RU2707423C2 - Method and system for diagnostics of industrial facility - Google Patents

Abstract

FIELD: monitoring systems.

SUBSTANCE: invention relates to monitoring and diagnosing the state of an industrial facility. Said technical result is achieved due to the fact that a system for diagnosing an industrial facility is developed, comprising a data collection unit configured to collect data from a set of sensors of an industrial facility; an industrial object model block configured to simulate an industrial object; an analysis unit configured to analyze the state of the industrial object based on data obtained from the data collection unit and the industrial object model; wherein the analysis unit is configured to conclude on the normal or abnormal operation of the industrial object based on the analysis; wherein the analysis unit is configured to receive data on changes made to the industrial object, and to command the model unit to change the model in accordance with the changes.

EFFECT: high accuracy of diagnosing an industrial facility in terms of detecting pre-fault conditions.

6 cl, 3 dwg

Description

The technical field to which the invention relates.

The invention relates to the field of monitoring and diagnosing the state of an industrial facility, in particular tracking the residual resource of elements of an industrial facility.

The level of technology.

A known method for predicting the need for maintenance of equipment (patent US 2009/0037206). The method described in the patent involves determining the time of failure of industrial equipment based on the assessment of the mutual influence of the parameters using mathematical, analytical or empirical models. The disadvantages of the described method is the inability to predict the influence of the processes described by the sensor signals implicitly or rarely encountered during the observation period. The fundamental difference between the proposed solution and the one described in US 2009/0037206 is the use of finite element models to take into account the influence of factors not directly measured on industrial equipment itself.

A solution is known that describes adaptive remote maintenance of rolling stock (US 8849732 B2, 09/30/2014), in which maintenance of rolling stock is provided by machine learning the rules. Existing rules or models are automatically updated. Machine learning is used to create a more efficient set of rules. Rules can be replaced, generalized or otherwise adapted based on the processing by the dispatchers of the results of existing rules. The acceptance or rejection of an event by the dispatcher is used as reference knowledge for machine learning controlled by the dispatcher of the new rule. Machine learning uses feedback from the controller to update the rule set. However, this decision, although it speaks of predicting future malfunctions, does not say anything about how it is implemented.

Known is selected as a prototype solution that describes the system and method for advanced tracking of the state of the system (US 7756678 B2, 07/13/2010). The known method for improved monitoring of the state of the system includes the use of many auto-associative neural networks to determine the estimates of actual values measured by at least one sensor in at least one of the many modes of operation; determining a difference between the estimated measured values and the actual values measured by at least one sensor; and combining the differences using a fuzzy controlled mixer model; fault diagnostics on combined differences; and determining a change in system performance by analyzing combined differences. A warning is provided if necessary. The smart sensor system includes an on-board processor for implementing the method according to the invention.

However, this decision does not say anything about the actions of the method in the case of updating elements of a controlled system. Also in the known method does not use an engineering model of the diagnosed industrial facility that describes its operation through the laws of the mutual influence of the parameters.

Disclosure of the invention.

In one aspect of the invention, a system for diagnosing an industrial facility is disclosed, comprising:

- a data collection unit configured to collect data from a set of sensors of an industrial facility;

- block model of an industrial facility, configured to simulate an industrial facility;

- an analysis unit configured to analyze the state of the industrial facility based on data received from the data collection unit and the model of the industrial facility;

moreover, the analysis unit is configured to make a conclusion about the normal or abnormal functioning of an industrial facility based on the analysis;

moreover, the analysis unit is configured to receive data on changes made to the industrial facility, and to command the model unit to change the model in accordance with the changes.

In additional aspects, it is disclosed that data on changes made to the industrial facility contain data on interference in the operation of the industrial facility in the elimination of failures and precautionary conditions; the analysis unit is configured to determine an element requiring repair or replacement leading to an abnormal operation of an industrial facility; an analysis unit is configured to modify the model based on data on repair or replacement of an element; the industrial model is a system of equations describing the dependence of the values of the determined parameter of the industrial object on the input data obtained from the industrial object at least by sensors, and the equation contains four groups of correction factors:

- coefficients of the physical model;

• general adjustment factors determining the effect of degradation and external factors on the operation of all elements of an industrial facility;

• individual correction factors for a given element of an industrial facility, which determine the influence of external factors on the operation of a given element of an industrial facility;

• temporary correction factors that determine the effect of repairs on the operation of a given element of an industrial facility.

In additional aspects, it is disclosed that the analysis unit is configured to zero out all temporary correction factors for a given element of an industrial facility when it receives a message about replacing a given element of an industrial facility, while the resettable time coefficients are stored in memory, while the analysis unit is configured to obtain new values temporary correction coefficients from the replaced element, comparing them with the saved coefficients and deciding on the lack of repair this element of an industrial facility.

In another aspect of the invention, a method for diagnosing an industrial facility is disclosed, comprising the steps of collecting sensor data of an industrial facility from the hardware-software complex characterizing the physical parameters of the industrial facility; create a model of an industrial facility; analyze the state of the industrial facility on the basis of the collected data and the model of the industrial facility; moreover, when analyzing, they make a conclusion about the normal or abnormal functioning of an industrial facility; moreover, when analyzing, they accept data on changes made to the industrial facility, and change the model in accordance with the changes made.

In additional aspects, it is disclosed that data on changes made to the industrial facility contain data on interference in the operation of the industrial facility in the elimination of failures and precautionary conditions; determine the element requiring repair or replacement, leading to the abnormal functioning of the industrial facility; change the model based on data on repair or replacement of an element; the industrial model is a system of equations describing the dependence of the values of the determined parameter of the industrial object on the input data obtained from the industrial object at least by sensors, and the equation contains four groups of correction factors:

- coefficients of the physical model;

- general correction factors determining the effect of degradation and external factors on the operation of all elements of an industrial facility;

- individual correction factors for a given element of an industrial facility, which determine the influence of external factors on the operation of a given element of an industrial facility;

- temporary correction factors that determine the effect of repairs on the operation of a given element of an industrial facility.

In additional aspects, it is disclosed that all temporary correction factors for a given element of an industrial object are canceled upon receipt of a message about replacing a given element of an industrial object, while the zeroed time coefficients are stored in memory, and new values of temporary correction coefficients are obtained from the replaced element, compared with stored coefficients and decide on the lack of repair of this element of an industrial facility.

The main tasks solved by the claimed invention are to determine the technical condition of an industrial facility in terms of failures, pre-failure states and operational modes violations according to the data of microprocessor-based systems for collecting data on the functioning of an industrial facility, determining the residual resource of elements of an industrial facility, automating the process of developing diagnostic algorithms for an industrial facility using models, as well as adaptation of the claimed method and system to the replacement of industrial elements The object is.

The essence of the invention lies in the fact that using a set of sensors located in the elements of an industrial facility, the state of the industrial facility is monitored. Further, the data from the sensors are processed to diagnose the condition of the industrial facility. The data from the sensors are compared with the data generated by a pre-developed model of the industrial facility, based on the results of the comparison, the conclusion is made about the serviceability of the industrial facility and its elements, about the need to repair and / or replace the elements of the industrial facility. A feature of the claimed invention is the use of an engineering model of an industrial facility, the presence of feedback that changes the model of an industrial facility according to interference with the industrial facility when eliminating failures, precautionary states and when performing super-cycle work on scheduled types of services (latent failures).

The technical result achieved by the solution is to increase the accuracy of diagnostics of an industrial facility in terms of identifying pre-failure conditions.

A brief description of the drawings.

FIG. 1 shows a functional diagram of the proposed diagnostic method.

FIG. 2 shows a method for obtaining the predicted parameter value of an industrial facility.

FIG. 3 shows a general outline of the claimed diagnostic system.

The implementation of the invention.

In general, a technical solution for servicing industrial facilities consists of a microprocessor-based data collection system on the operation of an industrial facility (hereinafter referred to as the ISU), a diagnostic system, and a resource management system (ERP system).

ISU consists of a set of sensors installed on the most important elements of the analyzed industrial facility, input signal converters, output signal converters and signal processing facilities. A key property of the signal processing tool is the ability to accumulate information on the values of the input and output signals in a given time period in its own memory, as well as transmit the accumulated information via a removable storage device or contactless via GPRS or Wi-Fi channel to the diagnostic system.

The developed Clover PMM diagnostic system consists of a hardware-software complex that receives data from the local government and which, based on these data and engineering model data, diagnoses an industrial facility. Data may

 arrive both through wired transmission and wireless transmission, also data can be loaded into the complex via a data storage device. One of the features of the diagnostic system is the possibility of self-training on the basis of an expert assessment of the identified anomalies, that is, the system is trained by an expert who indicates whether the identified anomaly is a sign of malfunction of an industrial facility (incident) or not. The diagnostic system is designed for automatic processing of MSU data and carries out data via GPRS, Wi-Fi, uploads it to the database, identifies anomalies using an engineering model, sends the identified anomalies for verification, as well as generates comments on identified incidents and sends them to the ERP system .

An ERP system can have a different structure depending on the specifics of an industrial facility, however, its key features from the point of view of the proposed technical solution are the ability to form work orders, record the fact that these orders are completed, and record information on the allocation of components (hereinafter - MPI) under execution of orders and transmit information about the work performed and decommissioned MPI to the diagnosis system.

The differences of the proposed technical solutions from known from the prior art are:

1. The ability to fill in the missing data areas by modeling the operation of an industrial facility with a finite element model

2. The presence of a set of criteria and a range of their boundary values, dynamically changed according to historical data on previously occurring failures and pre-failure conditions.

3. The presence of feedback that changes the predictive analysis model for data on interference in the operation of an industrial facility when eliminating failures, pre-failure conditions and when performing super-cycle work on planned types of services (latent failures).

DETAILED DESCRIPTION OF EMBODIMENTS.

Terms and Definitions:

ERP is a process management system.

An anomaly is a situation in which the values of the parameters of an industrial facility differ from normal for a given operating mode of an industrial facility.

Approximation - a description of the empirically observed pattern of parameter change by a mathematical function that describes the pattern with a given accuracy.

An incident is any case of abnormal operation of an industrial facility.

MPI - reusable materials - elements of an industrial facility that are dismantled during maintenance and repair and transferred to repair shops for further repairs.

Predictive analysis model - a model for predicting breakdowns of maintainable equipment, the operation of which can be adequately described by its sensors, as well as using calculation methods when constructing an engineering model.

MSU - microprocessor control system - a set of hardware and software that manage an industrial facility and accumulate information about its work for a certain period of time.

Violation of operating modes (NRE) is a type of incident characterized by improper operation of a working industrial facility due to incorrect actions by a human operator.

Failure is a type of incident characterized by the transition of an industrial facility from an operational state to an inoperative one.

The incident search rule is an automated incident search algorithm created by developers.

Pre-failure condition is a type of incident characterized by the transition of an industrial facility to a malfunctioning state while maintaining operability.

Hardware-software complex - a device or system consisting of software and hardware, a particular example may be a microprocessor system, a server, a specialized integrated circuit, a general-purpose computer, etc.

Industrial facility - an object related to the field of transport, engineering, production, private examples include a train, an airplane, a factory, a production line, a conveyor, a power plant, an industrial pump, etc.

An element of an industrial facility - a component, block, part, assembly, unit, which is part of a facility that is more complex in design and functions, examples are a train or airplane engine, a separate line or conveyor machine, assembly unit, etc.

In terms of the method, the task is solved, and the technical result is achieved by the fact that the diagnosis of an industrial facility (a flowchart of the method is shown in Fig. 1) is made by comparing (step 6) the “predicted” (calculated) values of the industrial facility operation parameters obtained at the stage 4 during the formation of the engineering model (stage 2), with real values obtained in stage 5 from the data received by the LSG in stage 1, and the identification of deviations that exceed statistically acceptable boundaries (stage 8) and do not go beyond these boundaries (e ap 7). Data sections in which parameter values went beyond the permissible limits are marked as “abnormal” (step 8) and sent to the expert for verification (confirmation) (step 9). The expert’s assessment serves to train the model (step 10), which further recognizes similar cases as an anomaly or not an anomaly, according to the decision of the expert. If the expert confirms the existence of an anomaly, at step 12, a record about the malfunction of the industrial facility is created in the diagnostic card, then at step 13 it is sent to the ERP system, and the data ratio is entered as inadmissible in the statistical database of the predictive analysis model of the industrial facility during training models. If the expert does not recognize the presence of an anomaly, a piece of data is entered into the model as a normal mode of operation. From the ERP-system, information about the replaced equipment goes to the predictive analysis model to account for the replacement of elements of an industrial facility. Since neither the repaired nor the new element can have absolutely identical characteristics in comparison with the replaced element, the system may erroneously consider that anomalous changes have occurred in the operation of the industrial facility, therefore it is important to take into account the fact of changing the equipment of the industrial facility.

One of the beneficial effects of the implementation of the claimed technical solution is the creation of an information base for the transfer of devices having on-board microprocessor systems for servicing according to the current technical condition.

The predictive analysis model of an industrial facility (2) calculates the value of the desired output parameter based on input parameters (obtained from the LSG data (1)) as well as the dependencies of parameter values obtained analytically (when creating an engineering model of an industrial facility) and statistically (when data on work for a long period of time in a virtual neural network). At the same time, the data on the parameter value accumulated in the database contain a mark on the belonging of the data section to the normal or abnormal mode of operation of the industrial facility, obtained in the verification process (9) by a technical expert. Additionally, the data contains marks:

- on the change of an element of an industrial facility during repairs (11) (zeros out the statistically accumulated data on the operation of an element of an industrial facility to eliminate the erroneous acceptance of differences in the data of a new element of an industrial facility as an anomaly);

- on the relation of the identified incident (provided that the anomaly is marked as an incident) to one of the types: NRE, failure or pre-failure condition. The accumulation of statistical data on the belonging of the value of the parameter difference to a certain type of incident allows you to build a function of the degradation of elements of an industrial facility;

- About the residual resource of elements of an industrial facility.

In the absence of one or more parameters necessary for the calculation, their values can be replaced by the readings of the "Virtual Sensor" - values obtained on the basis of the mnemonic dependences of the desired parameter on the values of the available parameters, obtained by modeling the parameter values with a separate engineering model of an industrial facility of high accuracy. The resulting mnemonic dependencies are subsequently introduced into the engineering model of the diagnosed elements of an industrial facility to replace the readings of missing sensors.

The engineering model of the diagnosed elements of an industrial facility (Fig. 2) consists of models of three levels:

1. CAD model. A three-dimensional model or a set of models built on the basis of design documentation and reproducing the geometric parameters of the diagnosed industrial facility.

2. CAE model. An analytical model of an industrial facility, built on the basis of technological documentation and reproducing the physical properties of the materials of which the model is made.

3. The physical model of an industrial facility, built on the basis of operational documentation and taking into account the nature of the physical processes that take place in an industrial facility during its operation.

As an output, when creating a model, the work parameter of an industrial facility is selected that most accurately describes the change in its objective function when the input parameters change.

Depending on the nature of the parameters obtained from the ISU, they are sent to the input, CAE- or physical model, changing the corresponding values of the parameters laid down earlier. The value of the output parameter is calculated based on the simulation of processes occurring inside an industrial facility using the finite element method.

The final model of operation of an industrial facility is a system of equations describing the dependence of the values of the output parameter on the input and having 4 groups of correction factors:

1. The coefficients of the physical model. The values set at the modeling stage and not changing during the training of the model. The physical law of the mutual influence of the parameters is determined.

2. General correction factors. Values that change identically for all observed elements of the same industrial object, depending on the absolute (since the start of observations) and relative (since the start of operation of this unit) time values. The influence of degradation and external factors on the operation of the entire industrial facility is determined.

3. Individual adjustment factors. Values that change for a given element of an industrial object over time, but not depending on the fact of a change in an element of an industrial object. The influence of external factors on the operation of a given element of an industrial facility is determined.

4. Temporary adjustment factors. Values that change for a given element of an industrial facility over time and depend on the fact of repair of an element of an industrial facility. The influence of repair on the operation of this element of an industrial facility is determined.

At the same time, the coefficient belongs to group (1) at the stage of model creation, and to groups (2 ... 4) - automatically in the process of statistical analysis of the model park work histories.

The determination of the values of the coefficients (2-4) is performed with the accuracy required for a given type of industrial plant element based on a statistical analysis of the available information on the operation of a known-good model (sections of the model’s working time not marked by the rules or by the operator as “anomalies” or “incidents”) with finding the most plausible solution. All found coefficients are initially determined as temporary correction factors, after which their belonging to groups (2) or (3) can be determined on the basis of the analysis of repeatability of values for different elements of an industrial facility before and after repair:

- when repeating one or several (but not all) values of the temporary correction factor in the data areas before and after at least two repairs, these coefficients are taken as individual correction ones.

- when repeating the values of individual correction factors for 60% + 1 element of an industrial facility, subject to the normal distribution of values, the coefficients are taken as the general correction factors for all elements of the same type of industrial object.

Adaptation of the model to the change of an element of an industrial facility during repair is carried out by zeroing all temporary correction factors for a given element of an industrial facility when a message is received about the replacement of an element of an industrial facility with preservation of their previous values in the history of the model. New values of time correction factors are determined based on observation of the model. After obtaining statistically reliable (both by the number of measurements and by the normality criterion), the new matrix of values of the time correction factors is compared with the old one. If the new matrix coincides with the old with a given accuracy, a message is generated about the fictitious repair of an industrial facility.

To reduce the diagnostic error, an expert is trained to model the model, which occurs in two ways:

1. Manual indication of anomalies by an expert by designating the beginning and end of the data area on which the anomaly was observed, as well as the anomaly's belonging to a group of elements of an industrial facility. For data sections marked by the user as an anomaly, absolute, relative values of all parameters characterizing the operation of the indicated group of elements of the industrial facility (as well as their rate of change, dispersion values and correlation coefficients) are estimated with similar values for sections not marked as an anomaly by the user. Specific values (deviating from the mathematical expectation by more than 3 variance values) obtained for this section are entered into the model database as new tolerances, and the names of the parameters for which they are obtained are added as a new model.

For example, if, when analyzing the operation of a centrifugal supercharger, the operator detects insufficient productivity (a slow increase in the rotational speed and pressure at the outlet at normal values of other parameters of the supercharger), he marks the time interval with low productivity, indicating its start and end times on the parameter graphs, as well as equipment at risk (“supercharger bearing”). For the marked time section, the search algorithm compares alternately the absolute, relative values of the input parameters (in the case of a supercharger, this is the motor current, rotor speed, inlet air temperature and pressure), as well as their rate of change, their dispersion values and cross-correlation coefficients with values obtained in areas not marked as abnormal. At the same time, parameters that have an atypical value (in this case, this is the rate of change of the motor current, the rotational speed of the compressor shaft, which are less than statistically normal) are marked as abnormal. At the same time, the influence of each parameter on the output parameter is estimated using the correlation coefficient (since the rate of change of parameters is estimated, the correlation coefficient will be calculated for the rate of change of pressure of the injected air). The parameter having the maximum deviation from typical values is recognized as the root cause of the anomaly (in this case, the stopping speed of the supercharger rotor (the rate of change of its rotational speed when the motor current is zero). As the anomaly indicates the anomaly, the “supercharger bearing” , the model automatically associates the exceeded rate of change of the rotational speed of the supercharger shaft when the motor current is zero with the precautionary state of the bearing.

2. Expert verification of the anomaly found by the model. Belonging to the anomaly of the data area marked by the model as an anomaly is confirmed or refuted by the expert during the verification process.

If the anomaly is confirmed by the user, the deviation of the actual value of the output parameter from the forecast for this section is marked as unacceptable, and if it is exceeded by the data in the future, such sections will be marked as an incident.

When the user refutes the anomaly, the deviation is marked as permissible, and in the future, areas with similar deviations as anomalies are not noted.

The residual resource of an element of an industrial facility is estimated by solving the problem of finding the most likely value of a parameter that best describes historical data for the period of observation using machine learning methods. If it is necessary to use a parameter (not measured at the industrial facility itself) for estimating the residual life of the parameter, the model is supplemented by a separate unit that solves the problem of determining the empirical regularity of changes in the calculated parameter at given points.

Depending on the amount of data and the nature of their change, the task of predicting the residual resource can be solved by one of the methods described below:

1. With a sufficient amount of data and sufficient discreteness, the estimation is made on the basis of the empirical distribution function of the failure frequency depending on the frequency of detected anomalies.

2. With a sufficient amount of data and insufficient discreteness, the assessment is made on the basis of an analysis of the discrepancy between the predicted and real values of the parameters characterizing the operation of the node by the trend analysis method. If there are several trends, the search for a working trend is carried out by the method of logistic regression.

3. If there is not enough data, the missing pieces of data are restored by modeling the operation of the diagnosed industrial facility with an engineering model. Based on the data obtained, analytical algorithms are optimized, after which the trends indicating degradation are identified in the reconstructed data set. Depending on the nature of the possible failure, the identified trends are approximated and superimposed on the empirical distribution function (with a gradual onset of failure) or the time it takes to reach the threshold level (with a discrete function of the onset of failure) is calculated.

As regards the industrial facility's diagnostic system, the task is solved, and the technical result is achieved using the data acquisition and processing system shown in FIG. 3, the system consists of a hardware-software complex, in particular, a server 309 of the diagnostic system, equipment 301, consisting of a microprocessor-based control system (hereinafter referred to as the ISU) (2) and a transmitter 303, a workstation 306 of a diagnostician and an ERP system 314 of the owner of an industrial facility or service company.

ISU 302 of the diagnosed industrial facility consists of a specialized computer of industrial design, a system of sensors and signal converters and control devices (if any) and, in addition to managing the industrial facility, accumulates information about its operation. The accumulated information is transmitted to the server 309 of the diagnostic system using the transmitter 303 through the channel 305 GPRS / Wi-Fi. If it is impossible to transfer data via GPRS / Wi-Fi, the information is read using a portable data carrier 304 during the scheduled maintenance procedures for the industrial facility and is downloaded to the diagnostics system server 309 through the diagnostics workstation 306, which is a standard computer with specialized software that has access to the Interner / Ethernet network through which data is transferred to the server 309 of the diagnostic system.

On the server 309 of the diagnostic system, by means of the data analysis unit 310, data is processed by the ISU 302 using the predictive analysis model 312, as a result of which the anomaly unit 311 detects anomalies that are sent for verification to the data verification unit 308 contained in the diagnostics workstation 306 and / or an expert. Both the anomalies confirmed and refuted during the verification process by the diagnostician and expert are transmitted to the server 309 of the diagnostic system for training the predictive analysis model 312. The anomalies confirmed during verification on the server 309 of the diagnostic system are marked in the block 313 of incidents as incidents and are sent to the ERP system 314 for generating comments via block 315 comments. In the ERP system 314, based on the comments issued by the block 315 comments, block 316 orders are formed orders for which from the warehouse are allocated reuse materials (hereinafter - MPI) 317.

After setting marks on the performance of work and installation of elements of the industrial facility in the ERP system using the block 318, the information on the work performed and the change of elements of the industrial facility is sent to the server 309 of the diagnostic system to verify the fact of the work and training the model.

DESCRIPTION OF EMBODIMENTS.

Option 1. Locomotive.

Modern locomotives are equipped with on-board microprocessor systems (MCU) designed to control the operation of power electric transmission and auxiliary equipment of the locomotive. In the general case, an ISU of a locomotive consists of a control rack, a set of sensors, controls (relays and servos) and a display module (DM) for interacting with the driver. In addition to the basic functionality, the design of most ISUs allows the accumulation and storage of information on the operation of locomotive equipment in the memory of the display module with the possibility of its further reading via portable flash drives or remote reading via wireless GPRS or Wi-Fi networks (depending on the version of the ISU) .

Based on the results of the decryption of the ISU data, incidents are detected (events in the operation of the locomotive that indicate a malfunction). They are divided into violations of operating modes (NRE) and pre-failure conditions and are entered into the ERP-depot system for taking measures when locomotives call for the nearest service or repair.

The main drawback of the existing system is the dependence of the quality of incident detection on the competence of the diagnostic operator. To reduce the influence of the diagnostician on the decryption result, automated diagnostician workstations (AWS MSU) with the function of automated incident search according to the specified diagnostic algorithms are introduced.

However, the diagnostic algorithms built into the AWS of the ISU used to have one significant drawback: in order to introduce new diagnostic algorithms into the AWS, it was necessary to refine the AWS of the ISU itself. As a result, with the accumulation of diagnostic experience and the emergence of new types of NRE and pre-failure conditions, the weight of incidents found using algorithms decreases and the problem of manually searching for incidents reappears. In the latest version of the AWS MASD, this problem was partially solved by introducing a built-in language for writing diagnostic algorithms, but user algorithms existed only in the local version of the AWS and could not be extended to all locomotives of the series.

At the same time, the development of computer technology has led to the emergence of new information processing methods, one of which is the BigData algorithms, specially created for the efficient processing of large amounts of data and machine learning methods. These methods allowed, on the basis of statistical processing of historical data, not only to reveal hidden relationships, but also to determine algorithms for the mutual influence of parameters.

The applicant has created a solution for the analysis of locomotive LSG data using BigData methods in order to identify hidden relationships between equipment operation parameters, determine trends in their changes (trends) and predict technical condition. As a pilot series of locomotives, it was decided to choose diesel locomotives 2 (3) TE116U, since at the time of the start of the experiment it was quite widespread, had a significant set of diagnostic data and, at the same time, considerable diagnostic experience was gained from it.

At the developed hardware-software complex, primary data analysis is performed using diagnostic algorithms and their comparison with the readings obtained for similar conditions on the model of locomotive equipment. Any deviation of the actually observed parameter values from the model readings is recognized as an anomaly - a previously unknown operating mode. The presence of diagnostic algorithms is explained both by the need to facilitate the work of diagnostic groups during the training and deployment of BigData methods, and by highlighting incidents from the normal operating modes of a locomotive. After processing the data, incidents identified using the model are sent to the diagnostician for verification. If he confirms that the anomaly is an incident (the type of incident is also indicated in the confirmation process - NRE or pre-failure condition, name of the failed equipment), then the anomaly is sent to the expert for verification, after which it is recorded in the model as an incident. Otherwise, the anomaly is recorded as the normal operating mode of the locomotive.

Incidents identified during data analysis are included in the locomotive comment list. After the integration of the hardware-software complex with the diesel locomotive monitoring system, information exchange is organized between the systems, in which incidents are sent to the diesel locomotive monitoring system in an automated mode, and after the repair is completed, data on repaired and replaced equipment is imported from the diesel locomotive monitoring system to the hardware-software complex for adjustment equipment models (in fact, the model of repaired equipment is separated in order to exclude the influence of data before repair ont to her work).

A preliminary study showed that, with all its flexibility, machine learning methods have a number of specific requirements for the information being processed:

1. The need for continuous information over a long period. The data should not contain time sections in which information on the operation of the equipment was not received.

2. The need for selection of the analyzed parameters. They should be clearly divided into input and output (the model was described by the principle of a black box), and the influence of inputs on the outputs should occur only through the described node.

3. Unimodality of the described process. If there are several processes in the node, the node must be divided into several models.

Traction electric motors (TED) were chosen as the primary objects for modeling, since they satisfy several selection criteria at once:

- TED is one of the most expensive types of equipment and the failure rate is high;

- TEDs are equipped with a sufficient number of sensors;

- The characteristics of TED are described in sufficient detail both in technical documentation and in scientific papers.

Based on the parameters available in the ISU, a TED model was built that, to ensure unimodality (requirement 3), combined three separate models:

1. The electrical model. The dependence of the current on the voltage at the terminals of the TED.

2. Electromechanical model. The dependence of the current TED on the speed of the wheelset.

3. Model of rheostatic braking. The dependence of the voltage at the terminals of the TED on the speed of the wheelset.

Subsequently, in a separate experiment, models (1) and (2) are combined into one.

Since the available volume of ISU data does not meet the requirements for training the model (even after connecting the data of the ASK and BRPD systems), additional data preparation was performed to ensure proper accuracy of work:

- TED operation modes were divided according to the field weakening mode. The sign of the transfer of the locomotive into rheostatic braking was used to separate the traction and rheostatic braking modes;

- signs of the operation of train contactors, motor circuit breakers and fuses were used to screen out modes in which the TED was disabled;

- in order to combat interference for all models, an additional screening of data sections was performed for 120 seconds. after changing the position of the controller (RMB) or 3 seconds. from the moment of turning on (turning off) the motor-fans, as well as the areas where the brake compressor was turned on.

By comparing the ISU data of diesel locomotives with the model, anomalies are detected.

Then it was decided to expand the study by building similar models for the rectifier installation, traction generator (TG) and the locomotive exciter, introducing a series of feedbacks between the nodes.

Thus, the application of BigData methods and machine learning to a model built on the basis of data from the local self-government of diesel locomotives made it possible to reveal hidden patterns in changing the parameters of the operation of diesel locomotive equipment, making it possible to detect cases of hidden non-scheduled repairs.

Option 2. Turbogenerator.

The developers implemented a project to analyze the work and assess the technical condition of the TZFP-220-UZ turbogenerator.

A turbogenerator is the main equipment of power units of thermal power plants (TPP) along with a power boiler, steam turbine and power transformer. For this reason, an emergency shutdown of a turbogenerator entails an emergency shutdown of the entire power unit, and the costs of restoring repair work.

Data collection of instrumentation and automation, an automated process control system, analog instruments; lists and defect magazines, operating instructions showed sufficient equipment of the investigated object with sensors for further construction of the engineering model and the predictive state analysis model. The failure data (insulation breakdown on the stator winding of the generator stator 23H) recorded in the act of technological violations served as the basis for marking the received telemetry set (data set) and training the developed model to identify similar malfunctions in the early stages of their development.

Based on the analysis of the collected data, a tree structure for the equipment was built with the binding of the controlled parameters to the nodes. The behavior of various values of the sensors until the failure occurs (phase currents, voltages, winding and active steel temperatures, etc.) is analyzed and key parameters are determined from the entire telemetry data sample provided.

The main model for assessing the technical condition was determined by the model of the temperature of the stator winding of the turbogenerator from full power and the stator cooling system. To ensure the required accuracy of training the model from the telemetry data set, sections with non-stationary modes of equipment operation (starts, shutdowns, sections of power change and transient processes) were removed and a training sample of data on normal operation of a turbogenerator for 1 month was determined.

To develop the model, algorithms and methods of machine learning (linear regression) were used.

Further analysis of the data obtained using the developed model of predictive analysis of the state of the turbogenerator, in comparison with the actual values of the parameters of the sensors revealed:

- Systematic excesses of the actual temperature of the stator winding over the calculated temperature in grooves No. 22 and No. 24 (these grooves are equipped with temperature sensors) by an average of 8-10 ° C for 1.5 years before a failure occurs.

- Temperature peaks of excesses for a year before failure by 17 ° С, for 7-9 months - by 12 ° С, for half a year - by 22 ° С.

The results obtained made it possible to generate algorithms for identifying similar violations during operation of the TZFP-220-UZ turbogenerator 1.5 years before the failure.

Because only the grooves No. 2, 4, 22, 24, 42, 44 are equipped with temperature sensors in the presented turbogenerator model, then for monitoring and localizing overheating zones in the intermediate stator grooves not equipped with temperature sensors, a heat map of active steel and stator windings was developed using CAD and CAE finite element models. To clarify the areas of overheating, optimization algorithms were used.

The interaction of the developed models and algorithms for the predictive analysis of the state of the TZFP-220-UZ turbogenerator together with the ERP system made it possible to timely receive information on the need for appropriate repair actions and an assessment of the maximum allowable period for such work.

A key feature of this application is the presence of feedback. Upon receipt of information from ERP on the repair and restoration (replacement) of insulation on the stator winding core (s) with a localized overheating area. The model allows you to control the change in parameter values, thereby checking the fact of the work and evaluate their quality. After that, the model is retrained using new parameter values to take into account the changed properties of the repaired (replaced) rods and to eliminate the erroneous acceptance of differences in data for the anomaly.

Embodiments are not limited to the embodiments described herein, and other embodiments of the invention will be apparent to a person skilled in the art based on the information set forth in the description and prior art without departing from the spirit and scope of the present invention.

The elements mentioned in the singular do not exclude the plurality of elements, unless specifically indicated otherwise.

The functional connection of elements should be understood as a connection that ensures the correct interaction of these elements with each other and the implementation of one or another functionality of the elements. Particular examples of functional communication can be communication with the possibility of exchanging information, communication with the possibility of transmitting electric current, communication with the possibility of transmitting mechanical motion, communication with the possibility of transmitting light, sound, electromagnetic or mechanical vibrations, etc. The specific type of functional connection is determined by the nature of the interaction of the mentioned elements, and, unless otherwise indicated, is provided by well-known means using principles well known in the art.

The methods disclosed herein comprise one or more steps or actions to achieve the described method. The steps and / or actions of the method can replace each other without going beyond the scope of the claims. In other words, unless a specific order of steps or actions is defined, the order and / or use of specific steps and / or actions can be changed without departing from the scope of the claims.

The application does not indicate specific software and hardware for the implementation of the blocks in the drawings, but a specialist in the field of technology should understand that the essence of the invention is not limited to a specific software or hardware implementation, and therefore, any software and hardware known in the art can be used to implement the invention. prior art. So the hardware can be implemented in one or more specialized integrated circuits, digital signal processors, digital signal processing devices, programmable logic devices, user programmable gate arrays, processors, controllers, microcontrollers, microprocessors, electronic devices, and other electronic modules configured to carry out the functions described in this document, a computer or a combination of the above.

Although not specifically mentioned, it is obvious that when it comes to storing data, programs, and the like, a computer-readable storage medium is meant, examples of computer-readable storage media include read-only memory, random access memory, register, cache semiconductor storage devices, magnetic media such as internal hard drives and removable drives, magneto-optical media and optical media such as CD-ROMs and digital versatile disks (DVDs), as well as any other Gia known in the prior art storage media.

Although exemplary embodiments have been described in detail and shown in the accompanying drawings, it should be understood that such embodiments are merely illustrative and not intended to limit the broader invention, and that the invention should not be limited to the specific arrangements and structures shown and described, since various other modifications may be apparent to those skilled in the art.

The features mentioned in the various dependent claims, as well as the implementations disclosed in various parts of the description, can be combined to achieve beneficial effects, even if the possibility of such a combination is not explicitly disclosed.

Claims (33)

1. A system for diagnosing an industrial facility, comprising: - a data collection unit configured to collect data from a set of sensors of an industrial facility; - block model of an industrial facility, configured to simulate an industrial facility; - an analysis unit configured to analyze the state of the industrial facility based on data received from the data collection unit and the model of the industrial facility; moreover, the analysis unit is configured to make a conclusion about the normal or abnormal functioning of an industrial facility based on the analysis; moreover, the analysis unit is configured to receive data on changes made to the industrial facility, and command the model unit to change the model in accordance with the changes; moreover, the analysis unit is configured to command the model unit to change the model of an industrial facility based on repair data of an element of an industrial facility; characterized in that the industrial model is a system of equations describing the dependence of the parameter value of the industrial object on the input data obtained from the industrial object at least by sensors, and the system of equations contains four groups of correction factors: - coefficients of the physical model; - general correction factors determining the effect of degradation and external factors on the operation of all elements of an industrial facility; - individual correction factors for a given element of an industrial facility, which determine the influence of external factors on the operation of a given element of an industrial facility; - temporary correction factors that determine the effect of repair on the operation of this element of an industrial facility, moreover, the analysis unit is configured to reset all temporary correction factors for a given element of an industrial object upon receipt of a repair message for this element of an industrial object, while the resettable time coefficients are stored in memory, while the analysis unit is configured to obtain new values of temporary correction factors from the repaired element, comparing them with the stored coefficients and deciding on the lack of repair of this element ennogo object. 2. The system according to claim 1, in which the data on changes made to the industrial facility contain data on interference in the operation of the industrial facility when eliminating failures and precautionary conditions. 3. The system of claim 1, wherein the analysis unit is configured to determine an element requiring repair, leading to an abnormal operation of an industrial facility. 4. A method for diagnosing an industrial facility, comprising stages in which using a hardware-software complex: collect sensor data of an industrial facility characterizing the physical parameters of the industrial facility; create a model of an industrial facility; analyze the state of the industrial facility on the basis of the collected data and the model of the industrial facility; moreover, when analyzing, they make a conclusion about the normal or abnormal functioning of an industrial facility; moreover, when analyzing, they accept data on changes made to the industrial object, and change the model of the industrial object in accordance with the changes made on the basis of data on the repair of an element of the industrial object; characterized in that the industrial model is a system of equations describing the dependence of the values of the determined parameter of the industrial object on the input data obtained from the industrial object at least by sensors, and the system of equations contains four groups of correction factors: - coefficients of the physical model; - general correction factors determining the effect of degradation and external factors on the operation of all elements of an industrial facility; - individual correction factors for a given element of an industrial facility, which determine the influence of external factors on the operation of a given element of an industrial facility; - temporary correction factors that determine the effect of repairs on the operation of a given element of an industrial facility; reset all temporary correction factors for a given element of an industrial object upon receipt of a message about the repair of a given element of an industrial object, while the resettable time coefficients are stored in memory; receive new values of temporary correction factors from the repaired element; compare them with stored coefficients; and decide on the lack of repair of this element of an industrial facility. 5. The method according to p. 4, in which the data on the changes made to the industrial facility, contain data on interference in the operation of the industrial facility in the elimination of failures and precautionary conditions. 6. The method according to p. 4, which determines the element requiring repair, leading to the abnormal functioning of the industrial facility.

Images