, deterioration of the heavy electric machine to be predicted according to the error trend with the optimal value set for each diagnostic item of the heavy electric machine to be predicted a degradation weight reflection unit that calculates a weight index (γ) and reflects the degradation weight index to a degradation simulation model;
Prevention of performing linear regression analysis for each data set according to the deterioration simulation model, generating deterioration prediction information based on the performed results, and generating preventive maintenance recommendation information according to the deterioration prediction time included in the deterioration prediction information Maintenance recommendation information generation unit; and
Among the diagnostic history data for a plurality of heavy electrical equipment, diagnosis history data pre-diagnosed in normal and abnormal states is received as learning data to learn an insulation state determination model and a failure determination model, and the learned insulation state determination model and failure determination Further comprising a health analysis information generation unit for generating health analysis information including failure prediction time information of the heavy electric equipment to be predicted using the model,
The degradation simulation model learning unit
The weight index β is calculated based on at least one information of the frequency of occurrence of stop events of heavy electrical equipment, driving patterns of heavy electrical equipment, and installation environment, and the calculated β is reflected in the equation,
The degradation weight reflection unit,
The sensor measurement result data is analyzed to calculate the error with the optimal value set for each diagnosis item of the heavy electric machine to be predicted, and the weight index γ having a higher value as the magnitude of the error increases, the equation as shown in Equation 2 A weight is applied to and reflected in the degradation simulation model,
The preventive maintenance recommendation information generation unit,
Calculate the weight δ at the time of prediction of degradation included in the degradation prediction information and the weight ε at the time of prediction of failure included in the health analysis information, and use the weighted degradation prediction information and health analysis information to obtain the preventive maintenance recommendation information create,
The weight δ and the weight ε are given the same value of 1 at first, and the weight in the direction of relatively increasing accuracy depending on whether or not the time information output by inputting the diagnosis history data matches the failure history data that occurred in the past. The weights are calculated by performing an update by adjusting the ratio of δ and the weight ε, respectively,
The preventive maintenance recommendation information calculates the weight δ at the time of prediction of deterioration included in the deterioration prediction information and the weight ε at the time of prediction of failure included in the soundness analysis information, and the average value of the weighted values for each time point is obtained. A preventive maintenance recommendation system for heavy electrical equipment based on real-time sensing data, characterized in that it is generated on the basis of standard.
[Equation 2]

The degradation simulation model learning unit of claim 1,
Grouping is performed on a plurality of data included in the diagnosis history data to generate a plurality of data sets;
Characterized in that the degradation simulation model is trained to calculate an equation having the form of a linear equation by performing regression analysis through a pre-stored regression model for each data set using the history data included in each generated data set as learning data. A preventive maintenance recommendation system for heavy electrical equipment based on real-time sensing data. The degradation simulation model learning unit of claim 2,
By comparing the analysis result value through the equation calculated by the pre-stored regression model and the diagnosis history data, a regression model having an analysis result value having a relatively minimum error with the diagnosis history data is a regression model of the corresponding data set. A system for recommending preventive maintenance of heavy electrical equipment based on real-time sensing data, characterized in that it designs a degradation simulation model composed of a plurality of regression analysis models. delete delete The soundness analysis information generating unit of claim 1,
Among the diagnostic history data for a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted from the database, diagnosis history data pre-diagnosed in normal and abnormal states is input as learning data, and a unit space is set, and a normal an insulation state determination model learning unit which calculates an average distance in the unit space for determining a state and an abnormal state and learns an insulation state determination model capable of determining an insulation state through the distance average;
Among the diagnosis history data for a plurality of heavy electrical devices included in the same heavy electrical device group as the heavy electrical equipment to be predicted from the database, diagnosis history data pre-diagnosed in normal and abnormal states and insulation state information for each time point of the plurality of heavy electrical devices It is entered into the failure determination model to divide each heavy electrical device into system layers, and based on the diagnosis history data pre-diagnosed in normal and abnormal states, for each heavy electrical device, each time point for important measurement items of the function level required for each system layer. a failure determination model learning unit that derives a state value and learns a failure determination model for determining a failure prediction time point according to whether or not the state value at each time point of a functional level required for each derived system layer is satisfied;
For each heavy electric machine derived from the failure determination model, a state value at each point in time of a plurality of functional levels required for each system layer is selected as an output variable, and a plurality of heavy electric machines included in the same heavy electric machine group as the received heavy electric machine to be predicted A model that analyzes diagnostic history data and output variables for , calculates a probability distribution, sets the calculated probability distribution as a prior distribution, and continuously updates the prior distribution for the output variables to a posterior distribution using a Bayesian approach. data learning execution unit; and
Diagnostic history data and heavy electric machine identification information of the heavy electrical equipment to be predicted are input to the insulation state determination model to receive insulation state information of the heavy electric equipment to be predicted by time, and insulation state information of the heavy electric equipment to be predicted by time and A heavy electrical equipment preventive maintenance recommendation system based on real-time sensing data further comprising a failure prediction time information generation unit for generating failure prediction time information by inputting diagnosis history data of the heavy electrical equipment to be predicted into a failure determination model. The method of claim 6, wherein the insulation state determination model learning unit,
Among the diagnostic history data of the received heavy electrical equipment, a plurality of major variable measurement data of the diagnostic history data pre-diagnosed in normal and abnormal states are input into the insulation judgment model,
A normal group is created based on the main variable measurement data of a plurality of diagnostic history data pre-diagnosed in a normal state among the input diagnostic history data of heavy electrical equipment,
Normalization is performed on the diagnosis history data included in the normal group to measure a probability distribution distance from the center point of the normal group to a plurality of diagnosis history data included in the normal group as a unit distance, and the measured unit distance A preventive maintenance recommendation system for heavy electrical equipment based on real-time sensing data, characterized in that a space in which the unit distance average of the is 1.0 is selected as a unit space. The method of claim 7, wherein the insulation state determination model learning unit,
Each diagnosis history data included in the normal group is obtained by subtracting the average value of the main variable measurement data of each diagnosis history data included in the normal group from the main variable measurement data of each diagnosis history data included in the normal group. Heavy electric equipment preventive maintenance recommendation system based on real-time sensing data, characterized in that it calculates a plurality of normalized vectors for each diagnosis history data by performing normalization by dividing by the standard deviation of the main variable measurement data of. The method of claim 8, wherein the insulation state determination model learning unit,
Heavy electrical equipment based on real-time sensing data, characterized in that for measuring a probability distribution distance with a plurality of diagnostic history data included in the normal group by applying an inverse matrix of a correlation matrix composed of correlation coefficients to the plurality of normalized vectors Preventive maintenance recommendation system. The method of claim 7, wherein the insulation state determination model learning unit,
Among the diagnostic history data of the received heavy electrical equipment, a plurality of major variable measurement data of the diagnostic history data pre-diagnosed in an abnormal state are input into the insulation judgment model,
An abnormal group is created based on the main variable measurement data of a plurality of diagnostic history data pre-diagnosed as abnormal among the input diagnostic history data of heavy electrical equipment,
The value obtained by subtracting the average value of the main variable measurement data of each diagnosis history data included in the normal group from the main variable measurement data of the plurality of diagnosis history data included in the abnormal group is determined as the main variable of each diagnosis history data included in the normal group. Dividing by the standard deviation of the variable measurement data to normalize the abnormal group to calculate a plurality of normalized vectors,
Based on real-time sensing data, characterized by measuring a probability distribution distance with a plurality of diagnostic history data included in the abnormal group as a unit distance by applying an inverse matrix of a correlation matrix composed of correlation coefficients to the plurality of normalized vectors. Heavy electrical equipment preventive maintenance recommendation system. The method of claim 10, wherein the insulation state determination model learning unit,
If the distribution of the unit distance of the plurality of diagnosis history data included in the normal group and the unit distance of the plurality of diagnosis history data included in the abnormal group do not differ by more than a preset standard,
Heavy electric equipment preventive maintenance recommendation system based on real-time sensing data, characterized in that it repeats the learning of the insulation state determination model until it is determined that the learning of the insulation state determination model is not completed and the difference is greater than or equal to a preset standard. The method of claim 11, wherein the insulation state determination model learning unit,
In order to increase the difference between the unit distances of the plurality of diagnostic history data included in the normal group and the distribution of the unit distances of the plurality of diagnostic history data included in the abnormal group, a plurality of predictive useful variables are selected among the main variables,
Heavy electric equipment preventive maintenance recommendation system based on real-time sensing data, characterized in that the insulation state determination model is re-learned to re-select the unit space and unit distance of the normal group based on the selected predictive useful variable. The method of claim 12, wherein the insulation state determination model learning unit,
In order to select the plurality of predictive useful variables, a two-level orthogonal array table is used to arrange variables and signal factors in the inner and outer arrays, calculate unit distances for each major variable, and calculate the signal of the calculated value. A heavy electrical equipment preventive maintenance recommendation system based on real-time sensing data, characterized in that the noise ratio is derived. The method of claim 12, wherein the insulation state determination model learning unit,
In order to determine whether each variable included in the plurality of predictive useful variables has predictive ability, a signal-noise ratio gain is calculated using a difference between signal-noise ratio averages of unit distances calculated for each variable. Heavy electric equipment preventive maintenance recommendation system based on real-time sensing data. The method of claim 14, wherein the insulation state determination model learning unit,
If the SNR gain has a negative value, it is determined that the variable has no predictive ability;
If the SNR gain has a positive value, it is determined that there is predictive ability;
Heavy electric equipment preventive maintenance recommendation system based on real-time sensing data, characterized in that the larger the value of the quantity, the higher the predictive ability. In the heavy electric equipment preventive maintenance recommendation system,
Storing diagnostic history data and heavy electric machine identification information and sensor measurement data collected by at least one sensor for a plurality of heavy electric machines generated by time-sequentially aligning diagnostic information for each heavy electric machine in a database; and
Including generating preventive maintenance recommendation information by analyzing diagnosis history data, heavy electric machine identification information, and sensor measurement data for a plurality of input heavy electric machines,
To store the data,
Managing diagnosis history data and heavy electric machine identification information for a plurality of heavy electric machines by receiving direct input from a user by using a data record file or providing a data input form;
Periodically uploading and managing sensor measurement result data from at least one sensor and identification information of a heavy electric machine for a plurality of heavy electric machines;
Based on the heavy electric machine identification information, a data group is created by grouping a plurality of heavy electric machines having a correlation with respect to at least one item of manufacturer, manufacturing period, and electric machine capacity, and diagnosis history for each heavy electric machine included in the data group Further comprising the step of managing data and sensor measurement result data;
The step of generating the preventive maintenance recommendation information,
From the database, a plurality of data sets are generated by grouping a plurality of data included in the diagnostic history data of a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted, and a plurality of data sets are generated. learning a degradation simulation model to calculate an equation having a form of a linear equation by performing regression analysis through a pre-stored regression model for each data set using the included history data as learning data;
Based on the sensor measurement result data of a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted from the database, deterioration of the heavy electric machine to be predicted according to the error trend with the optimal value set for each diagnostic item of the heavy electric machine to be predicted Calculating a weight index (γ) and reflecting the degradation weight index to a degradation simulation model; and
Performing linear regression analysis for each data set according to the degradation simulation model, generating degradation prediction information based on the performed result, and generating preventive maintenance recommendation information according to a degradation prediction time included in the degradation prediction information Including more,
Among the diagnostic history data for a plurality of heavy electrical equipment, diagnosis history data pre-diagnosed in normal and abnormal states is received as learning data to learn an insulation state determination model and a failure determination model, and the learned insulation state determination model and failure determination Further comprising generating soundness analysis information including failure prediction time information of the heavy electrical equipment to be predicted using the model,
The step of reflecting the degradation weight index to the degradation simulation model,
The sensor measurement result data is analyzed to calculate the error with the optimal value set for each diagnosis item of the heavy electric machine to be predicted, and the weight index γ having a higher value as the magnitude of the error increases, the equation as shown in Equation 2 A weight is applied to and reflected in the degradation simulation model,
The step of generating the preventive maintenance recommendation information,
Calculate the weight δ at the time of prediction of degradation included in the degradation prediction information and the weight ε at the time of prediction of failure included in the health analysis information, and use the weighted degradation prediction information and health analysis information to obtain the preventive maintenance recommendation information create,
The weight δ and the weight ε are given the same value of 1 at first, and the weight in the direction of relatively increasing accuracy depending on whether or not the time information output by inputting the diagnosis history data matches the failure history data that occurred in the past. The weights are calculated by performing an update by adjusting the ratio of δ and the weight ε, respectively,
The preventive maintenance recommendation information calculates the weight δ at the time of prediction of deterioration included in the deterioration prediction information and the weight ε at the time of prediction of failure included in the soundness analysis information, and the average value of the weighted values for each time point is obtained. A heavy electrical equipment preventive maintenance recommendation method based on real-time sensing data, characterized in that it is generated as a standard.
[Equation 2]

The step of learning the degradation simulation model of claim 16,
Grouping is performed on a plurality of data included in the diagnosis history data to generate a plurality of data sets;
Characterized in that the degradation simulation model is trained to calculate an equation having the form of a linear equation by performing regression analysis through a pre-stored regression model for each data set using the history data included in each generated data set as learning data. A method for recommending preventive maintenance of heavy electrical equipment based on real-time sensing data. The step of learning the degradation simulation model of claim 17,
By comparing the analysis result value through the equation calculated by the pre-stored regression model and the diagnosis history data, a regression model having an analysis result value having a relatively minimum error with the diagnosis history data is a regression model of the corresponding data set. A method for recommending preventive maintenance of heavy electrical equipment based on real-time sensing data, characterized in that designing a degradation simulation model composed of a plurality of regression analysis models. delete delete 17. The method of claim 16, wherein generating the soundness analysis information comprises:
Among the diagnostic history data for a plurality of heavy electric machines included in the same heavy electric machine group as the heavy electric machine to be predicted from the database, diagnosis history data pre-diagnosed in normal and abnormal states is input as learning data, and a unit space is set, and a normal Calculating an average distance in the unit space for determining a state and an abnormal state, and learning an insulation state determination model capable of determining an insulation state through a distance average;
Among the diagnosis history data for a plurality of heavy electrical devices included in the same heavy electrical device group as the heavy electrical equipment to be predicted from the database, diagnosis history data pre-diagnosed in normal and abnormal states and insulation state information for each time point of the plurality of heavy electrical devices It is entered into the failure determination model to divide each heavy electrical device into system layers, and based on the diagnosis history data pre-diagnosed in normal and abnormal states, for each heavy electrical device, each time point for important measurement items of the function level required for each system layer. deriving a state value and learning a failure determination model for determining a failure prediction time point according to whether or not the state value for each time point of a function level required for each derived system layer is satisfied;
For each heavy electric machine derived from the failure determination model, a state value at each point in time of a plurality of functional levels required for each system layer is selected as an output variable, and a plurality of heavy electric machines included in the same heavy electric machine group as the received heavy electric machine to be predicted Analyzing diagnostic history data and output variables for , calculating a probability distribution, setting the calculated probability distribution as a prior distribution, and continuously updating the prior distribution for the output variable as a posterior distribution using a Bayesian approach. ; and
Diagnostic history data and heavy electric machine identification information of the heavy electrical equipment to be predicted are input to the insulation state determination model to receive insulation state information of the heavy electric equipment to be predicted by time, and insulation state information of the heavy electric equipment to be predicted by time and A method for recommending preventive maintenance of heavy electric equipment based on real-time sensing data, further comprising the step of generating failure prediction time information by inputting diagnostic history data of the heavy electric equipment to be predicted into a failure determination model. 22. The method of claim 21, wherein the step of learning the insulation state determination model,
Among the diagnostic history data of the received heavy electrical equipment, a plurality of major variable measurement data of the diagnostic history data pre-diagnosed in normal and abnormal states are input into the insulation judgment model,
A normal group is created based on the main variable measurement data of a plurality of diagnostic history data pre-diagnosed in a normal state among the input diagnostic history data of heavy electrical equipment,
Normalization is performed on the diagnosis history data included in the normal group to measure a probability distribution distance from the center point of the normal group to a plurality of diagnosis history data included in the normal group as a unit distance, and the measured unit distance A method for recommending preventive maintenance of heavy electrical equipment based on real-time sensing data, characterized in that a space in which the unit distance average of s is 1.0 is selected as a unit space. 23. The method of claim 22, wherein the step of learning the insulation state determination model,
Each diagnosis history data included in the normal group is obtained by subtracting the average value of the main variable measurement data of each diagnosis history data included in the normal group from the main variable measurement data of each diagnosis history data included in the normal group. A method for recommending preventive maintenance of heavy electrical equipment based on real-time sensing data, characterized in that a plurality of normalized vectors for each diagnosis history data are calculated by performing normalization by dividing by the standard deviation of the main variable measurement data. 24. The method of claim 23, wherein the step of learning the insulation state determination model,
Heavy electrical equipment based on real-time sensing data, characterized in that for measuring a probability distribution distance with a plurality of diagnostic history data included in the normal group by applying an inverse matrix of a correlation matrix composed of correlation coefficients to the plurality of normalized vectors How to recommend preventive maintenance. 23. The method of claim 22, wherein the step of learning the insulation state determination model,
Among the diagnostic history data of the received heavy electrical equipment, a plurality of major variable measurement data of the diagnostic history data pre-diagnosed in an abnormal state are input into the insulation judgment model,
An abnormal group is created based on the main variable measurement data of a plurality of diagnostic history data pre-diagnosed as abnormal among the input diagnostic history data of heavy electrical equipment,
The value obtained by subtracting the average value of the main variable measurement data of each diagnosis history data included in the normal group from the main variable measurement data of the plurality of diagnosis history data included in the abnormal group is determined as the main variable of each diagnosis history data included in the normal group. Dividing by the standard deviation of the variable measurement data to normalize the abnormal group to calculate a plurality of normalized vectors,
Based on real-time sensing data, characterized by measuring a probability distribution distance with a plurality of diagnostic history data included in the abnormal group as a unit distance by applying an inverse matrix of a correlation matrix composed of correlation coefficients to the plurality of normalized vectors. Method of recommending preventive maintenance of heavy electrical equipment. 26. The method of claim 25, wherein the step of learning the insulation state determination model,
If the distribution of the unit distance of the plurality of diagnosis history data included in the normal group and the unit distance of the plurality of diagnosis history data included in the abnormal group do not differ by more than a preset standard,
It is determined that the learning of the insulation state determination model is not completed and the learning of the insulation state determination model is repeated until the difference is greater than or equal to a preset standard. 27. The method of claim 26, wherein the step of learning the insulation state determination model,
In order to increase the difference between the unit distances of the plurality of diagnostic history data included in the normal group and the distribution of the unit distances of the plurality of diagnostic history data included in the abnormal group, a plurality of predictive useful variables are selected among the main variables,
Heavy electric equipment preventive maintenance recommendation method based on real-time sensing data, characterized in that for re-learning the insulation state determination model to re-select the unit space and unit distance of the normal group based on the selected predictive useful variable. 28. The method of claim 27, wherein the step of learning the insulation state determination model,
In order to select the plurality of predictive useful variables, a two-level orthogonal array table is used to arrange variables and signal factors in the inner and outer arrays, calculate unit distances for each major variable, and calculate the signal of the calculated value. A heavy electrical equipment preventive maintenance recommendation method based on real-time sensing data, characterized in that the noise ratio is derived. 28. The method of claim 27, wherein the step of learning the insulation state determination model,
In order to determine whether each variable included in the plurality of predictive useful variables has predictive ability, a signal-noise ratio gain is calculated using a difference between signal-noise ratio averages of unit distances calculated for each variable. A method for recommending preventive maintenance of heavy electrical equipment based on real-time sensing data. The method of claim 29, wherein the step of learning the insulation state determination model,
If the SNR gain has a negative value, it is determined that the variable has no predictive ability;
If the SNR gain has a positive value, it is determined that there is predictive ability;
Heavy electrical equipment preventive maintenance recommendation method based on real-time sensing data, characterized in that the higher the value of the quantity, the higher the predictive ability.

Images