Abnormality diagnosis system and abnormality diagnosis method

Abstract

The present invention provides an abnormality diagnosis system and an abnormality diagnosis method that can suppress the erroneous diagnosis of the occurrence of an equipment abnormality when the equipment abnormality does not actually occur. The abnormality diagnosis system (20) comprises: a sampling unit (21) for sampling time series data representing the state of the equipment, the time series data repeating a pattern corresponding to a constant pattern; and a calculation unit (24) for performing principal component analysis based on the sampled time series data to diagnose the abnormality of the equipment. The calculation unit (24) performs a first principal component analysis on a pattern in the time series data representing the state, and makes a first abnormality determination based on the result. If the first abnormality determination is that there is an abnormality, a second principal component analysis is performed on a plurality of temporally continuous patterns in the time series data representing the state, including the pattern that has been subjected to the first principal component analysis, and makes a second abnormality determination based on the result. If the second abnormality determination is that there is an abnormality, it is diagnosed as an equipment abnormality.

Description

Technical Field

[0001] The present invention relates to an abnormality diagnosis system and an abnormality diagnosis method for diagnosing abnormality of a device having a device driver. Background Art

[0002] Conventionally, in equipment having equipment drive parts such as hydraulic devices and electric motors, equipment abnormality diagnosis is performed based on signals such as current and load obtained from the equipment drive parts. It has been proposed to use principal component analysis (hereinafter also referred to as PCA) for abnormality diagnosis when the operation of the equipment drive parts is a repetitive operation of a predetermined constant pattern (e.g., Patent Documents 1 and 2).

[0003] For example, in a sizing press, dies for side pressure are arranged on both sides of the slab being transported, and power is transmitted from the motor to the die via the crank to press the slab sideways at a constant cycle. The current waveform or load waveform that appears repeatedly at this time is collected for each material, and the current waveform or load waveform of one crank cycle is analyzed for each material using PCA to extract abnormal waveforms. By applying PCA, equipment abnormality diagnosis with higher accuracy than simple upper and lower limit checks can be performed.

[0004] Patent Document 1: Japanese Patent No. 5991042

[0005] Patent Document 2: Japanese Patent Application Publication No. 2017-32567

[0006] However, in the device drive unit, there is a situation where the waveform of the current and load etc. becomes extremely small for some reason even if the device is not actually damaged. For example, in the operation of fixed width extrusion, sometimes a waveform with almost no load appears from time to time due to insufficient material transportation caused by the slip of the pinch roller. In the case of such a phenomenon, an abnormality determination is made even if the device is not actually damaged. Summary of the invention

[0007] Therefore, an object of the present invention is to provide an abnormality diagnosis system and an abnormality diagnosis method that can suppress an erroneous diagnosis that a device abnormality has occurred when actually no device abnormality has occurred.

[0008] In order to solve the above-mentioned problems, the present invention provides the following (1) to (10).

[0009] (1) An abnormality diagnosis system is an abnormality diagnosis system for diagnosing abnormalities of a device, wherein the device performs an operation in a repeated constant pattern, and the abnormality diagnosis system is characterized by having:

[0010] a sampling unit that samples time series data representing the state of the device, the time series data repeating a pattern corresponding to the constant pattern; and

[0011] a computing unit that performs principal component analysis based on the time series data sampled by the sampling unit to diagnose abnormality of the device;

[0012] The above-mentioned operation unit is composed of:

[0013] A first principal component analysis is performed on a pattern in the time series data representing the above-mentioned state, and a first abnormality presence / abnormality determination is performed based on the result.

[0014] When the first abnormality determination is that there is an abnormality, a second principal component analysis is performed on a plurality of temporally continuous patterns including the pattern subjected to the first principal component analysis in the time series data representing the state, and a second abnormality determination is performed based on the result.

[0015] When the second abnormality presence / abnormality determination is that an abnormality exists, it is diagnosed that the device is abnormal.

[0016] (2) The abnormality diagnosis system according to (1) above, characterized in that:

[0017] The first abnormality presence / absence determination and the second abnormality presence / absence determination are performed based on a Q statistic obtained from results of the first principal component analysis and the second principal component analysis.

[0018] (3) The abnormality diagnosis system according to (1) or (2) above, characterized in that:

[0019] When the number of sampling points of the time series data for the first principal component analysis in the above-mentioned operation unit is set to N1 points, the dimensionality compression performed by the above-mentioned first principal component analysis is set to k1 dimensions, the number of sampling points of the time series data for the above-mentioned second principal component analysis is set to N2 points, the dimensionality compression performed by the above-mentioned second principal component analysis is set to k2 dimensions, and the number of the above-mentioned multiple patterns is set to M, where when M≥2, N2<N1×M or k2<k1×M holds.

[0020] (4) The abnormality diagnosis system according to any one of (1) to (3) above, characterized in that:

[0021] The above-mentioned device is a fixed-width extrusion device that reduces the width of the slab by driving a die with a driving motor, and the time series data representing the state of the above-mentioned device is the current waveform of the driving motor or the load waveform of the above-mentioned die.

[0022] (5) The abnormality diagnosis system according to (4) above, characterized in that:

[0023] The device further includes a data preprocessing unit configured to perform preprocessing including normalizing the height of a waveform and classifying the waveform according to a width reduction amount on the time series data sampled by the sampling unit.

[0024] (6) An abnormality diagnosis method is a method for diagnosing abnormality of a device, wherein the device performs an operation in a repeated constant pattern, and the abnormality diagnosis method is characterized by comprising:

[0025] a step of sampling time series data representing the state of the device, the time series data repeating a pattern corresponding to the constant pattern;

[0026] A step of performing a first principal component analysis on a pattern in the time series data representing the above state;

[0027] A step of determining whether a first abnormality exists based on the result of the first principal component analysis;

[0028] When the first abnormality is determined to be abnormal, a step of performing a second principal component analysis on a plurality of temporally continuous patterns including the pattern subjected to the first principal component analysis in the time series data representing the state; and

[0029] The step of determining whether a second abnormality exists based on the result of the second principal component analysis is as follows:

[0030] When the second abnormality presence / abnormality determination is that an abnormality exists, it is diagnosed that the device is abnormal.

[0031] (7) The abnormality diagnosis method according to (6) above, characterized in that:

[0032] The first abnormality presence / absence determination and the second abnormality presence / absence determination are performed based on a Q statistic obtained from results of the first principal component analysis and the second principal component analysis.

[0033] (8) The abnormality diagnosis method according to (6) or (7) above, characterized in that:

[0034] When the number of sampling points of the time series data used for the above-mentioned first principal component analysis is set to N1 points, the dimensionality compression based on the above-mentioned first principal component analysis is set to k1 dimensions, the number of sampling points of the time series data used for the above-mentioned second principal component analysis is set to N2 points, the dimensionality compression based on the above-mentioned second principal component analysis is set to k2 dimensions, and the number of the above-mentioned multiple patterns is set to M, where M≥2, N2<N1×M or k2<k1×M holds.

[0035] (9) The abnormality diagnosis method according to any one of (6) to (8) above, characterized in that:

[0036] The above-mentioned device is a fixed-width extrusion device that reduces the width of the slab by driving a die with a driving motor, and the time series data representing the state of the above-mentioned device is the current waveform of the driving motor or the load waveform of the above-mentioned die.

[0037] (10) The abnormality diagnosis method according to (9) above, characterized in that:

[0038] The method further includes performing pre-processing including normalizing the height of the waveform and classifying the waveform according to the amount of width reduction on the time series data sampled in the above step.

[0039] According to the present invention, it is possible to provide an abnormality diagnosis system and an abnormality diagnosis method that can suppress a misdiagnosis of the occurrence of a device abnormality when no device abnormality actually occurs. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 It is a schematic diagram and a block diagram showing an example of using the abnormality diagnosis system involved in one embodiment for a fixed-width extrusion machine.

[0041] Figure 2 FIG. 1 is a diagram showing an example of a sample waveform when an abnormal waveform is extracted.

[0042] Figure 3 Is used for use Figure 2 A graph of a sample waveform for illustration of PCA performed.

[0043] Figure 4 This is a flowchart showing an abnormality diagnosis method executed in the abnormality diagnosis system according to one embodiment.

[0044] Figure 5 This is a diagram showing a comparison between a case where a load waveform, which is time series data indicating a load, is a normal waveform and a case where it is an equipment abnormal waveform.

[0045] Figure 6 This is a diagram showing a comparison between a case where all load waveforms, which are time series data representing the load, are normal waveforms and a case where a slip occurs.

[0046] Figure 7 This is a diagram showing the relationship between the number of cycles and the Q statistic (mean square error) in the case of a slip occurrence waveform including one cycle (one waveform). DETAILED DESCRIPTION

[0047] Hereinafter, embodiments of the present invention will be described with reference to the drawings.

[0048] In this embodiment, a case where the present invention is applied to abnormality diagnosis of a fixed-width extrusion machine will be described.

[0049] Figure 1 It is a schematic diagram and a block diagram showing an example of using the abnormality diagnosis system involved in one embodiment for a fixed-width extrusion machine.

[0050] The fixed-width extrusion equipment 10 is an equipment for reducing the width (width reduction) of the slab 11 in a hot rolling line after continuous casting.

[0051] The fixed-width extrusion equipment 10 comprises: a pair of dies 2; an inlet-side pinch roller 3 and an outlet-side pinch roller 4; a pair of drive motors 5, which are arranged corresponding to each of the pair of dies 2 and drive the dies 2; and a pair of cranks 6, which connect the dies 2 and the drive motors 5 respectively.

[0052] In the fixed-width extrusion equipment 10, the slab 11 is conveyed between a pair of opposing dies 2 by the inlet-side pinch rollers 3, reduced in width by the pair of dies 2 swinging with power transmitted from the drive motor 5 via the crank 6, and conveyed out by the outlet-side pinch rollers 4. While the slab 11 is intermittently conveyed, the dies 2 repeatedly reduce in width at a constant cycle to form a slab of a specified width.

[0053] The abnormality diagnosis system 20 is a system for diagnosing equipment abnormalities of the fixed-width extrusion equipment 10 , and includes a sampling unit 21 , a data preprocessing unit 22 , a storage unit 23 , a calculation unit 24 , and an output unit 25 .

[0054] The sampling unit 21 is a component that samples the time series data representing the state of the device. In the present embodiment, the time series data represents the current value of the drive motor 5 when the slab 11 is widened by the die 2 and the load of the die 2 and other loads. As time information for obtaining the time series data, crank angle information can be used. At this time, the time series data representing the current value and the load and other loads is data in which one cycle of the crank 6 becomes a waveform (pattern) with a peak, and this waveform appears repeatedly. Specifically, it is a current waveform and a load waveform.

[0055] The data preprocessing unit 22 preprocesses the data sampled by the sampling unit 21. The preprocessing includes normalizing the height of the waveform corresponding to one cycle of the crank 6 and classifying the data according to the width reduction amount (width reduction amount) during the width reduction.

[0056] The storage unit 23 stores the preprocessed data. The storage unit 23 stores time series data (normal waveform) representing the load in a normal state obtained in advance as a reference waveform.

[0057] The calculation unit 24 performs abnormality diagnosis using principal component analysis (PCA) on the waveform of the preprocessed time series data (current value and load) of the load stored in the storage unit 23 . The diagnosis result is output to the output unit 25 .

[0058] The analysis technique using PCA is a technique for extracting a waveform (abnormality) having a shape different from a normal waveform from a plurality of waveforms. PCA and a multivariate statistical process management technique are used to extract abnormal waveforms.

[0059] Figure 2 FIG. 5 is a diagram showing an example of a sample waveform when an abnormal waveform is extracted using PCA. Figure 2 The horizontal axis is time, and the vertical axis is power value and load. Figure 2 In the example, the sample waveforms are waveform P[1], which is a reference waveform for a normal state, and waveform P[2], which is a waveform for an abnormal state (number of waveforms: P=2). For these waveforms P[1] and P[2], N samples (number of samples: N) are collected at a predetermined interval during the waveform collection period. For example, if the collection is performed at a period of 100 ms during a collection period of 5 seconds, the number of samples N is 50.

[0060] exist Figure 3 A conceptual diagram of the case where the PCA is used to analyze it is shown in , and for simplicity, a diagram of a two-dimensional space with two data points is shown. With respect to the points in the space represented by the first element axis (X axis) and the second element axis (Y axis) of the data, the axis with the largest dispersion is set as the first principal component. In addition, the axis orthogonal to the first principal component is set as the second principal component and shown as shown in the figure. In the case of multidimensional data with more than 3 elements, the second principal component axis with the largest dispersion is determined. In the case of N points of data elements, up to N principal components can be calculated. For the kth principal component (k<N), the axis with the largest dispersion among the orthogonal axes to the first to (k-1) principal components is set as the kth principal component.

[0061] In multivariate statistical process management, the Q statistic is evaluated. The Q statistic is the sum of the squares of the distances between each sample and the principal component axis (the sum of square errors), but in the present embodiment, it is preferred to use the mean square error obtained by dividing the value by the number of samples and normalizing it as the Q statistic. At this time, if the number of principal components R is 1, the Q statistic is evaluated only by the distance from the first principal component. Since the distance between abnormal samples and the principal component axis is large, if there are abnormal samples, the Q statistic increases, and it can be determined that the Q statistic is abnormal, and the possibility of equipment abnormality can be grasped. For example, a threshold value of the Q statistic is set, and when the Q statistic is above the threshold value, it can be determined that there is a possibility of equipment abnormality. In Figure 2In the example, waveform P[2] is an abnormal state including an abnormal sample, so the Q statistic becomes larger, and it is determined that there is an abnormality in the Q statistic.

[0062] The operation unit 24 first extracts a waveform of one crank cycle from the pre-processed load time series data stored in the storage unit 23, and performs PCA on the waveform to obtain the Q statistic. In addition, it is determined whether the Q statistic value is abnormal. For example, when the Q statistic is larger than the threshold value, it is determined that the Q statistic is abnormal. If there is no abnormality in the Q statistic, a signal indicating that there is no abnormality is sent to the output unit 25. On the other hand, when there is an abnormality in the Q statistic, there is a possibility of equipment abnormality, but there may also be a situation where the load disappears as described later. Therefore, when there is an abnormality in the Q statistic, in order to confirm which situation it is, multiple continuous waveforms of multiple cycles including the above-mentioned waveform are extracted from the storage unit 23, and PCA is performed on these multiple waveforms together to obtain the Q statistic. If there is no abnormality in the Q statistic, a signal indicating that there is no abnormality is sent to the output unit 25, and if there is an abnormality in the Q statistic, a signal indicating that there is an abnormality is sent to the output unit 25.

[0063] Next, the abnormality diagnosis method in the abnormality diagnosis system 20 will be described.

[0064] Figure 4 : is a flowchart showing the abnormality diagnosis method executed in the abnormality diagnosis system 20 .

[0065] In the fixed width extrusion equipment 10, while the slab 11 is intermittently conveyed, the driving motor 5 transmits power to the die 2 via the crank 6, and the die 2 reduces the width of the slab 11 at a constant cycle to perform fixed width extrusion. Such fixed width extrusion is performed continuously on a plurality of slabs 11.

[0066] In the abnormality diagnosis system 20, abnormality of the fixed width extrusion equipment 10 is diagnosed by the following method based on time series data indicating loads such as the current value of the drive motor 5 and the load of the die 2 during fixed width extrusion.

[0067] First, the abnormality diagnosis system 20 uses the sampling unit 21 to sample time series data of the load during the constant-width extrusion of the slab 11 of a predetermined material (step 1).

[0068] As described above, the time series data representing the load is data in which one cycle of the crank 6 forms a waveform having a peak and such a waveform appears repeatedly. Specifically, such time series data is the current waveform of the drive motor 5 and the load waveform of the mold 2 .

[0069] Next, the abnormality diagnosis system 20 uses the data preprocessing unit 22 to preprocess the data sampled by the sampling unit 21 (step 2).

[0070] As pre-processing of the data, the processing of normalizing the height of the waveform corresponding to one cycle of the crank 6 as described above and the processing of classifying the data according to the width reduction amount (width reduction amount) during the width reduction are performed. The reason for classifying the data according to the width reduction amount (width reduction amount) during the width reduction is that the crank angle at which the load is generated is different due to the difference in the width reduction amount, which causes the waveform deviation. Specifically, the smaller the width reduction amount, the later the time point when the load is generated.

[0071] Such pre-processed data is stored in the storage unit 23 .

[0072] Next, the abnormality diagnosis system 20 extracts one waveform corresponding to one cycle of the crank 6 from the preprocessed time series data of the slab of the material stored in the storage unit 23, and performs PCA on the one waveform (step 3).

[0073] In step 3, PCA is performed using the waveform of the normal time series data (normal waveform) as a reference waveform, and the Q statistic is obtained based on the result using the multivariate statistical process management method described above.

[0074] Next, it is determined whether the Q statistic is abnormal (step 4). For example, when the Q statistic is equal to or greater than a threshold value, it is determined that the Q statistic is abnormal.

[0075] In the event of an equipment failure, since it is an irreversible phenomenon, Figure 5 As shown in FIG. 1 , the load waveform as time series data representing the load becomes an abnormal equipment waveform that is different from the normal waveform in all cycles. Therefore, if there is no abnormality in a waveform equivalent to 1 cycle, it can be considered that there is no equipment abnormality. Figure 5 In the figure, the horizontal axis is the crank angle corresponding to time. In addition, the vertical axis is the load on the die. Figure 6 Same.

[0076] On the other hand, when the PCA process is performed on a waveform and the Q statistic is abnormal, it cannot be concluded that the abnormality is caused by an equipment abnormality. That is, in the actual fixed-width extrusion operation, the phenomenon that the load (load) almost does not work (load disappearance) occurs from time to time due to insufficient material (slab) transportation caused by the slip of the pinch rolls. Figure 6As shown in the figure, the waveform of the cycle where slip occurs is a slip occurrence waveform similar to the equipment abnormality waveform. The load waveform, which is the time series data representing the load, is a state in which the slip occurrence waveform is occasionally inserted into the normal waveform. Therefore, when the slip occurrence waveform is occasionally extracted for PCA, the Q statistic becomes large, just like when the equipment is abnormal. This phenomenon of load failure occurs in about 1 to 2 cycles out of about 10 cycles, but it cannot be predicted.

[0077] Therefore, when the Q statistic is determined to be abnormal by PCA based on a waveform in step 4, multiple continuous waveforms of multiple cycles including a waveform in step 3 are extracted from the preprocessed time series data of the slab of the material stored in the storage unit 23, and PCA is performed on the multiple waveforms (step 5).

[0078] In step 5, the waveform of the normal time series data (normal waveform) is used as a reference waveform, and PCA is performed on multiple waveforms of multiple cycles (multiple patterns continuous in time) together (simultaneously). Based on the result, the Q statistic is calculated using the multivariate statistical process management method as described above.

[0079] If there is no abnormality in the Q statistic in step 4 , a signal indicating that there is no device abnormality is sent to the output unit 25 .

[0080] Next, it is determined whether the Q statistic obtained in step 5 is abnormal (step 6).

[0081] like Figure 5 As shown, when the abnormality determined in step 4 is an equipment abnormality, the load waveform is different from the normal waveform in all cycles, so when PCA of multiple waveforms is performed in step 5, the Q statistic (mean square error) is the same value as when PCA of one waveform is performed.

[0082] On the other hand, when the abnormality determined in step 4 based on the PCA of one waveform is the load disappearance caused by the pinch roller slippage that occurs sporadically in about 1 to 2 cycles out of 10 cycles, the waveforms other than the slippage waveform in the multiple waveforms of the multiple cycles are as follows: Figure 6 The waveform shown is almost normal, so the Q statistic becomes smaller than the case where there is a device abnormality.

[0083] That is, by performing PCA on a plurality of waveforms of a plurality of cycles at once to obtain the Q statistic, it is possible to distinguish whether the abnormality of the Q statistic when performing PCA on one waveform in step 4 is caused by a device abnormality or by sporadic load disappearance.

[0084] Therefore, in step 6, it is possible to determine whether the device is abnormal or not based on the Q statistic. In the case of an abnormal Q statistic, for example, when the Q statistic is above the threshold, it is diagnosed that there is an equipment abnormality. In the case of a non-abnormal Q statistic, for example, when the Q statistic is less than the threshold, it is diagnosed that there is no equipment abnormality.

[0085] Here, the number of sampling points of the time series data of one mode used in the first principal component analysis is set to N1 points, and the dimension after the dimensionality compression performed by the first principal component analysis is set to k1 dimensions. In addition, the number of sampling points of the time series data of multiple modes used in the second principal component analysis is set to N2 points, the dimension after the dimensionality compression performed by the second principal component analysis is set to k2 dimensions, and the number of modes of the multiple modes is set to M (M≥2), that is, the waveform equivalent to one cycle used in the first principal component analysis is set to M waveform. At this time, in the second principal component analysis, the characteristics of the time series signal of the M waveform become more important than the disorder of the waveform of one cycle, so it is preferable to obtain data in a manner such that the relationship of N2<N1×M or k2<k1×M is established.

[0086] By acquiring such data, when abnormality diagnosis using detailed data of one mode is performed in abnormality diagnosis of one mode, and abnormality diagnosis is performed using multiple modes, the characteristics of one waveform can be preserved to a certain extent, and the shape of the data as an M waveform can be preserved at the same time. As a result, the calculation load can be suppressed, and both the evaluation of one mode and the evaluation of the waveform of the M mode can be achieved. In addition, while the slight disturbance of the waveform of one mode can be detected in the first abnormality determination, the detection focusing on the repeatability of the disturbance of the waveform and the offset (time delay) of the waveform can be performed in the second abnormality determination, and high-precision and high-reliability abnormality detection can be performed.

[0087] Figure 7 This is a graph showing the relationship between the number of cycles and the Q statistic (mean square error) in the case of a slip occurrence waveform containing 1 cycle (1 waveform). As shown in the figure, it can be seen that the greater the number of cycles (i.e., the number of waveforms) used for PCA calculation, the smaller the value of the Q statistic (mean square error), and the easier it is to distinguish between equipment abnormality and load loss. For example, if the Q statistic (mean square error) is greater than 10,000, it can be determined as equipment abnormality, and if it is less than 10,000, it can be determined as load loss caused by slip. In addition, according to Figure 7 It can be seen that if there are more than 3 cycles (the number of waveforms is more than 3), it is easier to distinguish between equipment abnormality and load loss caused by slip.

[0088] As described above, according to the present embodiment, in the abnormality diagnosis system 20 for diagnosing abnormalities of the fixed width extrusion equipment 10, it is possible to suppress the erroneous diagnosis of the occurrence of equipment abnormality when there is actually no equipment abnormality. Therefore, it is possible to reduce the stoppage of equipment accompanied by erroneous diagnosis, and it is possible to suppress the reduction of production efficiency.

[0089] In addition, the preprocessing unit 22, as a preprocessing, in addition to normalizing the height of the waveform corresponding to one cycle of the crank 6, classifies the data according to the width reduction amount (width reduction amount) during widening, thereby being able to suppress the deviation of the waveform caused by the width reduction amount (width reduction amount).

[0090] An embodiment of the present invention has been described above, but the present invention is not limited to the above embodiment, and various modifications can be made within the scope of the concept of the present invention. For example, in the above embodiment, the present invention is applied to abnormality detection of a fixed-width extrusion device, but it can be applied to any device in which the action of the device driving part repeats a constant pattern. In addition, as a repetitive pattern of time series data representing a load, an example of a waveform is shown, but it is not limited to this. In addition, as a time series data representing a load, an example of a current or a load is shown, but it is not limited to this.

[0091] Description of Reference Numerals

[0092] 2…die; 3…inlet side pinch roller; 4…exit side pinch roller; 5…drive motor; 6…crank; 10…fixed width extrusion equipment; 11…slab; 20…abnormal diagnosis system; 21…sampling unit; 22…data preprocessing unit; 23…storage unit; 24…calculation unit; 25…output unit

Claims

1. An abnormality diagnosis system for diagnosing abnormalities of a device, wherein the device performs an operation in a repetitive and constant pattern, wherein the abnormality diagnosis system has: a sampling unit that samples time series data representing a state of the device, the time series data repeating a pattern corresponding to the constant pattern; and a computing unit that performs principal component analysis based on the time series data sampled by the sampling unit to diagnose abnormality of the device, The computing unit is configured as follows: Perform a first principal component analysis on a pattern in the time series data representing the state, and perform a first abnormality determination based on the result. When the first abnormality is determined to be abnormal, a second principal component analysis is performed on a plurality of temporally continuous patterns including the pattern subjected to the first principal component analysis in the time series data representing the state, and a second abnormality is determined based on the result. When the second abnormality presence / abnormality determination is that an abnormality exists, it is diagnosed as an equipment abnormality.

2. The abnormality diagnosis system according to claim 1, characterized in that: The first abnormality presence / abnormality determination and the second abnormality presence / abnormality determination are performed based on a Q statistic obtained from results of the first principal component analysis and the second principal component analysis.

3. The abnormality diagnosis system according to claim 1 or 2, characterized in that: When the number of sampling points of the time series data for the first principal component analysis in the operation unit is set to N1 points, the dimensionality compression performed by the first principal component analysis is set to k1 dimensions, the number of sampling points of the time series data for the second principal component analysis is set to N2 points, the dimensionality compression performed by the second principal component analysis is set to k2 dimensions, and the number of the multiple patterns is set to M, where when M≥2, N2<N1×M or k2<k1×M holds.

4. The abnormality diagnosis system according to any one of claims 1 to 3, characterized in that: The device is a fixed-width extrusion device that reduces the width of a slab by driving a die with a driving motor, and the time series data representing the state of the device is a current waveform of the driving motor or a load waveform of the die.

5. The abnormality diagnosis system according to claim 4, characterized in that: The device further includes a data preprocessing unit that performs preprocessing including normalizing the height of a waveform and classifying the waveform according to a width reduction amount on the time series data sampled by the sampling unit.

6. An abnormality diagnosis method for diagnosing abnormality of a device, wherein the device performs an operation in a repeated constant pattern, wherein the abnormality diagnosis method comprises: a step of sampling time series data representing a state of the device, the time series data repeating a pattern corresponding to the constant pattern; a step of performing a first principal component analysis on a pattern in the time series data representing the state; A step of determining whether a first abnormality exists according to a result of the first principal component analysis; When the first abnormality is determined to be abnormal, a step of performing a second principal component analysis on a plurality of temporally continuous patterns including the pattern subjected to the first principal component analysis in the time series data representing the state; as well as The step of performing a second abnormality determination based on the result of the second principal component analysis, When the second abnormality presence / abnormality determination is that an abnormality exists, it is diagnosed as an equipment abnormality.

7. The abnormality diagnosis method according to claim 6, characterized in that: The first abnormality presence / abnormality determination and the second abnormality presence / abnormality determination are performed based on a Q statistic obtained from results of the first principal component analysis and the second principal component analysis.

8. The abnormality diagnosis method according to claim 6 or 7, characterized in that: When the number of sampling points of the time series data used for the first principal component analysis is set to N1, the dimensionality compression performed by the first principal component analysis is set to k1, the number of sampling points of the time series data used for the second principal component analysis is set to N2, the dimensionality compression performed by the second principal component analysis is set to k2, and the number of the multiple patterns is set to M, where M≥2, N2<N1×M or k2<k1×M holds.

9. The abnormality diagnosis method according to any one of claims 6 to 8, characterized in that: The device is a fixed-width extrusion device that reduces the width of a slab by driving a die with a driving motor, and the time series data representing the state of the device is a current waveform of the driving motor or a load waveform of the die.

10. The abnormality diagnosis method according to claim 9, characterized in that: The method further comprises the step of performing pre-processing including normalizing the height of the waveform and classifying the waveform according to the amount of width reduction on the time series data sampled in the above step.

Images