JP2021144054A - Abnormality sign detection method - Google Patents

Abstract

To provide an abnormality sign detection method which enables the cause of abnormality to be recognized.SOLUTION: The data recording unit 4 of an abnormality sign detection system 1 has recorded therein vibration waveform data based on the vibration frequency of a rotary machine 2 acquired by a vibration sensor 3, and a frequency calculation unit 6 calculates a multidimensional feature quantity from the vibration waveform data. A normal data recording unit 6 has accumulated therein multidimensional feature quantity data that is obviously normal, and a normal model calculation unit 7 creates a normal model on the basis of the multidimensional feature quantity data of the frequency calculation unit 6. A diagnosis data recording unit 8 has recorded therein multidimensional feature quantity data based on the diagnosis waveform data that constitutes diagnosis data, and a reconstruction error calculation unit 9 obtains the error distribution of input/output generated when the diagnosis data is input to the normal model. An abnormality determination unit 10 determines an abnormality sign of the rotary machine 2 when the error distribution of the diagnosis data deviates from the error distribution of normal data. A variable error output unit 11 calculates an input/output error of the diagnosis data for each frequency and estimates the cause of abnormality in accordance with the evaluation of the error.SELECTED DRAWING: Figure 1

Description

The present invention relates to a technique for detecting a sign of abnormality based on sensing data of a target facility.

Currently, while the population is declining and there is a shortage of engineers, a large amount of electrical equipment manufactured during the period of altimeter economic growth has reached the end of its design life, and there is a demand for the construction of an equipment diagnosis system that utilizes "ICT / IoT".

For example, Patent Document 1 describes an abnormality sign detection system that can handle a case where a plurality of normal patterns serving as a determination criterion for an abnormality sign exist. Further, Patent Document 2 describes an abnormality sign detection system capable of creating an appropriate class set even when the number of normal classes and the normal range are unknown.

JP-A-2017-102765 JP-A-2018-28845

In order to detect an abnormality sign of the mechanical system of the rotating machine, the systems of Patent Documents 1 and 2 learn only the data at the normal time by "One-class Support Vector Machine" (OCSVM) from the frequency component of the vibration, and the abnormality sign is detected. The method of detecting is used.

However, since "OCSVM" is classified on the feature space by the kernel method, it cannot be directly associated with the input dimension, it cannot be determined which frequency component is abnormal, and the cause of the abnormality may not be grasped.

The present invention has been made to solve such a conventional problem, and an object of the present invention is to identify the cause of abnormality when detecting a sign of equipment abnormality.

(1) One aspect of the present invention is an abnormality sign detection system that detects an abnormality sign based on vibration waveform data of the target equipment.
A multidimensional feature calculation unit that generates multidimensional features based on the vibration waveform data collected in advance when the target equipment is normal, and a multidimensional feature calculation unit.
A normal model creation unit that generates a normal model that reproduces the input as an output by learning the multidimensional features as normal data.
A reconstruction error calculation unit that uses a multidimensional feature amount generated based on the vibration waveform data of the target equipment as diagnostic data and obtains an input / output error distribution when the diagnostic data is input to the normal model.
If the error distribution of the diagnostic data deviates from the error distribution of the normal data, the abnormality determination unit for determining the abnormality sign and the abnormality determination unit
It is provided with an abnormality factor estimation unit that calculates an error of the diagnostic data for each frequency and estimates an abnormality factor of the target equipment according to the evaluation of the calculated error.

(2) Another aspect of the present invention is an abnormality sign detection method for detecting an abnormality sign based on vibration waveform data of the target equipment by a computer.
A multidimensional feature calculation step that generates a multidimensional feature based on the vibration waveform data collected in advance when the target equipment is normal, and
A normal model creation step that generates a normal model that reproduces the input as an output by learning the multidimensional features as normal data, and
A reconstruction error calculation step for obtaining an input / output error distribution when the diagnostic data is input to the normal model, using a multidimensional feature amount generated based on the vibration waveform data of the target equipment as diagnostic data.
If the error distribution of the diagnostic data deviates from the error distribution of the normal data, the abnormality determination step for determining the abnormality sign and the abnormality determination step
It has an abnormality factor estimation step of calculating an error of the diagnostic data for each frequency and estimating an abnormality factor of the target equipment according to the evaluation of the calculated error.

According to the present invention, it is possible to identify the cause of abnormality when detecting a sign of equipment abnormality.

The block diagram of the abnormality sign detection system which concerns on embodiment of this invention. Schematic diagram of the autoencoder. The relationship diagram between the same vibration frequency and abnormal factors. The graph which shows the transition of the reconstruction error E. A map showing the transition of the error e for each variable (for each frequency). The graph which shows the power generation amount of the water turbine generator A which concerns on Example 1. FIG. The graph which shows the transition of the reconstruction error E. The enlarged view of FIG. 7 excluding the value at the time of the failure. A map showing the transition of the error e for each variable (for each frequency). The graph which shows the power generation amount of the water turbine generator B which concerns on Example 2. The graph which shows the transition of the reconstruction error E. A map showing the transition of the error e for each variable (for each frequency).

Hereinafter, the abnormality sign detection system according to the embodiment of the present invention will be described. This abnormality sign detection system continuously collects vibration waveform data from sensors (accelerometers, acoustic sensors, etc.) installed in the target equipment such as electrical equipment, and based on the collected vibration waveform data, the abnormality sign of the target equipment is detected. To detect.

By catching the signs of this equipment abnormality, it is possible to take measures before the failure of the target equipment, and it is possible to reduce the downtime of the infrastructure system and the like. At this time, for a diagnostic system in which it is generally difficult to obtain abnormal data in advance, it is preferable to use a method based on normal data, and it is desirable to output data that leads to identification of the cause of the abnormality in order to deal with equipment abnormalities in advance. ..

Therefore, in the abnormality sign system, attention is paid to the reproducibility of the input dimension of the autoencoder (self-encoder: Auto Encoder), which is a kind of neural network, and the abnormality factor is estimated together with the abnormality sign detection of the target equipment.

≪System configuration example≫
A configuration example of the abnormality sign detection system will be described with reference to FIG. The abnormality sign detection system 1 of FIG. 1 targets the rotating equipment (rotating machine) 2. Here, the abnormality sign of the rotating machine 2 is detected based on the sensing data detected by the vibration sensor 3 installed in the rotating machine 2, that is, the vibration frequency.

Specifically, the abnormality sign detection system 1 is composed of a computer and includes hardware resources of a normal computer (for example, a main storage device such as a CPU, RAM or ROM, and an auxiliary storage device such as an HDD or SSD).

As a result of the collaboration between the hardware resource and the software resource (OS, application, etc.), the abnormality sign detection system 1 has a data recording unit 4, a frequency calculation unit 5, a normal data recording unit 6, a normal model calculation unit 7, The diagnostic data recording unit 8, the reconstruction error calculation unit 9, the abnormality determination unit 10, and the variable error output unit 11 are implemented. Each of the recording units 4, 6 and 8 is constructed as a database in the storage device.

Here, the data recording unit 4 accumulates and records vibration waveform data obtained by A / D converting the vibration of the rotating machine 2 acquired by the vibration sensor 3. Further, the frequency calculation unit 5 calculates a multidimensional feature amount as necessary from the vibration waveform data recorded in the data recording unit 4.

That is, the frequency calculation unit 5 converts the vibration waveform data into a frequency spectrum, and generates multidimensional feature data with each frequency of the converted frequency spectrum as a variable. For example, it can be transformed by using Fast Fourier Transform (FFT), cepstrum, constant Q transform, or the like.

Based on the converted multidimensional feature data, the learning stage is executed in each of the parts 6 and 7, while the diagnostic stage is executed in each of the parts 8 to 11. In this learning stage, a normal model is generated based on the multidimensional feature data obtained by converting the vibration waveform data during normal operation. Further, in the diagnosis stage, the presence or absence of an abnormality sign of the rotating machine 2 is diagnosed based on the multidimensional feature data obtained by converting the vibration waveform data to be diagnosed.

(1) Learning stage The normal data recording unit 6 stores multidimensional feature data in which it is obvious that the majority of the data collected in advance are normal. That is, the normal data recording unit 6 records the multidimensional feature amount data (normal data) obtained by converting the vibration waveform data acquired during the normal operation of the rotating machine 2 among the recorded data of the data recording unit 4.

Further, the normal model calculation unit 7 creates a normal model with an autoencoder based on the multidimensional feature data of the normal data recording unit 6, and normalizes the multidimensional feature data input as needed (standardization, 0). ^~ 1 Range normalization, etc.). Here, the autoencoder is an algorithm for dimensional compression using a neural network in machine learning, and learns parameters to be output so as to reproduce an input.

This autoencoder is supervised learning using the same data for the input layer and the output layer in a three-layer neural network, and the activation functions of the intermediate layer and the output layer can be arbitrarily selected. Further, as shown in FIG. 2, the input layer and the output layer have the same number of units, and the intermediate layer has a decimal number (z _d <x _D ) smaller than the number of units of the input layer, so that the input data is compressed and the dimensionality is increased. Reductions will be made.

Then, assuming that the dimension of the input data "x" is "D", the unit activation degree "z" of the intermediate layer is given by the equation (1).

Here, the dimension "d" of "z" is "D>d", and "W" and "b" indicate the coupling load and the bias parameter, respectively. Further, F () indicates an activation function, and a "Relu function", a "sigmoid function", or the like can be used. Further, the output layer "x'" is decoded to the same dimension as the input layer and is expressed by the equation (2).

“W”, “b”, “W'”, and “b'” are referred to by an error backpropagation method or the like so that the error L between the input layer and the output layer expressed by the equation (3) is minimized. Determined using an optimization algorithm.

Here, "|| · ||" indicates the L1 norm, and "m" indicates the number of samples (the number of normal data).

(2) Diagnosis Stage First, the diagnostic data recording unit 8 records multidimensional feature data (diagnosis data) obtained by converting the diagnostic waveform data to be diagnosed among the recorded data of the data recording unit 4. For example, the record of the diagnostic data recording unit 8 can record diagnostic data for each working day of the rotary machine 2, for example.

Next, the reconstruction error calculation unit 9 and the abnormality determination unit 10 execute the detection of the abnormality sign by the autoencoder for the diagnosis data of the diagnosis data recording unit 8. In the abnormality sign detection by this autoencoder, the internal parameters are learned from the normal data, and the diagnostic data cannot be reproduced correctly by the learning, so that the abnormality sign is determined.

At this time, the reconstruction error calculation unit 9 inputs the diagnostic data of the diagnostic data recording unit 8 to the normal model created by the normal model calculation unit 7, and obtains the reconstruction error E (Reconstruction error) of the equation (4). .. ^{“|| ・ || 2} ” in this equation (4) indicates the L2 norm.

On the other hand, the abnormality determination unit 10 sets an appropriate first threshold value based on the error distribution (L1 norm) of the normal data, and determines the abnormality of the diagnostic data based on the first threshold value. For example, assuming that the error distribution (L1 norm) of the normal data ^{follows the “mean μ variance σ 2} ”, the first threshold value can be set to “3σ”.

At this time, if the reconstruction error (L2 norm) of the equation (4) exceeds the first threshold value, it is assumed that the error distribution (L1 norm) of the normal data is deviated, and the abnormality judgment of the diagnostic data, that is, the abnormality sign “presence” To judge. By this abnormality determination, an abnormality sign of the rotating machine 2 is detected.

Further, the variable error output unit 11 sets the difference vector e between "x" and "x'" as shown in the equation (5) in order to indicate which frequency of the diagnostic data contributes abnormally. Calculate and evaluate the error for each variable (frequency).

When this difference vector "e" exceeds the second threshold value set for the error of the normal data, the frequency is evaluated as abnormal, and the abnormal factor is estimated from the region of the frequency evaluated as abnormal. The method of estimating the abnormal factor will be described below.

That is, it is known that a failure of the mechanical system of the rotating machine 2 appears as an inherent vibration. FIG. 3 shows a relationship diagram of anomalous factors between the vibration frequency and the rotating machine 2. Here, since the modulation in the low frequency region includes the rotation frequency, the possibility of imbalance or misunbalance of the rotating body is suspected. On the other hand, in the high frequency region, it is considered that the waveform of the impact system is included, and bearing scratches and local abnormalities of the rotating body are suspected.

Then, when the abnormality of the rotating machine 2 is detected, if it is known which variable (frequency) indicates the abnormality, the cause of the abnormality of the rotating machine 2 can be estimated. Regarding this point, according to the abnormality sign detection system 1, when the difference vector “e” exceeds the second threshold value, the frequency is evaluated as an abnormality. By collating the region of the frequency evaluated as the abnormality with FIG. 3, the cause of the abnormality of the rotating machine 2 can be estimated.

For example, in FIGS. 4 and 5, an acceleration sensor is attached to a pump motor that is actually in operation, vibration data is collected every hour, and the collected time-series data is used to diagnose the abnormality sign detection system 1. The result is shown. Here, constant Q conversion is used for frequency conversion, and the learning stage is executed during the learning period S in FIG.

During this simulation, an abnormal noise occurred in the pump motor about 100 days after the start of measurement. Here, the horizontal axis of FIG. 4 indicates the time, the vertical axis indicates the reconstruction error E, and it is shown that the reconstruction error E tends to increase after about 100 days.

Further, FIG. 5 shows an error (difference vector) e for each frequency, the horizontal axis represents time, the vertical axis represents frequency, and the color represents the magnitude of error e. Here, it can be seen that the error e becomes large at a high frequency of 2 kHz or more after about 100 days. According to FIG. 3, it is presumed that the self-excited system abnormality / wear system abnormality is the cause of the abnormality.

Then, the estimated abnormality factor is output together with the result of the abnormality determination and presented to the user. As a result, the user can identify and grasp the cause of the abnormality of the rotating machine 2, and can repair the rotating machine 2 before the failure. In this respect, it is possible to take measures against failures in advance, which can contribute to the reduction of downtime of infrastructure systems and the like.

<< Example >>
Hereinafter, examples of the abnormality sign detection system 1 will be described. Here, Examples 1 and 2 target the turbine generators A and B, respectively. Both the turbine generators A and B are actually in operation, and no abnormality is found at the start of measurement.

An acceleration sensor was attached to the turbine generators A and B as a vibration sensor 3, and vibration data was collected every hour, and an abnormality sign was diagnosed based on the collected time series data. At this time, only the data during the power generation operation was used, and the vibration data was measured every hour for 5 seconds.

Further, as frequency conversion, octave band analysis by constant Q conversion is performed, and a spectrum of 100 frames is calculated for each measurement data. Furthermore, the training data (normal data) is normalized so that it becomes [maximum 1, minimum 0] at each frequency. Table 1 shows the parameters used in each embodiment.

(1) Example 1
The water turbine generator A of the first embodiment is a three-phase induction generator on the horizontal axis and has a maximum output of "2500 kW", and an acceleration sensor is installed near the bearing on the water turbine side. Here, FIG. 6 shows the transition of the power generation amount of the water turbine generator A, the horizontal axis shows the number of days elapsed from the measurement start date of the acceleration sensor, and the vertical axis shows the power generation amount [kW]. The water turbine generator A actually failed (bearing damage) about 25 days after the measurement start date.

[Learning stage]
In the learning stage, samples of the normal state are created in advance when the water turbine generator A is in the normal state before the diagnosis, and they are learned.

[Diagnostic stage]
In the diagnosis stage, the next diagnosis steps (S01 to S09) are executed from the diagnosis start date.

S01, S02: An acceleration sensor is attached to the turbine generator A, and vibration data for 5 seconds is collected every hour (S01). The vibration data for 5 seconds collected here is divided into 100 (S02).

S03, S04: Octave band analysis by constant Q conversion is performed, and a spectrum of 100 frames is calculated (S03). The frequency spectrum calculated here is normalized so as to be [maximum 1, minimum 0] at each frequency, and used as diagnostic data (S04). As a result, 100 diagnostic data are created.

S05, S06: The frequency vector normalized in S04 is input to the autoencoder learned in the learning stage as diagnostic data (S05). As a result, 100 reconstruction errors E and errors e are output (S06). Here, the reconstruction error E is the final result by calculating the average of 100 pieces. Further, for the error e, the average of 100 pieces is calculated for each frequency and used as the final result.

S07, S08, S09: If the reconstruction error E exceeds the preset first threshold value, the water turbine generator A is determined to be abnormal (S07). At this time, if the error e exceeds the preset second threshold value, the frequency is evaluated as abnormal (S08). Here, the frequency region evaluated as abnormal is collated with FIG. 3 to estimate the abnormal factor (failure cause) of the water turbine generator A (S09).

The results of the diagnostic stage of Example 1 will be described with reference to FIGS. 7 to 9. FIG. 7 shows the transition of the reconstruction error E, FIG. 8 shows an enlarged view of FIG. 7 excluding the value at the time of failure, the vertical axis shows the reconstruction error E, and the horizontal axis shows the number of days from the measurement start date. Is shown. Further, FIG. 9 shows the transition of the error e for each frequency, the vertical axis shows the frequency, the horizontal axis shows the sample index, and the color represents the magnitude of the error e as in FIG.

Here, according to FIG. 7, the vibration data on the failure date (25th day from the start of diagnosis) shows a clear difference from that before that, and it can be seen that the autoencoder has a high abnormality detection capability. However, as shown in FIG. 8, there is no significant difference in the reconstruction error E at the stage before the failure, and the sign of the failure cannot be confirmed.

On the other hand, paying attention to the error e for each frequency shown in FIG. 9, the modulation of the error can be confirmed in the high frequency region around 6 kHz from the 12th day, which is presumed to be a wear system abnormality according to FIG. As a result, the abnormality sign detection system 1 can detect a sign before the occurrence of a failure that cannot be confirmed only by the construction error E as a modulation of the error e for each frequency.

(2) Example 2
The water turbine generator B of the second embodiment is a three-phase self-excited alternator on the vertical axis and has a maximum output of "12 MW", and has been operating normally for about 120 days from the measurement. Here, FIG. 10 shows the transition of the power generation amount of the water turbine generator B, the horizontal axis shows the number of days elapsed from the measurement start date of the acceleration sensor, and the vertical axis shows the power generation amount [MW].

The learning stage and the diagnosis stage are executed for such a water turbine generator B in the same manner as in the first embodiment. The results of this diagnostic stage will be described with reference to FIGS. 11 and 12. Here, in FIG. 11, the reconstruction error E is shown on the vertical axis and the number of days from the measurement start date is shown on the horizontal axis as in FIG. 7. Further, FIG. 12 shows the transition of the error e for each frequency as in FIG. 9, the vertical axis shows the frequency, the horizontal axis shows the sample index, and the color shows the magnitude of the error e.

First, according to FIG. 11, it can be seen that there is a case where the reconstruction error E becomes large around the 30th day in the data that is operating normally. Next, according to FIG. 12, an error e on the plus side can be confirmed in the vicinity of 100 Hz to 300 Hz. Since this is a low-frequency abnormality, it is presumed to be a structural system abnormality according to FIG. 3, and in this respect as well, the abnormal cause of the abnormality sign can be identified before the failure occurs.

The present invention is not limited to the above embodiment, and the system configuration and the like are not limited to FIG. 1, and can be modified and implemented within the scope described in each claim.

The present invention can also be configured as a program that causes the computer to function as the abnormality sign detection system 1. According to this program, the computer functions as each of the parts 4 to 10 and can detect an abnormality sign of the target equipment.

1 ... Abnormal sign detection system 2 ... Rotating equipment (target equipment)
3 ... Vibration sensor 4 ... Data recording unit 5 ... Frequency calculation unit (multidimensional feature amount calculation unit)
6 ... Normal data recording unit 7 ... Normal model calculation unit (normal model creation unit)
8 ... Diagnostic data recording unit 9 ... Reconstruction error calculation unit 10 ... Abnormality judgment unit 11 ... Variable error output unit (abnormal factor estimation unit)

Claims (4)

This is an abnormality sign detection method that detects an abnormality sign based on the vibration waveform data of the target equipment using a computer.
A multidimensional feature calculation step that generates a multidimensional feature based on the vibration waveform data collected in advance when the target equipment is normal, and
A normal model creation step that generates a normal model that reproduces the input as an output by learning the multidimensional features as normal data, and
A reconstruction error calculation step for obtaining an input / output error distribution when the diagnostic data is input to the normal model, using a multidimensional feature amount generated based on the vibration waveform data of the target equipment as diagnostic data.
If the error distribution of the diagnostic data deviates from the error distribution of the normal data, the abnormality determination step for determining the abnormality sign and the abnormality determination step
An abnormality factor estimation step of calculating an error for each frequency of the diagnostic data and estimating an abnormality factor of the target equipment according to the evaluation of the calculated error.
An abnormality sign detection method characterized by having. The abnormality sign detection method according to claim 1, wherein the normal model creation step is to create the normal model by a dimension compression type autoencoder using a neural network. The abnormality sign detection according to claim 1 or 2, wherein the abnormality determination step determines that the error distribution of the diagnostic data deviates from the error distribution of the normal data if the error distribution exceeds a preset threshold value. Method. In the abnormality factor estimation step, if the error of the diagnostic data exceeds a preset threshold value,
The method for detecting an abnormality sign according to any one of claims 1 to 3, wherein the error is highly evaluated.

Images