KR102226971B1 - Method for fault diagnosis based on multiple variables and apparatus using the method - Google Patents

Abstract

The present invention relates to a vibration-based failure diagnosis method considering complex conditions and an apparatus for performing such a method. The vibration-based failure diagnosis method considering complex conditions is a learning step that performs learning based on learning information, and a device diagnostics based on a first factor determined in the learning step and a second factor determined based on the diagnostic information. It may include a diagnostic step.

Description

A vibration-based fault diagnosis method taking into account complex conditions, and an apparatus for performing such a method {Method for fault diagnosis based on multiple variables and apparatus using the method}

The present invention relates to a vibration-based failure diagnosis method considering complex conditions and an apparatus for performing such a method. More specifically, it relates to a vibration-based fault diagnosis method considering complex conditions for determining a fault occurring in a device based on vibrations generated under various conditions such as speed, temperature, and output when the device is driven, and a device for performing such a method. .

With the development of the 4th industry, the use of various artificial intelligence technologies that determine and determine close to human intelligence is increasing. Various artificial intelligence technologies are also used to determine the failure of machines such as motors and pumps. This artificial intelligence technology can also be used to diagnose the health of facilities at the plant/power plant unit, not just a single machine unit. As power plants at home and abroad are aging in recent years, the cost of maintenance and management of power plants continues to increase. In order to reduce accidental failures and management costs of power plants, technology for predicting and diagnosing health conditions using big data obtained from power plants is required.

Until recently, various artificial intelligence algorithms have been proposed and used in relation to machine failure diagnosis, but it is difficult to say that the accuracy of failure diagnosis or failure prediction is still high. Therefore, there is a need for research on artificial intelligence technology that can increase the accuracy of prediction of machine failure.

It is an object of the present invention to solve all of the above-described problems.

In addition, an object of the present invention is to reduce the amount of learning information and the learning time for fault diagnosis by reducing the dimension of multidimensional input information.

In addition, an object of the present invention is to classify information into a plurality of classes, and to perform a fault diagnosis of a device more accurately through basic diagnosis and precise diagnosis of the device using at least one of the plurality of classes.

A typical configuration of the present invention for achieving the above object is as follows.

According to an aspect of the present invention, a vibration-based failure determination method considering a complex condition is based on a learning step of performing learning based on learning information, a first factor determined in the learning step, and a second factor determined based on diagnostic information. As a result, a diagnosis step of performing diagnosis on the device may be included.

Meanwhile, the diagnosis step may include a constant diagnosis step and a precise diagnosis step.

In addition, the first factor includes a standard vibration factor, the second factor includes a vibration factor, and the constant diagnosis step is performed based on the standard vibration factor and a first soundness factor determined based on the vibration factor. Can be.

In addition, the first factor further includes a standard precise diagnosis factor, the second factor further includes a precise diagnosis factor, and the precise diagnosis step includes a second determined based on the standard precise diagnosis factor and the precise diagnosis factor. It can be performed based on a health factor.

According to another aspect of the present invention, a vibration-based failure determination apparatus considering a complex condition includes a communication unit implemented for communication with an external device and a processor operatively implemented with the communication unit, wherein the processor includes learning information The device may be diagnosed based on a first factor determined in a learning step of performing learning based on and a second factor determined based on diagnostic information.

Meanwhile, the diagnosis may include constant diagnosis and precise diagnosis.

In addition, the first factor includes a standard vibration factor, the second factor includes a vibration factor, and the constant diagnosis step is performed based on the standard vibration factor and a first soundness factor determined based on the vibration factor. Can be.

In addition, the first factor further includes a standard precise diagnosis factor, the second factor further includes a precise diagnosis factor, and the precise diagnosis step includes a second determined based on the standard precise diagnosis factor and the precise diagnosis factor. It can be performed based on a health factor.

According to the present invention, the amount of learning information and a learning time for fault diagnosis can be reduced by reducing the dimension of the multidimensional input information.

In addition, according to the present invention, information is divided into a plurality of classes, and fault diagnosis of a device can be performed more accurately through basic diagnosis and precise diagnosis of the device using at least one of the plurality of classes.

1 is a conceptual diagram showing learning and diagnosis for fault diagnosis of a device according to an embodiment of the present invention.
2 is a flow chart showing a learning step for diagnosing a device failure according to an embodiment of the present invention.
3 is a flowchart showing a diagnostic step for diagnosing a device failure according to an embodiment of the present invention.
4 illustrates a method for determining a class according to an embodiment of the present invention.
5 is a conceptual diagram showing a standard vibration factor and a method of determining a vibration factor according to an embodiment of the present invention.
6 is a conceptual diagram showing a standard vibration factor and a method of determining a vibration factor according to an embodiment of the present invention.
7 is a conceptual diagram showing a learning step and a diagnosis step for each class according to an embodiment of the present invention.
8 is a conceptual diagram showing an angular resampling procedure according to an embodiment of the present invention.
9 is a conceptual diagram showing a method for performing a precise diagnosis according to an embodiment of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The detailed description of the present invention described below refers to the accompanying drawings, which illustrate specific embodiments in which the present invention may be practiced. These embodiments are described in detail sufficient to enable a person skilled in the art to practice the present invention. It should be understood that the various embodiments of the present invention are different from each other, but need not be mutually exclusive. For example, specific shapes, structures, and characteristics described in the present specification may be changed and implemented from one embodiment to another without departing from the spirit and scope of the present invention. In addition, it should be understood that the position or arrangement of individual elements in each embodiment may be changed without departing from the spirit and scope of the present invention. Therefore, the detailed description to be described later is not to be made in a limiting sense, and the scope of the present invention should be taken as encompassing the scope claimed by the claims of the claims and all scopes equivalent thereto. Like reference numerals in the drawings indicate the same or similar elements over several aspects.

Hereinafter, various preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings in order to enable those of ordinary skill in the art to easily implement the present invention.

A wind generator is a device that converts wind kinetic energy into electric energy using a turbine, does not emit carbon during power generation, and has relatively low power generation costs. Accordingly, the ratio of wind power generation to the total amount of global power generation is increasing, and the demand for wind power generators around the world is also increasing.

Currently, the share of wind power generation is increasing rapidly in Korea, but maintenance is important in order to efficiently operate a wind farm. In the case of a wind power generator failure, a plan for supplying parts and securing a crane should be made while considering the weather such as wind speed and precipitation. Therefore, downtime can be minimized by predicting failures in advance and scheduling maintenance accordingly.

On the other hand, the main mechanical elements of a wind power generator are composed of rotor-blades, main bearings, gearboxes, and rotating bodies such as a generator, and failure diagnosis of the mechanical elements is possible through vibration information of the mechanical elements.

The operating conditions such as the power generation amount and the rotor speed of the wind turbine are values that continuously change according to the wind speed, and the characteristics of the vibration change accordingly. Depending on the driving state, it may be classified into a stationary state and a non-stationary state. In addition, environmental factors such as temperature also change the characteristics of vibration. For accurate diagnosis, it is necessary to consider vibration information that changes by the above information. However, as the amount of information to be considered increases exponentially, the amount of information necessary for learning increases exponentially, so that the learning period may be longer, and if the amount of information necessary for learning is not satisfied, the accuracy decreases.

Therefore, in the vibration-based failure determination method considering the complex condition according to an embodiment of the present invention and the apparatus for performing this method, the vibration information is converted into a characteristic factor independent of the multidimensional input information through the normalization and correction procedure for the multidimensional input information. Therefore, it is possible to diagnose the device at all times. In addition, detailed diagnosis may be performed through additional analysis of some of the driving information.

Hereinafter, a method for determining a vibration-based failure in consideration of a complex condition according to an embodiment of the present invention and an apparatus for performing such a method are described using a wind power generator as an example for convenience of description. However, it may be used for determining failure of various devices as well as a wind power generator, and such an embodiment may also be included in the scope of the present invention.

1 is a conceptual diagram showing learning and diagnosis for fault diagnosis of a device according to an embodiment of the present invention.

In FIG. 1, a method for diagnosing a device failure through a learning step 100 and a diagnosis step 150 is disclosed.

Referring to FIG. 1, in the learning step 100, learning information 105 is collected, and a standard vibration factor and a standard precision diagnosis factor may be obtained based on the learning information 105.

The learning information 105 may include temperature information (learning), speed information (learning), power information (learning), and vibration information (learning). The learning information 105 may be divided into input learning information and output learning information. The input learning information may include temperature information (learning), speed information (learning), and power information (learning), and the output learning information may include vibration information (learning).

In the present invention, learning information 105 may be collected from each of at least one sensor installed in the device. In an embodiment of the present invention, the collected multi-dimensional input learning information (temperature information (learning), speed information (learning), power information (learning)) is considered to be applied to the device in consideration of output learning information (vibration information (learning)). Learning for fault diagnosis can be performed.

Multidimensional input learning information (temperature information (learning), speed information (learning), power information (learning)) is output learning information (vibration information) through dimensional reduction through normalization and correction procedures for each of the multidimensional input information (learning). (Learning)) may be determined as the standard vibration factor 110. The standard vibration factor 110 is a characteristic factor irrelevant to the multidimensional input learning information and may be used for constant diagnosis of the device. Through the process of reducing the dimension of the multidimensional input learning information used in the learning phase, the size of information required for learning and the learning time can be reduced.

In the learning step 100, normalization and correction functions for normalization and correction procedures for each of the multidimensional input information (learning) may be performed.

The standard vibration factor 110 for each of the at least one sensor installed in the device is determined through the learning step 100, and the standard vibration factor distribution 120 for at least one sensor is determined based on the standard vibration factor 110. I can. For example, if there is sensor 1, sensor 2, and sensor 3 in the device, the (standard vibration factor of sensor 1, standard vibration factor of sensor 2, standard vibration factor of sensor 3) determined by time is distributed on the three-dimensional axis. Can be achieved. This distribution may be the standard vibration factor distribution 120 of the device. The vibration factor 160 extracted later in the diagnosis step may be determined as the first soundness factor 170 determined based on the standard vibration factor distribution 120. The first health factor 170 may be used to determine a failure for a device (or an element included in the device).

In addition, in the learning step 100, a standard precision diagnosis factor 130 for precise diagnosis may be determined based on the collected learning information. Prior to the standard precision diagnostic factor determination step, angular resampling is performed as a preprocessing step. The second soundness factor 190, which is later extracted in the diagnosis step, may be determined based on the standard precise diagnosis factor 130 obtained in the learning step 100 and the precise diagnosis factor 180 obtained in the diagnosis step 150. In addition, failure determination of the device (or an element included in the device) may be made using the second health factor 190.

In the diagnosis step 150, diagnosis information is collected, and a first health factor 170 and a second health factor 190 may be determined based on the diagnosis information. The diagnostic information may include temperature information (diagnosis), speed information (diagnosis), power information (diagnosis), and vibration information (diagnosis). The diagnostic information may be divided into input diagnostic information and output diagnostic information. The input diagnosis information may include temperature information (diagnosis), speed information (diagnosis), and power information (diagnosis), and the output diagnosis information may include vibration information (diagnosis).

The diagnosis step may include a regular diagnosis step and a detailed diagnosis step.

The output diagnosis information (vibration information (diagnosis)) for multidimensional input diagnosis information (temperature information (diagnosis), speed information (diagnosis), power information (diagnosis)) collected in the regular diagnosis step is provided for each of the multidimensional input diagnosis information. It may be determined as the same vibration factor 160 as the method of calculating the standard vibration factor 110 in the learning step through dimension reduction through normalization and correction procedures. The vibration factor 160 may be determined as the first soundness factor 170 based on the standard vibration factor distribution 120 determined in the learning step. The determination of whether a device (or an element of the device) has a failure may be performed based on the first health factor 170 determined in the constant diagnosis step. A determination as to whether the current state is a normal state, a dangerous state, or a fault state may be performed based on the first health factor 170.

In the precise diagnosis step, at least one precise diagnosis factor 180 may be determined through an angular resampling procedure for the collected diagnosis information. The second health factor 190 may be determined based on the at least one precise diagnosis factor 180 determined in the detailed diagnosis step and the standard precise diagnosis factor 130 determined in the learning step. A failure determination of a device (or an element included in the device) may be made using the second health factor 190.

The above learning step and diagnosis step may be performed by a vibration-based failure determination device (or a vibration-based failure diagnosis device) in consideration of a complex condition. The vibration-based failure determination apparatus may include a communication unit implemented for communication with an external device and a processor operatively implemented with the communication unit. The processor may be implemented to perform at least one of a learning step and/or a diagnosis step to be described later. For example, the processor may perform diagnosis on the device based on the first factor determined in the learning step of performing learning based on the learning information and the second factor determined based on the diagnosis information.

Hereinafter, in an embodiment of the present invention, a specific learning step and a diagnosis step will be described.

2 is a flow chart showing a learning step for diagnosing a device failure according to an embodiment of the present invention.

In FIG. 2, a learning step for diagnosing a device failure is started.

Referring to FIG. 2, collection of learning information and classification based on the collected learning information may be performed (step S200).

The learning information may include temperature information (learning), speed information (learning), power information (learning), and vibration information (learning) as a factor for learning fault diagnosis of the device. Classification may be performed based on at least one piece of learning information among a plurality of pieces of learning information. For example, learning information may be classified based on speed information (learning) and power information (learning) excluding temperature information (learning) and vibration information (learning). Specifically, learning information may be classified into a plurality of classes based on speed information (learning) and power information (learning).

A standard vibration factor for constant diagnosis is determined (step S210).

A standard vibration factor may be determined based on output learning information (vibration information (learning)) for multidimensional input learning information (temperature information (learning), speed information (learning), and power information (learning)).

Vibration information (learning) is vibration information measured by a sensor and is at least one of the RMS (root mean square), peak-to-peak, crestfactor, and vibration kurtosis for the vibration value. It can contain one value.

A standard vibration factor may be determined based on input learning information and output learning information included in at least one of a plurality of classes for learning information.

Specifically, vibration information may be determined as a standard vibration factor through a normalization and correction procedure for each of the input learning information included in the multidimensional input learning information. For example, vibration information (learning) for three-dimensional input learning information consisting of (temperature information (learning), speed information (learning), and output information (learning)) is a normalization and correction procedure for temperature information (learning), It can be determined as a standard vibration factor through a normalization and correction procedure for output information (learning) and a normalization and correction procedure for speed information (learning). The standard vibration factor is a factor in which the alarm level does not change according to temperature, power, and speed, and may have characteristics that are not related to multidimensional input learning information.

The dimension of learning 4D data about speed information (learning), power information (learning), temperature information (learning), and vibration information (learning) is reduced through the normalization and correction procedure for the classification and input learning information. And, the amount of information and the learning time required for learning can be reduced. For a normalization and correction procedure for input learning information, learning about functions (prediction curve, standard deviation, etc.) for normalization and correction of input learning information may be performed.

A standard vibration factor distribution for each sensor may be determined (step S220).

In the above manner, a standard vibration factor for each sensor of each of a plurality of sensors installed in the device may be determined, and a standard vibration factor distribution for each sensor may be determined. The distribution of standard vibration factors for each sensor is learned and can be used for fault diagnosis in the future.

For example, a plurality of sensors may be installed in one device, and a statistical distance for an ordered pair of standard vibration factors may be determined through learning about a standard vibration factor distribution extracted from each of the plurality of sensors. When n sensors are installed in the device, an ordered pair of standard diagnostic factors of each of the n sensors determined by time may determine a standard vibration factor distribution. In a later diagnosis step, a first soundness factor to be described later may be determined based on a standard vibration factor distribution. Specifically, an alarm level for a set statistical distance may be set in consideration of a standard vibration factor distribution. The fault alarm level is set to a value higher than the critical alarm level, and can be displayed in the order of normal, dangerous, and fault according to the increase in risk. For example, when the standard vibration factor distribution is different, the determination of the current state of the device may be different even if the vibration factor is the same in the diagnosis step. In other words, the first soundness factor for the vibration factor determined in the diagnosis step is determined in consideration of the distribution of the standard vibration factor, and a determination of the current state of the device may be performed based on the first soundness factor.

Also, a standard precision diagnosis factor for precise diagnosis may be determined (step S250).

A standard precision diagnosis factor for precise diagnosis may be determined based on learning information corresponding to some of the plurality of classes classified based on the learning information.

In order to extract the standard precision diagnostic factor, constant velocity through angular resampling is performed (step S250-1).

The amount of change in speed information (learning) of some of the plurality of classes is lower than that of other classes and changes within a narrow area, so it can be regarded as a constant speed section. However, the speed information (learning) may be minutely changed depending on the surrounding conditions such as wind speed and wind direction. Such a minute change in speed information changes the position of the characteristic frequency in the frequency domain analysis, resulting in an error and lowering the accuracy of diagnosis. Therefore, after correcting at the same speed through angular resampling, it is possible to extract precise diagnostic feature factors. Angular resampling is a procedure that corrects the rotation at the same speed by changing the length of the time axis of the vibration signal in inverse proportion to the rotational speed measured by the tachometer. Through angular resampling, it is possible to improve the accuracy of frequency analysis (eg, characteristic frequency information, envelope signal frequency information) for each element (eg, bearing, gearbox) constituting the device.

The standard precise diagnosis factor is determined (step S250-2)

A standard precision diagnosis factor may be determined based on characteristic frequency information and envelope signal frequency information. Information such as status information and remaining life information of elements on the device may be determined through learning about a failure mode based on a standard precision diagnosis factor. Thereafter, the second health diagnosis factor may be determined based on the precise diagnosis factor determined in the diagnosis step and the standard precise diagnosis factor determined in the learning step. A fault diagnosis can be performed on the device (or element of the device) using the second health factor.

3 is a flowchart showing a diagnostic step for diagnosing a device failure according to an embodiment of the present invention.

In FIG. 3, a diagnostic step for diagnosing a device failure is started.

Referring to FIG. 3, diagnosis information may be collected and classification based on the collected diagnosis information may be performed (step S300).

The diagnostic information may include temperature information (diagnosis), speed information (diagnosis), power information (diagnosis), and vibration information (diagnosis) as a factor for diagnosing a failure of the device. Classification of diagnostic information may be performed in the same manner as classification of learning information in the learning step. Classification may be performed based on at least one diagnosis information among a plurality of diagnosis information. For example, classification of diagnostic information based on speed information (diagnosis) and power information (diagnosis) excluding temperature information (diagnosis) and vibration information (diagnosis) may be performed. Specifically, diagnosis information may be classified into a plurality of classes based on speed information (diagnosis) and power information (diagnosis).

Continuous diagnosis is performed based on the first health factor (step S320).

A vibration factor is determined based on output learning information (vibration information (diagnosis)) for multidimensional input diagnosis information (temperature information (diagnosis), speed information (diagnosis), and power information (diagnosis)) (step S320-1), The first soundness factor may be determined based on the vibration factor and the standard vibration factor distribution determined in the learning step (step S320-2).

A vibration factor may be determined based on input diagnosis information and output diagnosis information included in at least one of a plurality of classes for diagnosis information, and a first soundness factor may be determined based on a vibration factor and a standard vibration factor distribution.

Specifically, vibration information (diagnosis) may be determined as a vibration factor through a normalization and correction procedure for each of the input diagnostic information included in the multidimensional input diagnostic information. For example, vibration information (diagnosis) for 3D input diagnosis information consisting of (temperature information (diagnosis), speed information (diagnosis), and output information (diagnosis)) is a normalization and correction procedure for temperature information (diagnosis), It can be determined as a vibration factor through a normalization and correction procedure for output information (diagnosis) and a normalization and correction procedure for speed information (diagnosis). The vibration factor is a factor in which the alarm level does not change according to temperature, power, and speed, and may have characteristics that are not related to multidimensional input learning information. The process of determining the vibration factor in the diagnosis step may be the same as the process of determining the standard vibration factor in the learning step.

In the above manner, a vibration factor for each sensor of each of the plurality of sensors installed in the device is determined, and the vibration factor for each sensor may be determined as a first integrity factor based on a distribution of the standard vibration factor for each sensor determined in the learning step. Specifically, the first soundness factor may be determined based on a statistical distance set based on the distribution of the standard vibration factor. For example, an ordered pair of a plurality of sensors (eg, sensor 1 to sensor n) with respect to a first integrity factor (first integrity factor (sensor 1), first integrity factor (sensor 2), ??, first The statistical distance for the health factor (sensor n)) may be determined based on the Mahalanobis distance as shown in Equation 1 below.

<Equation 1>

By comparing the value of the first health factor with the set alarm level, the device may be constantly diagnosed (step S320-3).

The fault alarm level is set to a value higher than the critical alarm level, and can be displayed in the order of normal, dangerous, and fault according to the increase in risk.

Further, a precise diagnosis may be performed based on the second health factor (step S350).

A precise diagnosis factor may be extracted based on diagnosis information corresponding to some of the plurality of classes classified diagnosis information, and a second soundness factor may be determined based on the precise diagnosis factor and the standard precise diagnosis factor determined in the learning step. . The precise diagnosis may be performed based on the second health factor. The process of determining the precise diagnosis factor in the diagnosis stage and the process of determining the standard precision diagnosis factor in the learning stage may be the same.

In order to extract a precise diagnosis factor, uniform speed may be performed through angular resampling (step S350-1).

As described above, after correction at the same speed through angular resampling, a precise diagnosis factor may be determined (step S350-2).

The second health factor may be determined based on the precise diagnosis factor determined in the diagnosis step and the standard precise diagnosis factor determined in the learning step (step S350-3).

The precise diagnosis may be performed based on the second health factor, and a failure determination may be performed on the device (or element of the device) (step S350-4).

4 illustrates a method for determining a class according to an embodiment of the present invention.

In FIG. 4, a class classification method is disclosed for determining a standard vibration factor and a standard precision diagnosis factor in the learning step, and determining the vibration factor and the precise diagnosis factor in the diagnosis step. The class disclosed in FIG. 4 is an example of a class classified for a specific device. A class may be determined based on various other learning information and diagnosis information, and such class classification is also included in the scope of the present invention.

Referring to FIG. 4, classes may be classified based on speed information and power information.

Table 1 below is an example of a class classified based on speed information and power information for a 2MW class wind power generator.

<Table 1>

Learning information and/or diagnostic information may be classified into class 1 410 to class 5 450 based on the power measured by the installed power sensor and the rotation speed measured through the tachometer sensor. Class 5 450 may be in a stopped state, and Class 1 410 to Class 4 440 may be in a driving state. Among them, the class 1 410 and the class 2 420 may be in a state of constant speed rotation. The rotor is rotated below the cut-out speed by the wind power generator's own system, and below the cut-in speed (Class 3 (430), Class 4 (440) and Class 5 ( 450))), the generator does not work.

Based on the speed information and power information included in Class 1 (410), Class 2 (420) and Class 3 (430) among the classified classes, the standard vibration factor and the standard vibration factor distribution may be determined in the learning step, and the diagnosis step In, the vibration factor for basic diagnosis can be determined.

Based on the speed information and power information included in the class 1 410 and class 2 420, which are the highest speed sections among the classified classes, a precise diagnosis factor may be determined in the learning step, and the diagnosis factor may be determined in the diagnosis step. .

The characteristic frequencies of bearings, gearboxes and generators, which are diagnostic elements included in the device, are dependent on the rotational speed. Therefore, it is easy to perform a frequency analysis such as Fast Fourier Transform (FFT) in the class 1 410 and class 2 420, which are constant velocity sections, and a time synchronization average (TSA: Time) that removes noise and amplifies the defect signal. In the case of Synchronous Average), it is easy to apply in the constant velocity section.

Therefore, based on the speed information and power information included in the class 1 410 and class 2 420, a standard precision diagnosis factor may be determined in the learning process, and a precise diagnosis factor for precise diagnosis may be determined in the diagnosis process. .

Continuous monitoring is not possible because high wind speed is required to rotate the rotor at full speed during the diagnosis phase. Therefore, for class 1 (410), class 2 (420), and class 3 (430), where wind power generators generate power for continuous monitoring, the basic vibration characteristic factor (RMS, peak-to-peak), which is a basic feature of vibration analysis, is used. , crestfactor, kurtosis), the distribution of standard vibration factors may be determined in the learning stage through analysis of vibration information, and the vibration factors for basic diagnosis may be determined in the diagnosis stage.

That is, in the embodiment of the present invention, a standard vibration factor distribution (or standard vibration factor) and a standard precision diagnosis factor are determined in the learning step by using the class classified through classification of learning information and diagnostic information, and the vibration factor in the diagnosis step And a precise diagnosis factor can be determined.

5 is a conceptual diagram showing a standard vibration factor and a method of determining a vibration factor according to an embodiment of the present invention.

In FIG. 5, output learning/diagnosis information for multidimensional input learning/diagnosis information is determined as a standard vibration factor in the learning stage and a vibration factor in the diagnosis stage through normalization and correction procedures for each of the multidimensional input learning/diagnosis information. Is initiated.

Referring to FIG. 5, a standard vibration factor and a vibration factor may be determined through dimensionality reduction for multidimensional input learning/diagnosis information. Vibration information, which is output learning/diagnosis information for multidimensional input learning/diagnosis information, may be extracted for each sensor, and a standard vibration factor and a vibration factor may be determined for each sensor.

For example, 3D {(speed information, power information, and temperature information), (vibration information)} and output learning/diagnosis information (vibration information) are dimensionally reduced to input learning/diagnosis information to reduce the standard vibration factor. And a vibration factor can be extracted.

First, the input learning/diagnosis information 500 consisting of multi-dimensional (speed information, power information, and temperature information) is input learning/diagnosis information for each class as one-dimensional input learning/diagnosis information and two-dimensional input learning/diagnosis information. Reduction 510 may be performed.

Specifically, for input learning/diagnosis information of class 1 and class 2 of the 3D input learning/diagnosis information 500, based on the speed-power curve, two-dimensional input learning/diagnosis information (temperature information, power Information) 510 may be reduced in dimensionality. Among the three-dimensional input learning/diagnosis information 500, the input learning/diagnosis information of class 3 is reduced to one-dimensional input learning/diagnosis information (temperature information) 510 according to the speed information based on the speed-power curve. Can be done.

The speed-power curve is a curve containing information on power according to speed, and in the case of class 3, as a section without separate power monitoring, one-dimensional input learning/diagnosis information (temperature information) according to speed information 510 ), and in the case of Class 1 and 2, there is a section in which separate output monitoring exists, and data is reduced to 2D input learning/diagnosis information (temperature information, power information) 510 according to speed information. Can be done.

Thereafter, for the two-dimensional input learning/diagnosis information (temperature information, power information) 510 according to the speed information of class 1 and class 2, a normalization and correction procedure 520 for temperature information and power information may be performed. . For the one-dimensional input learning/diagnosis information (temperature information) 510 according to the speed information of class 3, a normalization and correction procedure 520 for temperature information may be performed. Dimensions for multidimensional input learning/diagnosis information may be sequentially decreased through a normalization and correction procedure for temperature information and/or power information.

Equation 2 below is an equation for normalizing and correcting temperature information, and Equation 3 may be an equation for normalizing and correcting power information.

<Equation 2>

<Equation 3>

One-dimensional data (vibration information for speed information) 530 may remain through the normalization and correction procedure for temperature information and power information, and after that, normalization for speed information is additionally performed to perform standard vibration factor (learning step). )/Vibration factor (diagnosis step) 540 may be determined.

Equation 4 below is an equation for determining a standard vibration factor.

<Equation 4>

The standard vibration factor/vibration factor 540 may be extracted for each sensor. The plurality of standard vibration factors 540 for the plurality of sensors may form a standard vibration factor distribution 550.

In the case of normalization and correction for temperature information and power information, in the case of class 3, the rotational speed of the generator and the amount of power produced are dependent on each other, so the normalization and correction procedures for temperature information can be performed except for the correction for power information. . On the other hand, in the case of Class 1 and Class 2, which are the maximum speed sections, power information and temperature information are changed while the speed of the device is fixed. According to an embodiment of the present invention, the temperature correction function for correcting the temperature information of class 1 and class 2 borrows the temperature correction function of the highest speed section in class 3, and then normalizes and corrects the power information. I can.

Through the above method, as a result, the learning amount and learning time in the learning phase can be reduced by changing the learning and diagnosis of 4D data on speed information, power information, temperature information, and vibration information into multiple stages of 2D data learning and diagnosis. have.

6 is a conceptual diagram showing a standard vibration factor and a method of determining a vibration factor according to an embodiment of the present invention.

In FIG. 6, a process of normalizing and correcting multidimensional input learning/diagnosis information for determining a standard vibration factor (learning step) and a vibration factor (diagnosing step) is disclosed. The learning step of learning a function for normalizing and correcting temperature information, power information, and speed information is started.

Referring to FIG. 6, first, a normalization and correction procedure for temperature information may be performed (step S600).

A normalization and correction procedure for temperature information may be performed based on a vibration information distribution, a prediction curve (temperature information), and a standard deviation according to the temperature information. Such a prediction curve can be generated using a polynomial function and a sigmoid function. Vibration information (temperature, measured value) and vibration information (temperature , The predicted value) can be calculated as a standard deviation. Thereafter, a normalization and correction procedure for a difference between the vibration information (temperature, measured value) and the vibration information (temperature, predicted value) according to the temperature information may be performed.

According to an embodiment of the present invention, in the learning step, learning about a prediction curve (temperature information) and a standard deviation may be performed. The prediction curve (temperature information) and the standard deviation may change according to the vibration information (temperature, measured value) accumulated in the learning stage. In the learning step, based on the prediction curve (temperature information) and the standard deviation determined based on the learning result, a normalization and correction procedure for temperature information may be performed in a later diagnosis step.

Next, a normalization and correction procedure for power information may be performed (step S620).

A normalization and correction procedure for output information may be performed based on a vibration information distribution and a prediction curve (power information) and a standard deviation according to the power information. The vibration information (temperature, power, measured value) and the prediction curve (power information) according to the power information may have values changed based on the normalization and correction procedure for the above-described temperature information. The prediction curve (power information) may include information on vibration information (temperature, power, prediction value).

Standard deviation of vibration information (temperature, power, measured value) and vibration information (temperature, power, predicted value) is calculated, and vibration information (temperature, power, measured value) and vibration information (temperature, power, predicted value) are considered. Thus, normalization and correction procedures for power information can be performed.

Similarly, according to an embodiment of the present invention, in the learning step, learning about a prediction curve (power information) and a standard deviation may be performed. The prediction curve (power information) and the standard deviation may change according to the vibration information (temperature, power, measured value) accumulated in the learning stage. A normalization and correction procedure for power information may be performed in a later diagnosis step based on a prediction curve (power information) and a standard deviation determined based on the learning result in the learning step.

Next, a normalization and/or correction procedure for the speed information may be performed (step S640).

Normalization and correction procedures for speed information may be performed based on vibration information (temperature, output, speed, measured value), prediction curve (speed information), and standard deviation according to the speed information. Vibration information (temperature, output, speed, measured value) and prediction curve (speed information) according to the speed information may be changed by the above-described normalization and correction procedure for temperature information and a normalization and correction procedure for power information. Standard deviation for vibration information (temperature, output, speed, predicted value) based on vibration information (temperature, output, speed, measured value) and predicted curve (speed information) according to speed information is calculated, and vibration information (temperature , Output, measured value) and vibration information (temperature, output, speed, predicted value), the normalization and correction procedure for the speed information can be performed.

Similarly, according to an embodiment of the present invention, in the learning step, learning about a prediction curve (speed information) and a standard deviation may be performed. The prediction curve (speed information) and the standard deviation may change according to the vibration information (temperature, power, speed, measured value) accumulated in the learning stage. In the learning step, based on the prediction curve (speed information) and the standard deviation determined based on the learning result, a normalization and correction procedure for power information may be performed in a later diagnosis step.

Through the above process, the standard vibration factor and/or the vibration factor may be finally determined. The standardized vibration factor and/or vibration factor may be a generalized factor in which the alarm level does not change according to temperature information, output information, and speed information determined through the above normalization and correction procedure.

The above process may be performed for each of the plurality of sensors, and each of a plurality of standardized vibration factors for each of the plurality of sensors may be extracted to obtain a standardized vibration factor distribution.

7 is a conceptual diagram showing a learning step and a diagnosis step for each class according to an embodiment of the present invention.

In FIG. 7, a method for performing a learning step and a diagnosis step for each class is disclosed.

Referring to FIG. 7, in the learning step, temperature information and vibration information for each speed section for class 3 710 are collected, and learning through normalization and correction for input learning information may be performed as described above. In addition, learning about functions for normalization and correction procedures may be performed.

In the case of Class 1 and Class 2 700 in the learning stage, vibration information changes according to speed information, temperature information, and power information, and thus a lot of learning data is required. Accordingly, the temperature correction function for the class 1 and class 2 700 is replaced by the temperature correction function of the speed section closest to the class 1 and class 2 700, and the learning time for the function may be reduced.

In the diagnosis step, temperature information and vibration information for each speed section corresponding to the class 3 760 may be collected, and diagnosis may be performed through normalization and correction of the input diagnosis information.

In the diagnosis step, temperature information, vibration information, and output information corresponding to the class 1 and class 2 750 are directly collected, and diagnosis may be performed through normalization and correction of the input diagnosis information as described above. That is, in the diagnosis step, different from the learning step, the diagnosis may be made based on the diagnosis information of the class 1 and class 2 750 that are directly collected.

8 is a conceptual diagram showing an angular resampling procedure according to an embodiment of the present invention.

In FIG. 8, an angular resampling procedure for obtaining a precision diagnosis learning factor and a second health factor is disclosed.

Referring to FIG. 8, there is a small change in speed even in Class 1 and Class 2, which are regarded as constant speed sections. The position of the frequency peak varies according to the change in speed, and errors may occur during frequency analysis.

Therefore, it is possible to measure the RPM every moment through the tachometer and adjust the length of the signal based on this. When the angular resampling 800 is performed, since data is resampled based on the same rotation angle, the same number of data per rotation may be obtained even when the RPM changes.

For example, when a change from 700 RPM to 1300 RPM occurs for 1 minute, the time unit is adjusted to 1000 RPM, and the characteristic frequency 850 may be extracted by recombining the signal.

Frequency analysis based on the angular resampling 800 may also be performed on an envelope signal, and a characteristic frequency 850 for the envelope signal may be determined.

9 is a conceptual diagram showing a method for performing a precise diagnosis according to an embodiment of the present invention.

In FIG. 9, a method of determining a second health factor based on a precise diagnosis factor (diagnosis step) for each device element is disclosed.

Referring to FIG. 9, in the case of a bearing, bearing vibration information included in Class 1 and Class 2 may be collected. A plurality of precision diagnosis factors 920 for the bearing may be determined based on each of the plurality of envelope signal frequency analysis 910 using the angular resampling 900 for bearing vibration information. A second integrity factor 930 for the bearing may be determined based on a plurality of precision diagnostic factors 920 for the bearing. The plurality of precision diagnosis factors 920 for the bearing may be factors using a characteristic frequency determined through frequency analysis of the envelope signal.

Equations 5 to 8 below are equations for determining the characteristic frequency of each element of the bearing.

Equation 5

Equation 6

Equation 7

Equation 8

In the case of a gearbox, the first integrity factor 950 for each gearbox position determined based on information included in Class 1, Class 2, and Class 3 may be used to determine the second integrity factor 990 for the gearbox. have.

In addition, in the case of a gear box, gear box vibration information corresponding to class 1 and class 2 may be collected. A precision diagnosis factor for the gearbox may be extracted based on the frequency analysis 970 for each sensor position and the time-frequency analysis 975 for each sensor position by using the angular resampling 960 for the gearbox vibration information.

Specifically, the precision diagnosis factor 980 for the gearbox n stage is extracted based on the frequency analysis for each position, and the precision diagnosis factor 985 for the gearbox n stage is extracted based on the time-frequency analysis for each sensor location. I can. Since the gearbox of the wind turbine is generally multi-stage, the characteristic frequency for each stage is determined as a precision diagnosis factor, and a plurality of precision diagnosis factors may determine the second integrity factor 990 for the gearbox. . In the case of the gearbox, the second soundness factor 990 for each position may be determined using the characteristic frequency and the sensor position.

Equations 9 and 10 below are equations for determining a characteristic frequency for a gearbox.

Equation 9

Equation 10

That is, in the method of determining a precision diagnosis factor according to an embodiment of the present invention, a characteristic frequency for each element of a device is set, and frequency analysis, time-frequency analysis, and envelope signal frequency analysis may be performed based on this. In the case of bearings, a precise diagnosis characteristic factor can be determined by performing frequency analysis on the envelope signal and calculating the size of the peak generated at the characteristic frequency of the defect. At this time, the characteristic frequency of the bearing is equal to n times (n=1,2,3,4,...) of the frequency indicated in Equations 5 to 8. Since the gearbox of a wind turbine is generally multi-stage including planetary gears, the characteristic frequency is calculated for each stage, and the peak size at the characteristic frequency and the width and size of the sideband around the characteristic frequency are determined through frequency analysis. By analyzing, precise diagnostic feature factors can be determined. The gear ratio of the planetary gear may be calculated through Equation 9, and the characteristic frequency GMF (gear mesh frequency) may be determined through Equation 10. Additionally, precise diagnostic feature factors can be determined through time-frequency analysis.

The embodiments according to the present invention described above may be implemented in the form of program instructions that can be executed through various computer components and recorded in a computer-readable recording medium. The computer-readable recording medium may include program instructions, data files, data structures, and the like alone or in combination. The program instructions recorded in the computer-readable recording medium may be specially designed and constructed for the present invention or may be known and usable to those skilled in the computer software field. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical recording media such as CD-ROMs and DVDs, magnetic-optical media such as floptical disks. medium), and a hardware device specially configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those produced by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware device can be changed to one or more software modules to perform the processing according to the present invention, and vice versa.

In the above, the present invention has been described by specific matters such as specific elements and limited embodiments and drawings, but this is provided only to aid in a more general understanding of the present invention, and the present invention is not limited to the above embodiments. Anyone with ordinary knowledge in the technical field to which the invention pertains can make various modifications and changes from these descriptions.

Therefore, the spirit of the present invention is limited to the above-described embodiments and should not be defined, and all ranges equivalent to or equivalently changed from the claims to be described later as well as the claims to be described later are the scope of the spirit of the present invention. It will be said to belong to.

Claims (8)

The vibration-based failure diagnosis method considering complex conditions,
A learning step of performing learning based on learning information; And
A diagnosis step of performing diagnosis on the device based on a first factor determined in the learning step and a second factor determined based on diagnostic information,
The diagnosis step includes a constant diagnosis step and a precise diagnosis step,
The first factor includes a standard vibration factor,
The second factor includes a vibration factor,
The constant diagnosis step is performed based on the standard vibration factor and a first soundness factor determined based on the vibration factor,
The first factor further includes a standard precision diagnostic factor,
The second factor further includes a precise diagnosis factor,
The precise diagnosis step is performed based on the standard precision diagnosis factor and a second soundness factor determined based on the precise diagnosis factor,
The standard vibration factor is determined based on output learning information for multidimensional input learning information,
The standard precision diagnosis factor is determined based on frequency analysis information determined based on angular resampling. The method of claim 1,
The standard vibration factor is determined for each of a plurality of sensors installed in the device,
And the first soundness factor is determined based on a standard vibration factor distribution of a plurality of standard vibration factors corresponding to each of the plurality of sensors. The method of claim 2,
Wherein the frequency analysis information includes characteristic frequency information or envelope signal frequency information. The method of claim 3,
The data used in the learning step and the diagnosis step are divided into a plurality of classes,
The standard vibration factor and the class for determining the standard precision diagnostic factor are individually determined based on the rotational speed of the rotor,
A method, characterized in that the class for determining the standard precision diagnosis factor is determined in consideration of whether the rotational speed of the rotor is constant. The vibration-based fault diagnosis device considering complex conditions,
A communication unit implemented for communication with an external device; And
Including a processor operatively implemented with the communication unit,
The processor is implemented to perform diagnosis on the device based on a first factor determined in a learning step of performing learning based on learning information and a second factor determined based on diagnostic information,
The diagnosis step includes a constant diagnosis step and a precise diagnosis step,
The first factor includes a standard vibration factor,
The second factor includes a vibration factor,
The constant diagnosis step is performed based on the standard vibration factor and a first soundness factor determined based on the vibration factor,
The first factor further includes a standard precision diagnostic factor,
The second factor further includes a precise diagnosis factor,
The precise diagnosis step is performed based on the standard precision diagnosis factor and a second soundness factor determined based on the precise diagnosis factor,
The standard vibration factor is determined based on output learning information for multidimensional input learning information,
The standard precision diagnosis factor is determined based on frequency analysis information determined based on angular resampling. The method of claim 5,
The standard vibration factor is determined for each of a plurality of sensors installed in the device,
The vibration-based failure determination device, characterized in that the first soundness factor is determined based on a standard vibration factor distribution of a plurality of standard vibration factors corresponding to each of the plurality of sensors. The method of claim 6,
The frequency analysis information comprises characteristic frequency information or envelope signal frequency information. The method of claim 7,
The data used in the learning step and the diagnosis step are divided into a plurality of classes,
The standard vibration factor and the class for determining the standard precision diagnostic factor are individually determined based on the rotational speed of the rotor,
The class for determining the standard precision diagnosis factor is determined in consideration of whether the rotational speed of the rotor is constant.

Images