JP7750101B2 - Abnormality diagnosis device and abnormality diagnosis method - Google Patents

Description

The present invention relates to an abnormality diagnosis device and an abnormality diagnosis method for a signal processing device, such as a control board, that performs a plurality of signal processes involving the transmission and reception of signals for the operation of a predetermined manufacturing facility.

Manufacturing equipment used in manufacturing processes is equipped with field devices such as sensors, control circuit boards, or control devices that drive loads such as motors in response to instructions from a programmed logic computer. Because control devices can fail due to external factors or aging, the devices themselves are equipped with functions to detect and display overcurrent, overvoltage, power loss, and sensor failure, which are useful for responding to abnormalities after they occur. The latest technology also includes functions that allow data from sensors, devices, and control circuit boards (control boards) to be collected using external data collection devices and used to check trend waveforms.

However, control boards are electronic circuit boards that operate specific manufacturing equipment using control devices and information processing devices, and when diagnosing abnormalities, the above functions alone are not enough to identify the location of the abnormality. For this reason, troubleshooting takes time because the location of the failure must be diagnosed using data stored within the control board itself that is not output externally, data collected from sensors, and information such as overcurrent, overvoltage, power loss, and sensor failure indications.

Furthermore, for control devices where the manufacturer has discontinued maintenance or withdrawn the equipment, there are no experts available, and even if the control board is examined at a later date, the root cause may not be identified. In such cases, this does not lead to improvements in response, such as the preparation of manuals after a failure. As a result, there is an issue of the time required to deal with problems through analysis (mean time to repair = MTTR) increasing.

To address this issue, for example, Patent Document 1 proposes a system that collects control board data and sensor data in advance, compares the data with expected values, and analyzes logs before and after an abnormality is detected.

In addition, as part of preventive maintenance, control boards are replaced regularly, and elements that are expected to deteriorate are replaced or repaired. The capacity of the electrolytic capacitors removed during this process is investigated to determine any deterioration trends, and inspections of control device boards of the same age are prioritized. However, with this preventive maintenance, even control boards that do not need repair can be repaired, leading to cost overruns. In addition, there is often no testing environment after repairs, making it impossible to fully guarantee the integrity of the control circuit.

To address this issue, for example, Patent Document 2 measures the ripple in the power supply waveform to check for deterioration of the electrolytic capacitors on the power supply board, and performs a deterioration diagnosis of the programmer logic computer using threshold judgment, leading to preventive maintenance.

JP 2008-250594 A Japanese Patent Application Laid-Open No. 2003-248515

However, in Patent Document 1, the diagnostic method is based on expected values, so depending on the settings, a failure may be overlooked. Furthermore, only peripheral information is available at the time of the board failure, and it is not possible to determine which part of the board is the direct cause of the failure, so the time required for analysis to diagnose the true cause of the failure remains unresolved. Meanwhile, the technology in Patent Document 2 is only applicable to power supplies, and no degradation diagnosis technology has been established for the control board itself.

Therefore, the present invention provides technology that enables simple and reliable abnormality diagnosis of signal processing devices, such as control boards, that perform multiple signal processes involving the transmission and reception of signals for the operation of specified manufacturing equipment.

To solve the above problems, the present invention provides the following [1] to [19].

[1] An abnormality diagnosis device for diagnosing abnormalities in a signal processing device that performs multiple signal processes involving signal transmission and reception for the operation of a predetermined manufacturing facility,
a signal acquisition unit that acquires a plurality of signals from the signal processing device;
an anomaly score calculation unit that calculates an anomaly score representing a deviation of the actual results of each signal acquired by the signal acquisition unit during operation of the specified manufacturing equipment from a prediction made by the learning model, using a learned model between signals generated by performing sparse structure analysis on a plurality of signals when the specified manufacturing equipment is operating normally;
an abnormality determination unit that determines whether a signal is abnormal based on the abnormality score calculated by the abnormality score calculation unit;
An abnormality diagnosis device having the above.

[2] The anomaly diagnosis device described in [1], wherein the trained model is constructed by constructing a model structure in which each of a plurality of signals acquired by the specified manufacturing equipment under normal conditions can be an output, and in which signals other than the signal selected as an output from the plurality of signals are input, and training the model by performing the sparse structure analysis.

[3] The abnormality diagnosis device described in [1] or [2], wherein the specified equipment is manufacturing equipment that transports transported objects, and the input signals used to create the trained model are stratified according to when the transported object is accelerating, in a steady state, or decelerating.

[4] The abnormality diagnosis device described in any one of [1] to [3], wherein the signal processing device is a control board provided in a control device that controls the specified equipment.

[5] Further comprising a map creation processing unit that creates a map that visualizes the anomaly scores of the plurality of signals;
the map is a color map in which the horizontal axis represents time and the vertical axis represents the type of signal, and the degree of the abnormality score of the plurality of signals is represented by color gradation,
The abnormality diagnosis device according to any one of [1] to [4], wherein the map display is capable of displaying the map in units of acquisition cycles of the plurality of signals and enlarging the map display up to a predetermined time.

[6] A method for diagnosing an abnormality in a signal processing device that processes a plurality of signals transmitted to and received from a signal processing device that performs a plurality of signal processes involving the transmission and reception of signals for the operation of a predetermined facility, comprising:
a signal acquisition step of acquiring a plurality of signals from the signal processing device;
an anomaly score calculation step of calculating an anomaly score representing a deviation of the actual results of each signal during operation of the specified manufacturing equipment acquired in the signal acquisition step from a prediction by the learning model, using a learned model between the signals generated by performing sparse structure analysis on the plurality of signals when the specified manufacturing equipment is operating normally;
an abnormality determination step of determining whether or not the signal is abnormal based on the abnormality score calculated in the abnormality score calculation step;
An abnormality diagnosis method comprising:

[7] The abnormality diagnosis method described in [6], wherein the trained model is constructed by constructing a model structure in which each of a plurality of signals acquired by the specified manufacturing equipment under normal conditions can be an output, and inputting signals other than the signal selected as an output from the plurality of signals, and training the model by performing the sparse structure analysis.

[8] Further comprising a map creation processing step of creating a map that visualizes the anomaly scores of the plurality of signals;
the map is a color map in which the horizontal axis represents time and the vertical axis represents the type of signal, and the degree of the abnormality score of the plurality of signals is represented by color gradation;
The abnormality diagnosis method according to [6] or [7], wherein the map display can be expanded from a display in units of acquisition cycles of the plurality of signals to a predetermined time.

[9] A method for generating an abnormality diagnosis model for a signal processing device that performs a plurality of signal processes involving signal transmission and reception for operation of a predetermined facility, comprising:
acquiring a plurality of signals from the signal processing device;
A method for generating an abnormality diagnosis model, which constructs a model structure in which a signal other than a signal selected as an output from among the plurality of signals obtained when the specified manufacturing equipment is operating normally can be output, and generates a trained model between the signals by performing sparse structure analysis and training.

This invention provides technology that can easily and reliably diagnose abnormalities in signal processing devices, such as control boards, that perform multiple signal processes involving the transmission and reception of signals for the operation of specified manufacturing equipment. This makes it possible to prevent the MTTR from expanding. Furthermore, this invention can be applied to control devices that do not inherently have self-diagnostic functions, and abnormality diagnosis is possible even without in-depth knowledge or skilled techniques.

1 is a functional block diagram of an abnormality diagnosis device according to an embodiment of the present invention; FIG. 1 is a diagram for explaining sparse structure analysis. FIG. 2 is a block diagram showing the main configuration of the static thyristor Leonard 2. FIG. 4 is a diagram showing an example in which the signal of the current variation control (ACCR) in FIG. 3 is used as the objective variable, all analyzable signals of the control device are structurally analyzed in a normalized form, and the top 16 explanatory variables with the highest influence on the objective variable are listed. FIG. 10 is a diagram showing a two-dimensional map that is an example of a map of abnormality scores created by a map creation processing unit. 2 is a flowchart showing a process flow when an abnormality diagnosis method is carried out by the abnormality diagnosis device of FIG. 1. 10A and 10B are diagrams showing the results of an experiment in which characteristic changes due to deterioration of a capacitor in a control board were simulated to confirm whether they could be detected by an abnormality diagnosis device.

Embodiments of the present invention will be described below with reference to the accompanying drawings.

<Abnormality diagnosis device>
FIG. 1 is a functional block diagram of an abnormality diagnosis device according to an embodiment of the present invention.
In this embodiment, an example will be described in which a control board of a static thyristor Leonard 2 is used as a signal processing device that performs a plurality of signal processes involving the transmission and reception of signals for the operation of a predetermined manufacturing facility.

The abnormality diagnosis device 1 takes in multiple signals from the control board (signal processing device) of the static thyristor Leonard 2, performs sparse structure analysis on the multiple signals under normal conditions, and creates a model between the signals by learning from them.The created model is then used to calculate an abnormality score for signals under operation, normalizing and quantifying the degree of deviation from normality of the signals under operation and determining whether the signals are abnormal.

The Static Thyristor Leonard 2 is used, for example, in manufacturing equipment that transports objects such as steel plates by rotating a table motor with an electric motor. It is a control device that converts AC to DC and controls the voltage supplied to the DC motor (electric motor), providing smooth speed control. The Static Thyristor Leonard 2 has 10 or more control boards that can process (analyze) multiple signals, for example, several tens of signals, and is controlled by a higher-level control PLC 3. The control boards of the Static Thyristor Leonard 2 have analog circuits that amplify and control external signals.

The host control PLC 3 issues control commands to the control board of the static thyristor Leonard 2, and at the same time provides the abnormality diagnosis device 1 with analysis triggers, such as whether or not the device is in operation, and if an object is being transported, whether it is accelerating, in steady-state transport, or decelerating.

Next, the abnormality diagnosis device 1 will be specifically described.
The abnormality diagnosis device 1 includes an analysis unit 10 and a storage unit 20. The analysis unit 10 is configured as an analysis PLC, and includes a signal sampling processing unit (signal acquisition unit) 11, an abnormality score calculation unit 12, an abnormality determination unit 13, an alarm output unit 14, and a map creation processing unit 15.

The signal sampling processing unit 11 receives an analysis trigger (collection start trigger) from the upper control PLC 3 and, based on that, samples (acquires) signals from the control board, which is the signal processing device for the static thyristor Leonard 2.

The anomaly score calculation unit 12 uses a trained model between signals generated by performing sparse structure analysis on multiple signals when the specified manufacturing equipment is operating normally to calculate an anomaly score that represents the deviation of the actual results of each signal sampled by the signal sampling processing unit 11 during operation from the prediction made by the trained model. Note that the normal state of the specified manufacturing equipment refers to a state in which the specified manufacturing equipment can operate normally using signals from the control board, which is a signal processing device. Furthermore, when applied to equipment that transports transport objects, the input signals used to create the normal state model are preferably stratified into (1) acceleration, (2) steady state (constant speed), and (3) deceleration.

Sparse structure analysis is a method for expressing causal relationships (correlations) between variables, characterized by ignoring weak correlations and retaining only the essential relationships between strongly correlated variables. Output variables can be predicted by performing sparse structure analysis based on lasso regression, as shown in Figure 2, for example. Figure 2(a) schematically illustrates a model under normal conditions, in which lasso regression is performed on data from a certain section of the target signal, resulting in minute regression coefficients of 0. For example, the X2 variable is modeled so that it has no causal relationships (correlations) other than those between X1 and X4. A distinctive feature of this method is that it creates a model of all variables so that the regression coefficients between variables that do not have a causal relationship are 0. Figure 2(b) schematically illustrates an example of an abnormal condition, in which the causal relationship (correlation) between X2 and X4 has been disrupted for some reason.

This will be explained in more detail.
The trained model between signals is generated for a certain control board by an external offline computer (such as a server computer or a cloud computer). The trained model between signals is generated by constructing a model structure in which each of multiple signals (e.g., X1 to X5 in FIG. 2 ) acquired by a specific manufacturing facility under normal conditions can be an output, and inputting signals other than the signal selected as an output from these multiple signals, and then performing sparse structure analysis based on Lasso regression to train the model and predict output variables. An anomaly score is then calculated, representing the degree to which the actual performance of each signal acquired during operation deviates from the prediction based on the trained model. The anomaly score is defined for each signal by the following formula:
Anomaly score = (Xi actual - Xi predicted) ² / σ ²

When an abnormality occurs, as shown in Figure 2(b), the value of this anomaly score increases, so this property is used to detect a breakdown in the causal relationship (correlation) between signals and thereby detect an anomaly.

This section provides a specific example of model generation for the control board of a static thyristor Leonard 2 used when the specified equipment is a conveying facility for transporting steel plates, etc. Figure 3 is a block diagram showing the main components of a static thyristor Leonard 2. In this example, the ACCR signal in Figure 3 is used as the objective variable (input). A structural analysis was performed on all analyzable signals (50 signals) of the control device in a normalized form, and Figure 4 shows an example of the 16 explanatory variables (outputs) with the highest influence on the objective variable. As shown in Figure 4, the explanatory variables are, from top to bottom, voltage feedback, ACR output, ACCR intermediate, forward rotation ON signal, etc. Normal signals are collected for each signal, and the top explanatory variables with the highest contribution are automatically selected to create a regression model.

In the case of conveying equipment that transports steel plates and the like, as mentioned above, it is preferable that the input signals used to create a model under normal conditions be stratified into (1) acceleration, (2) steady state (constant speed), and (3) deceleration. In particular, in conveying equipment using static thyristor Leonard 2, the control board is a board with analog circuits, so the input signals used to create a model under normal conditions behave differently during acceleration, steady state (constant speed), and deceleration. For this reason, there is a strong need to stratify these and create multiple models.

The anomaly determination unit 13 is capable of setting a threshold for the anomaly score, and normalizes and quantifies the deviation of the signal from normality based on the anomaly score calculated for the signal during operation, to determine whether the signal is abnormal. The anomaly determination unit 13 is capable of setting a threshold for the anomaly score. For example, the unit can be configured to determine that a signal is abnormal if the proportion of times the anomaly score is equal to or greater than the threshold during a data (signal) sampling period is equal to or greater than a predetermined value (e.g., 10%).

The alarm output unit 14 outputs an alarm contact output and issues an external alarm when the signal is determined to be abnormal based on the abnormality score.

The map creation processing unit 15 retrieves the data for multiple signals for which anomaly scores have been calculated from the storage unit 20 and creates a map of the anomaly scores for multiple signals over a certain period of time. The anomaly score map is sent to the web browser display PC 4 and displayed. The map is a visualization method that visualizes the anomaly scores for multiple signals over a certain period of time, allowing for at-a-glance understanding. For example, as shown in Figure 5, it is a two-dimensional map with time on the horizontal axis and signal type (11 types in Figure 5) on the vertical axis, and displays the map so that each anomaly score value can be distinguished. While the map in Figure 5 is displayed in grayscale, it is preferable to use a color map that displays color gradations according to the anomaly score value to make the anomaly score easier to understand. The map display can be expanded from a display in units of the signal sampling period to a specified time (e.g., up to one hour). This makes it possible to visualize overall anomalies, and also to use the zoom function to visualize instantaneous anomalies and determine which point in the signal in the control amplifier circuit first becomes abnormal.

<Abnormality diagnosis method>
Next, a flow of the abnormality diagnosis method in the abnormality diagnosis device configured as above will be described.

FIG. 6 is a flowchart showing a process flow when the abnormality diagnosis method is carried out in the abnormality diagnosis device of FIG.
First, an analysis trigger (acquisition start trigger), such as whether or not the vehicle is being transported and the transport mode (accelerating, steady, decelerating) is acquired from the upper control PLC 3 (step ST1), and a signal collection period is calculated (step ST2). The signal collection period is set for each of the acceleration, steady, and deceleration states.

Then, data (signals) from the control board, which is the signal processing device for the static thyristor Leonard 2, is stored (step ST3). This data is acquired and analyzed by the signal sampling processing unit 11 (step ST4). The signal sampling period at this time is preferably approximately 100 μs to 10 ms.

Next, the anomaly score calculation unit 12 uses a trained model between signals generated by performing sparse structure analysis on multiple signals when the specified manufacturing equipment is operating normally to calculate an anomaly score for each signal sampled by the signal sampling processing unit 11 during operation of the specified equipment (step ST5). The analysis results are stored in the memory unit 20 (step ST6).

The abnormality determination unit processing unit 13 calculates an abnormality score, quantifying the deviation from normality, for example by normalizing it, and determines whether the signal is abnormal (step ST7). Specifically, if the proportion of abnormality scores equal to or greater than a threshold during the data collection period exceeds a predetermined value (e.g., 10%), the signal is determined to be abnormal (score abnormality). If the score is determined to be abnormal, an alarm contact output is output from the alarm output unit 14 (step ST8), and an external alarm is sounded (step ST9).

Next, the map creation processing unit 15 updates the map of anomaly scores for multiple signals, launches the external web browser (step ST10), and performs map display processing on the external web browser display computer 4 to update the display (step ST11).

The Static Thyristor Leonard 2 has more than 10 control boards as signal processing devices, and each control board is designed to collect multiple (dozens of) signals, resulting in the exchange of a huge amount of signals. For this reason, in the past, when an abnormality occurred, determining the abnormal signal, its cause, and the action to be taken required experience and skilled techniques, and was extremely complicated and difficult.

In contrast, in this embodiment, an abnormality diagnosis device 1 is installed, and a sparse structure analysis of multiple normal signals taken from the control board of the static thyristor Leonard 2 is performed. A model between the generated signals is used to calculate an abnormality score for the signal during operation, and an abnormality in the signal is determined based on this. This allows for easy and reliable abnormality diagnosis of signal processing devices such as control boards without relying on experience or skilled techniques. This makes it possible to prevent the MTTR from expanding. Furthermore, it can be applied to control devices that do not originally have self-diagnosis functions, and abnormality diagnosis is possible even without in-depth knowledge of the device being diagnosed.

<Experimental Example>
Next, an experimental example will be described. In this example, the ACCR output processing of the ACR board of the static thyristor Leonard shown in Figure 3 was performed by changing the gain of the ACCR assuming deterioration of the control board, and simulating changes in characteristics due to deterioration of the capacitors in the control board, and confirming whether the abnormality could be detected by the abnormality diagnosis device. Figure 7 shows the results of such an experimental example.

Figure 7(a) plots the difference between predicted and actual ACCR output during operation at normal gain values (prediction - actual: difference between predicted and actual) and the ACCR gain abnormality score. The left side is a scatter plot of the difference between predicted and actual ACCR gain output and the ACCR gain abnormality score, and the right side is a histogram of the difference between predicted and actual ACCR gain output.

Figure 7(b) plots the difference between the predicted and actual values and the ACCR gain anomaly score when the ACCR gain is changed to simulate capacitor degradation. Similarly, the left side is a scatter plot of the difference between the predicted and actual values of the ACCR gain output and the ACCR gain anomaly score, and the right side is a histogram of the difference between the predicted and actual values of the ACCR gain output.

As shown in Figure 7(a), the difference between the predicted and actual ACCR gain output and the ACCR gain anomaly score during operation at normal gain values are small. However, when the ACCR gain is changed to simulate capacitor degradation, as shown in Figure 7(b), the anomaly score increases and the distribution of the difference between the predicted and actual ACCR output also doubles. This suggests that anomaly score monitoring can identify deterioration anomalies in the control board.

<Other applications>
Although the embodiments of the present invention have been described above, these are merely examples and should not be considered limiting. The above embodiments may be omitted, substituted, or modified in various ways without departing from the spirit of the present invention.

For example, in the above embodiment, a static thyristor Leonard control board was used as the signal processing device, but the control board may be a control device other than a static thyristor Leonard. Furthermore, the present invention is not limited to control boards, and can be applied to any signal processing device that performs multiple signal processes involving the transmission and reception of signals for the operation of specified manufacturing equipment.

Furthermore, while conveying equipment for transporting objects such as steel plates has been given as an example of specified equipment, it is not limited to this.

1. Abnormality diagnosis device 2. Static thyristor Leonard 3. Upper control PLC
4. Web browser display PC
REFERENCE SIGNS LIST 10 Analysis unit 11 Signal sampling processing unit 12 Abnormality score calculation unit 13 Abnormality determination processing unit 14 Warning output processing unit 15 Map creation processing unit 20 Storage unit

Claims (4)

An abnormality diagnosis device for diagnosing abnormalities in a signal processing device that performs a plurality of signal processes involving signal transmission and reception for operation of manufacturing equipment that transports an object to be transported by a DC motor ,
the signal processing device is a static thyristor Leonard control board that controls the supply of power to the DC motor to control the manufacturing equipment,
a signal acquisition unit that acquires a plurality of signals from the control board that is the signal processing device;
an anomaly score calculation unit that calculates an anomaly score representing a deviation of the actual results of each signal acquired by the signal acquisition unit during operation of the manufacturing equipment from a prediction made by the learned model, using a learned model between signals generated by performing sparse structure analysis on a plurality of signals when the manufacturing equipment is operating normally;
an abnormality determination unit that determines whether a signal is abnormal based on the abnormality score calculated by the abnormality score calculation unit;
and
the trained model is a regression equation trained by constructing a model structure in which a signal selected from a plurality of signals acquired by the manufacturing equipment under normal conditions is used as a response variable and signals of the plurality of signals excluding the selected signal are used as explanatory variables, and performing the sparse structure analysis ;
An abnormality diagnosis device in which the input signal for creating the trained model is stratified according to when the transport object is accelerating, steady, or decelerating . a map creation processing unit that creates a map that visualizes the anomaly scores of the plurality of signals;
the map is a color map in which the horizontal axis represents time and the vertical axis represents the type of signal, and the degree of the abnormality score of the plurality of signals is represented by color gradation,
2. The abnormality diagnosis device according to claim 1 , wherein the map display is capable of displaying the plurality of signals in units of acquisition cycles and enlarging the display up to a predetermined time. 1. A method for diagnosing an abnormality in a signal processing device that performs a plurality of signal processes involving signal transmission and reception for operation of a manufacturing facility that transports an object to be transported by a DC motor , comprising:
the signal processing device is a static thyristor Leonard control board that controls the supply of power to the DC motor to control the manufacturing equipment,
a signal acquisition step of acquiring a plurality of signals from the control board which is the signal processing device;
an anomaly score calculation step of calculating an anomaly score representing a deviation of the actual results of each signal during operation of the manufacturing equipment acquired in the signal acquisition step from a prediction by the learned model, using a learned model between the signals generated by performing sparse structure analysis on the plurality of signals when the manufacturing equipment is operating normally;
an abnormality determination step of determining whether or not the signal is abnormal based on the abnormality score calculated in the abnormality score calculation step;
and
the trained model is a regression equation trained by constructing a model structure in which a signal selected from a plurality of signals acquired by the manufacturing equipment under normal conditions is used as a response variable and signals of the plurality of signals excluding the selected signal are used as explanatory variables, and performing the sparse structure analysis ;
An abnormality diagnosis method , in which the input signal for creating the trained model is stratified according to when the transport object is accelerating, steady, or decelerating . a map creation processing step of creating a map that visualizes the anomaly scores of the plurality of signals;
the map is a color map in which the horizontal axis represents time and the vertical axis represents the type of signal, and the degree of the abnormality score of the plurality of signals is represented by color gradation;
4. The abnormality diagnosis method according to claim 3 , wherein the map display can be expanded from a display in units of acquisition cycles of the plurality of signals to a predetermined time.