Determination system, determination method, and program

Abstract

The purpose of the present disclosure is to provide a determination system for determining an abnormal state on the basis of waveform data. A determination system (1) comprises an acquisition unit (11), a feature amount calculation unit (12), a learning unit (14), and a determination unit (15). The acquisition unit (11) acquires waveform data (31) from a sensor (22) installed in production equipment (2) that is a die press device. The feature amount calculation unit (12) calculates a plurality of feature amounts (33) from the waveform data (31). The learning unit (14) generates a trained model on the basis of the plurality of feature amounts (33) of each of a plurality of pieces of reference waveform data (41). The plurality of pieces of reference waveform data (41) are acquired from the sensor (22). The determination unit (15) determines whether the operation of the production equipment (2) corresponding to the waveform data (31) is normal or abnormal on the basis of a degree of abnormality calculated from the plurality of feature amounts (33) of the waveform data (31) using the trained model. An abnormality in the operation of the production equipment is caused by material buildup.

Description

Determination System, Determination Method, and Program

[0001] The present disclosure relates to a determination system, a determination method, and a program, and particularly to a determination system, a determination method, and a program using waveform data.

[0002] Patent Document 1 discloses a burr detection device that detects burr formation (abnormality) in a press device (production equipment). In the burr detection device of Patent Document 1, the presence or absence of burr formation is determined based on the displacement of a stripper that holds the workpiece.

[0003] However, in the burr detection device of Patent Document 1, when the stripper crushes the burr or the displacement of the stripper is small, it may not be possible to detect burr formation even if it occurs. On the other hand, in the burr detection device of Patent Document 1, if the threshold value of the displacement of the stripper is small, there is a possibility of false detection of burr formation when the operation of the press device is normal.

[0004] Also, in general criteria such as the maximum value and integral value of waveform data, it is necessary to set the criteria according to each mold and workpiece, and it is necessary to set the criteria for waveform data according to the workpiece and mold, making it difficult to use a uniform criterion for multiple production facilities.

[0005] Japanese Patent Application Laid-Open No. 2015-412

[0006] An object of the present disclosure is to provide a determination system, a determination method, and a program that generally determine an abnormal state based on waveform data.

[0007] A determination system according to one aspect of this disclosure comprises an acquisition unit, a feature calculation unit, a learning unit, and a determination unit. The acquisition unit acquires waveform data from a sensor installed in production equipment. The production equipment is a mold press machine. The feature calculation unit calculates a plurality of features from the waveform data. The learning unit generates a trained model based on the plurality of features of each of the plurality of reference waveform data. The plurality of reference waveform data are acquired from the sensor. The determination unit uses the trained model to determine whether the operation of the production equipment corresponding to the waveform data is normal or abnormal based on the degree of abnormality calculated from the plurality of features of the waveform data. The abnormality of the production equipment is a build-up of waste.

[0008] A determination method according to one aspect of this disclosure includes an acquisition step, a feature calculation step, and a determination step. In the acquisition step, waveform data is acquired from a sensor installed on the production equipment. The production equipment is a die press machine. In the feature calculation step, a plurality of features are calculated from the waveform data. In the determination step, a trained model generated based on a plurality of reference waveform data is used to determine whether the operation of the production equipment corresponding to the waveform data is normal or abnormal based on the degree of abnormality calculated from the plurality of features of the waveform data. The plurality of reference waveform data are acquired from the sensor. The abnormality of the production equipment is waste buildup.

[0009] A program according to one aspect of this disclosure causes one or more processors to execute the determination method.

[0010] Figure 1 is a functional block diagram of the judgment system according to Embodiment 1. Figure 2 is a schematic diagram of the production equipment whose operation is judged by the judgment system. Figure 3 is an example of waveform data in the judgment system. Figure 4 is a schematic diagram showing an example of Mahalanobis distance. Figure 5 is a flowchart showing the operation of the judgment system. Figure 6 is a schematic diagram showing the method for calculating feature quantities in the judgment system according to Embodiment 2. Figure 7 is a flowchart showing a part of the operation of the judgment system according to Embodiment 3. Figure 8 is a flowchart showing another part of the operation of the judgment system.

[0011] The determination system, determination method, and program according to the embodiments will be described in detail below with reference to the drawings. However, the figures described in the embodiments below are schematic diagrams, and the ratios of the size and thickness of each component do not necessarily reflect the actual dimensional ratios. The configurations described in the embodiments below are merely examples of the disclosure. The disclosure is not limited to the embodiments below, and various modifications are possible depending on the design, etc., as long as the effects of the disclosure can be achieved.

[0012] (Embodiment 1) (1) Overview The judgment system 1 (see Figure 1) is a system that determines whether the operation of the production equipment 2 (see Figure 2) is in an abnormal state based on waveform data 31 (see Figure 3) that shows one of a plurality of operations that are repeatedly performed in the production equipment 2 (see Figure 2). The production equipment 2 is, for example, a die press machine that uses a die 21. Specifically, an abnormal state in the operation of the production equipment 2 refers to a slag ejection failure in the die 21. Slag ejection refers to the fact that in a hole-punching process using the die 21, a portion 24 of the punched material 23 (hereinafter referred to as "slag") is not discharged properly and remains inside the die 21. A slag ejection failure means that the processing of the material 23 cannot be performed properly because of the presence of slag 24 inside the die 21.

[0013] A sensor 22 is installed in the production equipment 2. The sensor 22 outputs sensor data 32 that indicates multiple operations that are repeatedly performed in the production equipment 2. Specifically, the sensor 22 is a strain sensor that detects the strain of the mold 21. The sensor data 32 is data that shows the time series of the strain value of the mold 21. The sensor data 32 includes one or more waveform data 31. Specifically, one waveform data 31 is the part of the sensor data 32 that indicates one operation among the operations that are repeatedly performed in the production equipment 2. Among the multiple pressing operations, the part from the start time t0 (see Figure 3) to the end time of one pressing operation corresponds to one waveform data 31. The waveform data 31 is discontinuous data that has a start point and an end point. Here, "discontinuous data" means data of finite length. Also, the waveform data 31 is data that shows the time series change of one or more values. In Embodiment 1, the waveform data 31 shows the time series change of the strain value of the mold 21. Note that the strain values ​​in Figure 3 are relative to the reference value.

[0014] (2) Configuration of the Judgment System The judgment system 1, as shown in Figure 1, comprises an acquisition unit 11, a feature calculation unit 12, a storage unit 13, a learning unit 14, a judgment unit 15, and an output unit 16.

[0015] The acquisition unit 11 acquires waveform data 31. Specifically, the acquisition unit 11 acquires sensor data 32 from the sensor 22 and extracts one or more waveform data 31 from the sensor data 32.

[0016] As shown in Figure 3, the waveform data 31 is data that shows the time-series change of a single data point. Here, the single data point is the strain value of the mold 21.

[0017] The acquisition unit 11 acquires information from the production equipment 2, for example, indicating the start and end times of each of a plurality of repeated operations, and extracts one or more waveform data 31. For example, if the production equipment 2 is a die press machine as shown in Figure 2, the acquisition unit 11 acquires the height of the upper die 26 (stripper) or the rotation angle of the crank that operates the upper die 26, and extracts one or more waveform data 31 from the sensor data 32. The sensor 22 may also output one waveform data 31 each time the production equipment 2 performs each of the plurality of operations, in conjunction with the operation of the production equipment 2.

[0018] The feature calculation unit 12 calculates a plurality of feature quantities 33 from the waveform data 31. Each of the plurality of feature quantities 33 is calculated from partial data, which is a part of the waveform of the waveform data 31. The plurality of feature quantities 33 includes one or more first feature quantities 34 and one or more second feature quantities 35. The partial data is, for example, the initial waveform portion 351, the waveform corresponding to the stripper process 352, the waveform corresponding to the extraction process 353, or the waveform corresponding to the bottom dead center process 354, as will be described later.

[0019] The feature calculation unit 12 extracts one or more partial data from the waveform data 31 and calculates multiple feature quantities 33 from the one or more partial data. The feature calculation unit 12 calculates multiple feature quantities from the one or more partial data, either from the values ​​contained in the partial data or from the values ​​contained in the partial data. The values ​​calculated from the values ​​contained in the partial data include, for example, the rate of change, the integral value, the frequency feature quantity, entropy, variance, and curve fitting coefficient. The frequency feature quantity includes, for example, the spectral energy and the value that can be calculated by the Fast Fourier Transform (FFT).

[0020] The first feature 34 has a higher sensitivity than the second feature 35 in indicating whether the operation of the production equipment 2 corresponding to the waveform data 31 is normal or abnormal. Here, "feature 33 has a high sensitivity in indicating whether the operation of the production equipment 2 corresponding to the waveform data 31 is normal or abnormal" means that the difference between feature 33 of waveform data 31 where the operation of the corresponding production equipment 2 is normal and feature 33 of waveform data 31 where the operation of the corresponding production equipment 2 is abnormal is significant. In other words, "feature 33 has a high sensitivity in indicating whether the operation of the production equipment 2 corresponding to the waveform data 31 is normal or abnormal" means that there is a high degree of accuracy in determining whether the operation of the production equipment 2 corresponding to the waveform data 31 is normal or abnormal based on feature 33.

[0021] Furthermore, the second feature 35 has a lower dependence than the first feature 34 on the relationship between a portion of the waveform data 31 corresponding to the second feature 35 and the start time t0 of the waveform data 31 (see Figure 3). Here, "feature 33 has a low dependence on the relationship between a portion of the waveform data 31 corresponding to feature 33 and the start time t0 of the waveform data 31" means that feature 33 is less affected by the time difference between the start time t0 of the waveform data 31 and the start time of the operation of the production equipment 2, or by the time difference between a specific process of the production equipment 2 estimated by the feature calculation unit 12 for a portion of the waveform data 31 and a specific process of the actual production equipment 2. In other words, "feature 33 has a low dependence on the relationship between a portion of the waveform data 31 corresponding to feature 33 and the start time t0 of the waveform data 31" means that even if there is a time difference between the time of some or all of the waveforms in the waveform data 31 and the time of the actual operation of the production equipment 2, the value of feature 33 is less affected by the time difference.

[0022] The first feature 34 includes, for example, one or more of the following: the time rate of change of the initial waveform portion 351 of the waveform data 31, the integral value of the initial waveform portion 351 of the waveform data 31, and the integral value of the waveform 352 corresponding to the stripper process among the waveform data 31. In other words, the first feature 34 is a value calculated from the values ​​included in the partial data.

[0023] The initial waveform section 351 corresponds to the period from time t1 to time t2 in the waveform data 31, before the upper die 26 contacts the material 23 in the production equipment 2. If no sludge buildup occurs, the material 23 does not contact the upper die 26 during the period from time t1 to time t2. In contrast, if sludge buildup occurs, the upper die 26 may come into contact with the material 23 during the period from time t1 to time t2. Therefore, when sludge buildup occurs, the value of the first feature quantity 34 tends to change more than when sludge buildup does not occur. Furthermore, the waveform 352 corresponding to the stripper process corresponds to the period from time t2 to time t3 in the waveform data 31, when the upper die 26 presses the material 23 in the production equipment 2. When sludge buildup occurs, sludge 24 may get stuck between the upper die 26 and the mold 21, so the value of the first feature quantity 34 tends to change more than when sludge buildup does not occur. In other words, the partial data includes partial data showing the strain before and after the time when the upper mold 26 clamps the material 23.

[0024] The second feature 35 includes, for example, one or more of the maximum value of waveform 353 corresponding to the extraction process of waveform data 31, and the maximum value of waveform 354 corresponding to the bottom dead center process of waveform data 31. In other words, the second feature 35 is a value included in the partial data.

[0025] The waveform 353 corresponding to the punching process is the portion of the waveform data 31 from time t4 to time t5, which corresponds to the period when the punch punches out the material 23 in the production equipment 2. When sludge buildup occurs, sludge 24 may get stuck between the upper die 26 and the mold 21, so the value of the second feature 35 tends to change more than when sludge buildup does not occur. The waveform 354 corresponding to the bottom dead center process is the portion of the waveform data 31 from time t6 to time t7, which corresponds to the period when the upper die 26 has fully lowered in the production equipment 2. When sludge buildup occurs, deformation of the mold 21 may occur due to the sludge 24, so the value of the second feature 35 tends to change more than when sludge buildup does not occur.

[0026] Each of the multiple feature quantities 33 includes both one or more first feature quantities 34 and one or more second feature quantities 35. This allows the determination unit 15, described later, to determine with high accuracy when the operation of the production equipment 2 is abnormal, and to reduce the possibility of falsely detecting an abnormality when the operation of the production equipment 2 is normal.

[0027] Furthermore, in the multiple feature quantities 33, it is preferable that the number of first feature quantities 34 and the number of second feature quantities 35 are equal. This reduces both the possibility of the determination unit 15 overlooking an abnormality in the operation of the production equipment 2 and the possibility of it mistakenly detecting the normal operation of the production equipment 2 as abnormal. In addition, it is preferable that the multiple feature quantities include multiple first feature quantities 34 and multiple second feature quantities 35. This makes it possible to further improve the accuracy of the determination.

[0028] The memory unit 13 is a storage medium that stores multiple waveform data. The multiple waveform data includes waveform data 31 for determining whether the operation of the corresponding production equipment 2 is normal or abnormal. The multiple waveform data also includes multiple reference waveform data 41 that represent the normal operation of the corresponding production equipment 2. Here, "waveform data that represents the normal operation of the corresponding production equipment 2" refers to waveform data that corresponds to the operation of the production equipment 2 so that the processed material 23 does not become a defective product, and does not necessarily mean that there are absolutely no abnormalities in the operation of the production equipment 2 corresponding to the reference waveform data. The memory unit 13 also stores feature quantities 33 that correspond to each of the multiple reference waveform data 41.

[0029] The learning unit 14 generates a trained model based on the multiple feature quantities 33 of each of the multiple reference waveform data 41. The trained model includes, for example, a unit space 51. More specifically, the learning unit 14 creates a unit space 51 using the multiple feature quantities 33 of each of the multiple reference waveform data 41. Here, the unit space 51 is a region in a multidimensional space with each of the multiple feature quantities 33 as an axis, where multiple points corresponding to the multiple reference waveform data 41 are gathered. For example, if the feature quantity 33 includes one first feature quantity 34 and one second feature quantity 35, the unit space 51 is a region in a two-dimensional plane, as shown in Figure 4.

[0030] Here, the multiple feature quantities 33 calculated by the feature quantity calculation unit 12 from the waveform data 31 and the multiple feature quantities 33 of each of the multiple reference waveform data 41 used to generate the unit space 51 correspond to each other. More specifically, one feature quantity 33 of the waveform data 31 and one feature quantity 33 of one reference waveform data 41 that correspond to each other represent the same feature of the waveform data 31 or the one reference waveform data 41. For example, one or more feature quantities 33 calculated by the feature quantity calculation unit 12 from the waveform data 31 and one or more feature quantities 33 of each of the multiple reference waveform data 41 include the integral value of the initial waveform section 351 and the maximum value of the waveform 354 corresponding to the bottom dead center. The integral value of the initial waveform section 351 of the waveform data 31 and the integral value of the initial waveform section 351 of one reference waveform data 41 correspond to each other. The maximum value of the waveform 354 corresponding to the bottom dead center of the waveform data 31 and the maximum value of the waveform 354 corresponding to the bottom dead center of one reference waveform data 41 correspond to each other.

[0031] The determination unit 15 determines whether the operation of the production equipment 2 corresponding to the waveform data 31 is normal or abnormal, based on the abnormality score calculated from multiple feature quantities 33 of the waveform data 31. More specifically, the determination unit 15 calculates the abnormality score from multiple feature quantities of the waveform data 31 using a trained model generated by the learning unit 14. The abnormality score calculated from the multiple feature quantities 33 is based, for example, on the multiple feature quantities 33 and the unit space 51. Specifically, the abnormality score calculated from the multiple feature quantities 33 is the Mahalanobis distance d1 (see Figure 4) or Euclidean distance between the multiple feature quantities 33 and the unit space 51, or the cosine similarity between the multiple feature quantities 33 and the unit space 51. The trained model including the unit space 51 may also be a support vector machine (SVN), and the abnormality score based on the multiple feature quantities 33 is a value indicating whether or not it is an outlier. The trained model may also be created using k-nearest neighbors, neural networks (NN), or clustering. In this embodiment, the degree of abnormality is the Mahalanobis distance d1 between the multiple feature quantities 33 and the unit space 51. The Mahalanobis distance d1 is the distance between a point corresponding to one of the multiple feature quantities 33 of the waveform data 31 and the unit space 51 in a multidimensional space with each of the multiple feature quantities 33 as an axis. The determination unit 15 determines that the operation of the production equipment 2 corresponding to the waveform data 31 is abnormal if the Mahalanobis distance d1 is greater than or equal to a threshold. The determination unit 15 also determines that the operation of the production equipment 2 corresponding to the waveform data 31 is normal if the Mahalanobis distance d1 is less than a threshold.

[0032] More specifically, the learning unit 14 and the determination unit 15 perform the following processes, for example.

[0033] The learning unit 14 normalizes each feature 33 of one or more reference waveform data 41. Here, the feature calculation unit 12 calculates k (k is an integer of 2 or more) feature 33 for each of the waveform data 31 and one or more reference waveform data 41. The k feature 33 include, for example, the integral value of the initial waveform unit 351 and the maximum value of the waveform 354 corresponding to the bottom dead center. The number of reference waveform data 41 is n (n is an integer of 2 or more). That is, k feature 33 are calculated for each of the n reference waveform data 41.

[0034] The learning unit 14 normalizes the n reference waveform data 41. Specifically, for each of the k feature quantities 33, it normalizes them so that the mean of the n feature quantities 33 is 0, and the standard deviation of the n feature quantities 33 is 1.

[0035] Specifically, the normalized feature Yij of the j-th feature (j is an integer between 1 and k) among the k feature quantities 33, which corresponds to the i-th reference waveform data 41 (i is an integer between 1 and n) among the n reference waveform data 41, is expressed by the following formula, using the j-th feature quantity yij corresponding to the i-th reference waveform data 41, the mean value yj of the j-th feature quantity 33, and the standard deviation σj of the j-th feature quantity 33.

[0036] Yij = (yij - yj) / σj ... (Equation 1) Here, the mean value yj of the j-th feature 33 is the arithmetic mean of the j-th feature y1j to ynj corresponding to the n reference waveform data 41. Also, the standard deviation σj of the j-th feature 33 is the standard deviation of the j-th feature y1j to ynj corresponding to the n reference waveform data 41. Note that the j-th feature yij corresponding to the i-th reference waveform data 41 represents the same feature of the waveform data for all i, for example, the integral value of the initial waveform section 351.

[0037] The learning unit 14 calculates the normalized feature Yij for all i and j using the above (Equation 1).

[0038] Next, the learning unit 14 creates a correlation coefficient matrix. The correlation coefficient matrix is ​​a k x k square matrix. The diagonal elements of the correlation coefficient matrix are 1. Also, the elements r in the p x q column of the correlation coefficient matrix (where p and q are integers less than or equal to k, and p ≠ q) ｐｑ It is given by the following formula.

[0039]

[0040] Next, the learning unit 14 calculates the inverse matrix of the correlation coefficient matrix.

[0041] The determination unit 15 normalizes each feature amount 33 of the waveform data 31. Here, the determination unit 15 performs normalization based on the same criteria as the feature amounts 33 of the n reference waveform data 41. That is, the j-th normalized feature amount Ytj of the waveform data 31 is represented by the following formula using the j-th feature amount ytj of the waveform data 31, the average value yj of the j-th feature amount 33, and the standard deviation σj of the j-th feature amount 33.

[0042] Ytj = (ytj - yj) / σj... (Formula 3) Next, the determination unit 15 calculates the Mahalanobis distance d1 (see FIG. 4) using the inverse matrix of the correlation coefficient matrix and the feature amounts 33 of the normalized waveform data 31. The Mahalanobis distance d1 is given by the following formula using the normalized feature amount Ytj of the waveform data 31 and the inverse matrix A of the correlation coefficient matrix.

[0043]

[0044] The determination unit 15 determines whether the operation of the production facility 2 corresponding to the waveform data 31 is normal or abnormal by determining whether the calculated Mahalanobis distance d1, which is the degree of abnormality, is greater than or equal to the threshold value or less than the threshold value. Specifically, if the Mahalanobis distance d1 is greater than or equal to the threshold value, the determination unit 15 determines that the operation of the production facility 2 corresponding to the waveform data 31 is abnormal. On the other hand, if the Mahalanobis distance d1 is less than the threshold value, the determination unit 15认定 that the operation of the production facility 2 corresponding to the waveform data 31 is normal.

[0045] The output unit 16 outputs the determination result of the determination unit 15 to the storage unit 13. Further, the output unit 16 outputs the plurality of feature amounts 33 used for the determination to the storage unit 13. That is, the storage unit 13 stores the determination result of the determination unit 15 and the plurality of feature amounts 33.

[0046] For example, when it is determined that the operation of the production facility 2 corresponding to the waveform data 31 is abnormal, the output unit 16 outputs a stop command to the production facility 2. Further, the output unit 16 performs an output for presenting the determination result to, for example, the operator of the production facility 2. The output unit 16 has, for example, a display control function for causing a display device to display an image indicating the determination result to the operator of the production facility 2. The image includes, for example, the presence or absence of occurrence of an abnormality in the operation of the production facility 2, the frequency of occurrence of the abnormality or the tendency of occurrence of the abnormality, and candidates for abnormal locations of the production facility 2 assumed from the frequency of occurrence of the abnormality or the tendency of occurrence. The display device may be a display included in the determination system 1 or an information terminal device capable of communicating with the determination system 1.

[0047] The determination system 1 includes a computer system. The computer system mainly consists of a processor and memory as hardware. The function of the determination system 1 in this disclosure is realized by the processor executing a program recorded in the memory of the computer system. The program may be pre-recorded in the memory of the computer system, provided via a telecommunications line, or provided on a non-temporary recording medium such as a memory card, optical disk, or hard disk drive that can be read by the computer system. The processor of the computer system consists of one or more electronic circuits including semiconductor integrated circuits (ICs) or large-scale integrated circuits (LSIs). The integrated circuits such as ICs and LSIs referred to here are named differently depending on the degree of integration, and include integrated circuits called system LSIs, VLSIs (Very Large Scale Integrations), or ULSIs (Ultra Large Scale Integrations). Furthermore, FPGAs (Field-Programmable Gate Arrays) that are programmed after the manufacture of the LSI, or logic devices that allow for the reconstruction of junction relationships or circuit compartments within the LSI, can also be used as processors. Multiple electronic circuits may be integrated onto a single chip or distributed across multiple chips. Multiple chips may be integrated onto a single device or distributed across multiple devices. The computer system referred to here includes a microcontroller having one or more processors and one or more memories. Therefore, the microcontroller also consists of one or more electronic circuits, including semiconductor integrated circuits or large-scale integrated circuits.

[0048] Furthermore, it is not essential for the judgment system 1 to have multiple functions integrated into a single enclosure; the components of the judgment system 1 may be distributed across multiple enclosures. In addition, at least some of the functions of the judgment system 1, for example, some of the functions of the judgment unit 15, may be implemented by the cloud (cloud computing), etc. Also, some or all of the storage unit 13 may be implemented by cloud storage, a database server, etc.

[0049] Conversely, at least some of the functions of the determination system 1, which are distributed among multiple components in Embodiment 1, may be consolidated within a single housing.

[0050] (3) Operation diagram 5 is a flowchart showing the operation of the determination system 1 according to embodiment 1.

[0051] The acquisition unit 11 of the judgment system 1 acquires waveform data 31 (step ST1). The acquisition unit 11 acquires, for example, sensor data 32 from the sensor 22, acquires the start time and end time of operation from the production equipment 2, and extracts waveform data 31 from the sensor data 32.

[0052] Next, the feature calculation unit 12 of the judgment system 1 calculates multiple features 33 (step ST2). The feature calculation unit 12 obtains, for example, the start and end times of each process within the operation of the production equipment 2 from the production equipment 2 and extracts one or more partial data from the waveform data 31. The feature calculation unit 12 calculates multiple features 33 based on the partial waveform. The feature calculation unit 12 may also calculate multiple features 33 using, for example, a trained model that outputs one or more features 33 when waveform data 31 or a partial waveform is input.

[0053] Next, the learning unit 14 of the judgment system 1 generates a trained model. Specifically, the learning unit 14 generates a unit space 51 (step ST3). The learning unit 14 generates the unit space 51 based, for example, on multiple feature quantities 33 of each of the multiple reference waveform data 41 stored in the storage unit 13.

[0054] Next, the determination unit 15 of the determination system 1 determines whether the operation of the production equipment 2 corresponding to the waveform data 31 is normal or abnormal (step ST4). The determination unit 15 calculates the degree of abnormality from the feature quantities 33 of the waveform data 31 using a trained model. For example, the determination unit 15 calculates the Mahalanobis distance d1 between multiple feature quantities 33 of the waveform data 31 and the unit space 51 as the degree of abnormality. Based on the Mahalanobis distance d1, the determination unit 15 determines whether the operation of the production equipment 2 corresponding to the waveform data 31 is normal or abnormal.

[0055] Through the above operations, the judgment system 1 can determine whether the operation of the production equipment 2 is normal or abnormal based on the waveform data 31.

[0056] Here, the judgment system 1 may output the judgment result from the output unit 16. For example, if the output unit 16 determines that the operation of the production equipment 2 is abnormal, it outputs a stop command to the production equipment 2. This makes it possible to prevent a decrease in yield when a decrease in yield may occur due to an abnormality in the production equipment 2. Alternatively, for example, the output unit 16 displays an image on a display device that shows the judgment result to the operator of the production equipment 2. The image includes whether or not an abnormality has occurred in the operation of the production equipment 2, the frequency or tendency of the abnormality to occur, and candidate locations of the abnormality in the production equipment 2 that can be assumed from the frequency or tendency of the abnormality to occur. This makes it easy for the operator of the production equipment 2 to determine whether the abnormality in the production equipment 2 is a temporary problem or whether maintenance of the production equipment 2 is required.

[0057] Furthermore, the determination system 1 determines whether the operation of the production equipment 2 is normal or abnormal based on the similarity between the multiple reference waveform data 41 and the waveform data 31. Therefore, if the production equipment 2 and mold 21 equipped with the sensor 22 are the same for the multiple reference waveform data 41 and the waveform data 31, the determination system 1 can determine whether the operation of the production equipment 2 is normal or abnormal based on the waveform data 31. In other words, if the conditions for the operation of the production equipment 2 related to the waveform data 31 and the operation of the production equipment 2 related to the multiple reference waveform data 41 can be made the same, the determination system 1 can generally determine whether the operation of the production equipment 2 is normal or abnormal. Here, the condition for the operation of the production equipment 2 being the same means that the same waveform data 31 is obtained when the production equipment 2 performs the exact same operation.

[0058] (4) Effect The determination system 1 according to Embodiment 1 comprises an acquisition unit 11, a feature calculation unit 12, a learning unit 14, and a determination unit 15. The acquisition unit 11 acquires waveform data 31 from a sensor 22 installed in the production equipment 2. The production equipment 2 is a mold press machine. The feature calculation unit 12 calculates a plurality of feature quantities 33 from the waveform data 31. The learning unit 14 generates a trained model based on the plurality of feature quantities 33 of each of the plurality of reference waveform data 41. The plurality of reference waveform data 41 are acquired from the sensor 22. The determination unit 15 determines whether the operation of the production equipment 2 corresponding to the waveform data 31 is normal or abnormal based on the Mahalanobis distance d1, which is the degree of abnormality calculated from the plurality of feature quantities 33 of the waveform data 31 using the trained model. An abnormality in the operation of the production equipment 2 is a waste buildup. Thus, according to the determination system 1 according to Embodiment 1, it is possible to determine the waste buildup, which is an abnormal state, based on the waveform data 31.

[0059] Furthermore, in the determination system 1 according to Embodiment 1, each of the multiple feature quantities 33 of the multiple reference waveform data 41 is a value that indicates whether the operation of the production equipment 2 corresponding to each of the multiple reference waveform data 41 is normal or abnormal. The multiple feature quantities 33 of the waveform data 31 and each of the multiple feature quantities 33 of the multiple reference waveform data 41 correspond to each other. When the learning unit 14 generates a trained model, it generates a unit space 51 based on each of the multiple feature quantities 33 of one or more reference waveform data 41 from among the multiple reference waveform data 41 in which the operation of the corresponding production equipment 2 is normal. As a result, according to the determination system 1 according to Embodiment 1, it is possible to determine with high accuracy whether the operation of the production equipment 2 corresponding to the waveform data 31 is normal or abnormal based on the similarity of the feature quantities 33 between the one or more reference waveform data 41 in which the operation of the corresponding production equipment 2 is normal and the waveform data 31.

[0060] Furthermore, in the determination system 1 according to Embodiment 1, the determination unit 15 determines that the operation of the production equipment 2 corresponding to the waveform data 31 is abnormal if the Mahalanobis distance d1, which is the degree of abnormality, is greater than or equal to a threshold. The determination unit 15 determines that the operation of the production equipment 2 corresponding to the waveform data 31 is normal if the Mahalanobis distance d1, which is the degree of abnormality, is less than a threshold. Thus, according to the determination system 1 according to Embodiment 1, it is possible to easily determine whether the operation of the production equipment 2 corresponding to the waveform data 31 is normal or abnormal based on the similarity between the waveform data 31 and a plurality of reference waveform data 41.

[0061] Furthermore, in the determination system 1 according to Embodiment 1, the sensor 22 acquires the distortion of the mold 21 included in the production equipment 2 as waveform data 31. As a result, the determination system 1 according to Embodiment 1 makes it possible to detect abnormalities in the manufacturing process using a mold press device at an early stage.

[0062] Furthermore, in the determination system 1 according to Embodiment 1, the feature calculation unit 12 extracts one or more partial data 351 to 354 from the waveform data 31 and calculates multiple feature quantities 33 from the one or more partial data 351 to 354. Thus, according to the determination system 1 according to Embodiment 1, the feature calculation unit 12 can easily calculate multiple feature quantities 33 from the waveform data 31.

[0063] Furthermore, in the determination system 1 according to Embodiment 1, one or more partial data 351 to 354 include partial data 351 to 352 that show the distortion before and after the time when the upper die 26 clamps the material. As a result, according to the determination system 1 according to Embodiment 1, at least one of the multiple feature quantities 33 has high sensitivity in indicating whether the operation of the production equipment 2 corresponding to the waveform data 31 is normal or abnormal. Therefore, it is possible to increase the abnormality sensitivity of the determination system 1 according to Embodiment 1.

[0064] Furthermore, in the determination system 1 according to Embodiment 1, the feature calculation unit 12 calculates a plurality of feature quantities 33 from each of the one or more partial data 351 to 354, either from the values ​​contained in the partial data 351 to 354 or from the values ​​calculated from the values ​​contained in the partial data 351 to 354. Therefore, the determination system 1 according to Embodiment 1 improves the accuracy of determining whether the operation of the production equipment 2 corresponding to the waveform data 31 is normal or abnormal, and makes it possible to easily calculate the plurality of feature quantities 33.

[0065] Furthermore, the determination system 1 according to Embodiment 1 further includes an output unit 16. The output unit 16 outputs a result indicating whether the operation of the production equipment 2 corresponding to the waveform data 31 determined by the determination unit 15 is normal or abnormal. As a result, according to the determination system 1 of Embodiment 1, for example, the operator of the production equipment 2 can take early action against any abnormalities in the production equipment 2.

[0066] Furthermore, in the determination system 1 according to Embodiment 1, the output unit 16 outputs the result as an image. As a result, according to the determination system 1 according to Embodiment 1, the operator of the production equipment 2 can easily determine whether the abnormality of the production equipment 2 is a temporary problem or whether the production equipment 2 requires maintenance.

[0067] Furthermore, the determination system 1 according to Embodiment 1 further includes a storage unit 13. The storage unit 13 stores the results output by the output unit 16, waveform data 31, and a plurality of feature quantities 33 calculated from the waveform data 31. As a result, according to the determination system 1 of Embodiment 1, the operator of the production equipment 2 can make a more accurate judgment about the state of the production equipment 2.

[0068] Furthermore, the determination method according to Embodiment 1 includes an acquisition step ST1, a feature calculation step ST2, and a determination step ST4. In the acquisition step ST1, waveform data 31 is acquired from a sensor 22 installed on the production equipment 2. The production equipment 2 is a mold press machine. In the feature calculation step ST2, a plurality of feature quantities 33 are calculated from the waveform data 31. In the determination step ST4, a trained model generated based on a plurality of reference waveform data 41 is used to determine whether the operation of the production equipment 2 corresponding to the waveform data 31 is normal or abnormal, based on the degree of abnormality (Mahalanobis distance d1) calculated from the plurality of feature quantities 33 of the waveform data 31. The plurality of reference waveform data 41 are acquired from the sensor 22. The abnormality in the operation of the production equipment 2 is a build-up of waste. The program according to Embodiment 1 causes one or more processors to execute the determination method according to Embodiment 1. As a result, the determination method and program according to Embodiment 1 can determine a build-up, which is an abnormal state, based on the waveform data 31.

[0069] Furthermore, the determination method according to Embodiment 1 includes a learning step ST3. In the learning step ST3, a trained model is generated based on multiple feature quantities 33 of each of the multiple reference waveform data 41. As a result, according to the determination method according to Embodiment 1, it is possible to determine with high accuracy whether the operation of the production equipment 2 corresponding to the waveform data 31 is normal or abnormal, based on the similarity of the feature quantities 33 between the reference waveform data 41 and the waveform data 31.

[0070] (Embodiment 2) (1) In the determination system 1 according to configuration embodiment 2, the method by which the feature calculation unit 12 (see Figure 1) calculates multiple feature quantities 33 from one or more partial data differs from that of embodiment 1.

[0071] In the determination system 1 according to Embodiment 1, for example, the feature calculation unit 12 calculates the integral value of the initial waveform portion 351 (see Figure 2) of the waveform data 31 as the feature quantity 33. In contrast, in the determination system 1 according to Embodiment 2, the feature calculation unit 12 calculates the integral value of the initial waveform portion 351 and the portion of the partial data 352 prior to the peak as the feature quantity 33. In other words, the feature calculation unit 12 calculates the integral value of the partial data extracted using the maximum strain value before and after the time when the upper mold 26 clamps the material 23 as the reference point as the feature quantity 33. As a result, even if there is an error in the time when the upper mold 26 clamps the material 23 between the waveform data 31 and the plurality of reference waveform data 41, it is possible to detect sludge buildup with high accuracy.

[0072] Furthermore, in the determination system 1 according to Embodiment 1, for example, when the feature calculation unit 12 extracts multiple features from partial data, it calculates multiple values ​​included in the partial data as multiple features 33. For example, the feature calculation unit 12 extracts partial data 351 from the waveform data 31a shown in Figure 6 and calculates the values ​​of each of the multiple points 311a to 316a as multiple features 33. Also, for example, the feature calculation unit 12 extracts partial data 351 from the waveform data 31b shown in Figure 6 and calculates the values ​​of each of the multiple points 311b to 316b as multiple features 33.

[0073] Here, points 311a to 316a are extracted using time ta1, which corresponds to the maximum strain value at the time when the upper mold 26 clamps the material 23 in the partial data 351, as the reference point. Similarly, points 311b to 316b are extracted using time tb1, which corresponds to the maximum strain value at the time when the upper mold 26 clamps the material 23 in the partial data 352, as the reference point. Points 311a and 311b correspond, and the time difference between point 311a and time ta1 is equal to the time difference between point 311b and time tb1. Likewise, points 312a and 312b correspond, and the time difference between point 312a and time ta1 is equal to the time difference between point 312b and time tb1. Similarly, points 313a and 313b correspond, and the time difference between point 313a and time ta1 is equal to the time difference between point 313b and time tb1. Similarly, points 314a and 314b correspond, and the time difference between point 314a and time ta1 is equal to the time difference between point 314b and time tb1. Similarly, points 315a and 315b correspond, and the time difference between point 315a and time ta1 is equal to the time difference between point 315b and time tb1. Similarly, points 316a and 316b correspond, and the time difference between point 316a and time ta1 is equal to the time difference between point 316b and time tb1.

[0074] In other words, there is a one-to-one correspondence between the multiple points 311a to 316a and the multiple points 311b to 316b, and the reference point is the maximum value of the strain at the time when the upper mold 26 clamps the material 23. Therefore, even if there is a difference in the time when the upper mold 26 clamps the material 23 between point 311a and point 311b, the feature calculation unit 12 can appropriately calculate multiple feature quantities 33. Furthermore, the feature calculation unit 12 can calculate the waveform of partial data as multiple feature quantities 33. Therefore, compared to the case where one feature quantity 33 (for example, maximum value, minimum value, or integral value) is calculated from one partial data, the possibility of the feature quantities 33 being coincidentally similar when the waveforms of the partial data are different is low.

[0075] Furthermore, the feature calculation unit 12 can, for example, use relative values ​​from point 316a, which is a local maximum, as multiple feature quantities 33 based on a plurality of points 311a to 316a. Similarly, the feature calculation unit 12 can, for example, use relative values ​​from point 316b, which is a local maximum, as multiple feature quantities 33 based on a plurality of points 311b to 316b. This makes it possible to determine whether the operation of the production equipment 2 corresponding to the waveform data 31 is normal or abnormal, based on the multiple feature quantities 33, even when there is a shift in the reference of absolute values ​​among multiple waveform data 31.

[0076] (2) Effects In the determination system 1 according to Embodiment 2, the feature quantity calculation unit 12 calculates a plurality of feature quantities 33 from each of the one or more partial data 351 to 354, based on the values ​​contained in the partial data 351 to 354. Therefore, the determination system 1 according to Embodiment 2 makes it possible to further improve the accuracy of determining whether the operation of the production equipment 2 corresponding to the waveform data 31 is normal or abnormal.

[0077] Furthermore, in the determination system 1 according to Embodiment 2, one or more partial data points 351 to 354 are extracted using the maximum strain values ​​before and after the time when the upper die 26 clamps the material 23 as reference points. Therefore, the determination system 1 according to Embodiment 2 can further improve the accuracy of determining whether the operation of the production equipment 2 corresponding to the waveform data 31 is normal or abnormal.

[0078] (Embodiment 3) (1) Configuration In the determination system 1 according to embodiment 1, partial data is extracted based on the start time and end time of each process within the operation of the production equipment 2. In contrast, the determination system 1 according to embodiment 3 extracts partial data based on the waveform characteristics of the waveform data 31.

[0079] The feature calculation unit 12 extracts partial data based on reference points detected based on the waveform of the waveform data. The feature calculation unit 12 detects the reference points based on the waveform of the waveform data.

[0080] Specifically, the feature calculation unit 12 detects the strain peak (see Figure 3) generated when the upper die 26 presses the material 23 in the production equipment 2 as a reference point. The feature calculation unit 12 extracts partial data from the waveform data 31 based on the reference point. The feature calculation unit 12 may extract partial data so as to include the reference point, or it may extract partial data after or before the reference point. For example, the feature calculation unit 12 extracts waveform 352 as partial data so as to include the reference point, and extracts partial data that occurs before waveform 352 as waveform 351. Alternatively, the feature calculation unit 12 may detect the strain peak (see Figure 3) generated when the punch punches out the material 23 in the production equipment 2 as a reference point, and extract waveform 353 as partial data based on the reference point. Furthermore, the feature calculation unit 12 detects the peak of strain that occurs when the upper mold 26 is fully lowered in the production equipment 2 as a reference point, and extracts the waveform 354 as partial data based on the reference point.

[0081] The feature calculation unit 12 extracts partial data from each of the waveform data 31 using the above-described process, and extracts multiple feature quantities 33. This makes it possible to determine whether the operation of the production equipment 2 is normal or abnormal, even if the time required for each process is not constant in each operation of the production equipment 2.

[0082] Figure 7 is a flowchart detailing the operation of step ST3 (see Figure 5), in which the learning unit 14 of the judgment system 1 generates a unit space 51. The feature calculation unit 12 searches for a reference point for each of the waveform data 31 (step S31). Next, the feature calculation unit 12 extracts the partial data using the reference point as the reference position (step S32) and extracts multiple feature quantities 33 (step S33). The learning unit 14 generates a unit space 51 based on the multiple feature quantities 33 obtained by repeating steps S31 to S33.

[0083] Figure 8 is a flowchart detailing the operation of step ST4 (see Figure 5), in which the determination unit 15 of the determination system 1 determines whether the operation of the production equipment 2 corresponding to the waveform data 31 is normal or abnormal. The feature calculation unit 12 searches for a reference point for partial data in the waveform data 31 (step S41). Next, the feature calculation unit 12 extracts partial data using the reference point as the reference position (step S42) and extracts multiple feature quantities 33 (step S43). The determination unit 15 calculates the degree of abnormality based on the multiple feature quantities 33 of the waveform data 31 and the unit space 51 (step S44).

[0084] (2) Effects In the determination system 1 according to Embodiment 3, one or more partial data 351 to 354 are extracted using the maximum or minimum value of strain before and after the time when the upper mold 26 clamps the material 23 as the reference point. As a result, according to the determination system 1 according to Embodiment 1, even if there is a time lag between multiple operations repeated in the production equipment 2, the accuracy of the feature quantities is improved, making it possible to determine with high accuracy whether the operation of the production equipment 2 corresponding to the waveform data 31 is normal or abnormal.

[0085] Furthermore, in the determination system 1 according to Embodiment 3, the feature calculation unit 12 detects reference points based on the waveform of the waveform data 31. As a result, the determination system 1 according to Embodiment 1 can determine with high accuracy whether the operation of the production equipment 2 corresponding to the waveform data 31 is normal or abnormal.

[0086] (Other modifications according to the embodiment) (1) The determination system 1 according to embodiments 1 to 3 includes a learning unit 14, but the determination system 1 may acquire a unit space 51 generated by another device or another system, for example.

[0087] Furthermore, in the determination system 1 according to embodiments 1 to 3, a unit space 51 is generated each time an operation is performed (step ST3). However, if the multiple reference waveform data 41 are the same, a unit space 51 that has already been generated based on the multiple reference waveform data 41 may be used.

[0088] (Aspect) The determination system (1) according to the first aspect comprises an acquisition unit (11), a feature calculation unit (12), a learning unit (14), and a determination unit (15). The acquisition unit (11) acquires waveform data (31) from a sensor (22) installed in the production equipment (2). The production equipment (2) is a mold press machine. The feature calculation unit (12) calculates a plurality of feature quantities (33) from the waveform data (31). The learning unit (14) generates a trained model based on the plurality of feature quantities (33) of each of the plurality of reference waveform data (41). The plurality of reference waveform data (41) are acquired from the sensor (22). The determination unit (15) determines whether the operation of the production equipment (2) corresponding to the waveform data (31) is normal or abnormal based on the degree of abnormality (Mahalanobis distance d1) calculated from the plurality of feature quantities (33) of the waveform data (31) using the trained model. The malfunction in production equipment (2) is slag buildup.

[0089] According to the determination system (1) described above, it is possible to determine an abnormal condition of "scratchy buildup" based on waveform data (31).

[0090] In the second embodiment of the determination system (1), in the first embodiment, each of the multiple feature quantities (33) of the multiple reference waveform data (41) is a value that indicates whether the operation of the production equipment (2) corresponding to each of the multiple reference waveform data (41) is normal or abnormal. The multiple feature quantities (33) of the waveform data (31) and each of the multiple feature quantities (33) of the multiple reference waveform data (41) correspond to each other. When the learning unit (14) generates a trained model, it generates a unit space (51) based on each of the multiple feature quantities (33) of the multiple reference waveform data (41).

[0091] According to the judgment system (1) described above, it is possible to determine with high accuracy whether the operation of the production equipment (2) corresponding to the waveform data (31) is normal or abnormal, based on the similarity of the feature quantities (33) between a plurality of reference waveform data (41) and the waveform data (31).

[0092] In the third embodiment of the determination system (1), in the second embodiment, the determination unit (15) determines that the operation of the production equipment (2) corresponding to the waveform data (31) is abnormal if the degree of abnormality (Mahalanobis distance d1) is greater than or equal to a threshold. The determination unit (15) determines that the operation of the production equipment (2) corresponding to the waveform data (31) is normal if the degree of abnormality (Mahalanobis distance d1) is less than a threshold.

[0093] According to the determination system (1) described above, it is possible to easily determine whether the operation of the production equipment (2) corresponding to the waveform data (31) is normal or abnormal based on the similarity between the waveform data (31) and one or more reference waveform data (41) in which the operation of the production equipment (2) corresponding to the waveform data (31) is normal.

[0094] In the determination system (1) relating to the fourth embodiment, in any of the first to third embodiments, the production equipment (2) is a mold press machine. The sensor (22) acquires the distortion of the mold (21) included in the production equipment (2) as waveform data (31).

[0095] According to the judgment system (1) described above, it becomes possible to detect abnormalities in the manufacturing process using a mold press device at an early stage.

[0096] In the determination system (1) according to the fifth embodiment, in any of the first to fourth embodiments, the feature calculation unit (12) extracts one or more partial data (351 to 354) from the waveform data (31) and calculates multiple features (33) from one or more partial data (351 to 354).

[0097] According to the determination system (1) described above, the feature calculation unit (12) can easily calculate multiple feature quantities (33) from the waveform data (31).

[0098] In the determination system (1) according to the sixth embodiment, in the fifth embodiment, one or more partial data (351 to 354) includes partial data (351 to 352) that show the strain before and after the time when the upper die (26) clamps the material.

[0099] According to the judgment system (1) described above, at least one of the multiple feature quantities (33) has high sensitivity in indicating whether the operation of the production equipment (2) corresponding to the waveform data (31) is normal or abnormal. Therefore, it is possible to increase the abnormality sensitivity of the judgment system (1).

[0100] In the seventh aspect of the determination system (1), in the fifth or sixth aspect, the feature calculation unit (12) calculates from each of one or more subdata (351 to 354) values ​​contained in the subdata (351 to 354), or values ​​calculated from the values ​​contained in the subdata (351 to 354), as a plurality of features (33).

[0101] According to the judgment system (1) described above, the accuracy of determining whether the operation of the production equipment (2) corresponding to the waveform data (31) is normal or abnormal is improved, and it becomes possible to easily calculate multiple feature quantities (33).

[0102] In the determination system (1) relating to the eighth aspect, in any of the fifth to seventh aspects, one or more partial data (351 to 354) are extracted using the maximum or minimum strain value before and after the time when the upper mold (26) clamps the material (23) as the reference point.

[0103] According to the judgment system (1) described above, it is possible to further improve the accuracy of determining whether the operation of the production equipment (2) corresponding to the waveform data (31) is normal or abnormal.

[0104] In the determination system (1) according to the ninth embodiment, in the eighth embodiment, the feature calculation unit (12) detects the reference point based on the waveform of the waveform data (31).

[0105] According to the judgment system (1) described above, it is possible to determine with high accuracy whether the operation of the production equipment (2) corresponding to the waveform data (31) is normal or abnormal.

[0106] The determination system (1) according to the tenth embodiment further comprises an output unit (16) in any of the first to ninth embodiments. The output unit (16) outputs a result indicating whether the operation of the production equipment (2) corresponding to the waveform data (31) determined by the determination unit (15) is normal or abnormal.

[0107] According to the judgment system (1) described above, for example, an operator of the production equipment (2) can take early action against any abnormalities in the production equipment (2).

[0108] In the determination system (1) according to the eleventh embodiment, in the tenth embodiment, the output unit (16) outputs the result as an image.

[0109] According to the determination system (1) described above, the operator of the production equipment (2) can easily determine whether the abnormality of the production equipment (2) is a temporary problem or whether the production equipment (2) requires maintenance.

[0110] In the twelfth embodiment of the determination system (1), a memory unit (13) is further provided in the tenth or eleventh embodiment. The memory unit (13) stores the result output by the output unit (16), waveform data (31), and a plurality of feature quantities (33) calculated from the waveform data (31).

[0111] According to the judgment system (1) described above, the operator of the production equipment (2) will be able to make a more accurate judgment about the state of the production equipment (2).

[0112] The determination method relating to the 13th embodiment includes an acquisition step (ST1), a feature calculation step (ST2), and a determination step (ST4). In the acquisition step (ST1), waveform data (31) is acquired from a sensor (22) installed on the production equipment (2). The production equipment (2) is a mold press machine. In the feature calculation step (ST2), multiple feature quantities (33) are calculated from the waveform data (31). In the determination step (ST4), a trained model generated based on multiple reference waveform data (41) is used to determine whether the operation of the production equipment (2) corresponding to the waveform data (31) is normal or abnormal based on the degree of abnormality (Mahalanobis distance d1) calculated from the multiple feature quantities (33) of the waveform data (31). The multiple reference waveform data (41) are acquired from the sensor (22). An abnormality in the operation of the production equipment (2) is waste buildup.

[0113] According to the determination method described above, it is possible to determine that the condition of sludge buildup is abnormal based on the waveform data (31).

[0114] The determination method according to the 14th embodiment further includes a learning step (ST3) in the 13th embodiment. In the learning step (ST3), a trained model is generated based on multiple feature quantities (33) of each of the multiple reference waveform data (41).

[0115] According to the determination method described above, it is possible to determine with high accuracy whether the operation of the production equipment (2) corresponding to the waveform data (31) is normal or abnormal, based on the similarity of the feature quantities (33) between the reference waveform data (41) and the waveform data (31).

[0116] The program according to the 15th embodiment causes one or more processors to execute the determination method according to the 13th or 14th embodiment.

[0117] According to the program described above, it is possible to determine an abnormal condition of "scratchy buildup" based on the waveform data (31).

[0118] 1 Judgment System 11 Acquisition Unit 12 Feature Calculation Unit 13 Memory Unit 14 Learning Unit 15 Judgment Unit 16 Output Unit 2 Production Equipment 21 Mold 22 Sensor 23 Material 26 Upper Mold 31 Waveform Data 33 Feature Quantity 34 First Feature Quantity 35 Second Feature Quantity 351 Initial Waveform Unit (Partial Data) 352 Waveform Corresponding to Stripper Process (Partial Data) 353 Waveform Corresponding to Punching Process (Partial Data) 354 Waveform Corresponding to Bottom Dead Center Process (Partial Data) 41 Reference Waveform Data 51 Unit Space ST1 Acquisition Step ST2 Feature Calculation Step ST3 Learning Step ST4 Judgment Step d1 Mahalanobis Distance (Anomaly Score) t0 Start Time

Claims

1. A determination system comprising: an acquisition unit that acquires waveform data from a sensor installed in a mold press machine, which is a production facility; a feature calculation unit that calculates a plurality of feature quantities from the waveform data; a learning unit that generates a trained model based on the plurality of feature quantities of each of the plurality of reference waveform data acquired from the sensor; and a determination unit that determines whether the operation of the production facility corresponding to the waveform data is normal or abnormal (a waste buildup) based on the degree of abnormality calculated from the plurality of feature quantities of the waveform data using the trained model.

2. The determination system according to claim 1, wherein each of the plurality of feature quantities of the plurality of reference waveform data is a value indicating whether the operation of the production equipment corresponding to each of the plurality of reference waveform data is normal or abnormal, the plurality of feature quantities of the waveform data and each of the plurality of feature quantities of the plurality of reference waveform data correspond to each other, and the learning unit generates a unit space based on each of the plurality of feature quantities of the plurality of reference waveform data when generating the trained model.

3. The determination system according to claim 2, wherein the determination unit determines that the operation of the production equipment corresponding to the waveform data is abnormal if the degree of abnormality is greater than or equal to a threshold, and determines that the operation of the production equipment corresponding to the waveform data is normal if the degree of abnormality is less than the threshold.

4. The determination system according to any one of claims 1 to 3, wherein the sensor acquires the distortion of a mold included in the production equipment as waveform data.

5. The determination system according to any one of claims 1 to 4, wherein the feature calculation unit extracts one or more partial data from the waveform data and calculates the plurality of feature quantities from the one or more partial data.

6. The determination system according to claim 5, wherein the one or more of the aforementioned partial data include partial data showing the strain before and after the time when the upper mold clamps the material.

7. The determination system according to claim 5 or 6, wherein the feature calculation unit calculates from each of the one or more partial data values ​​a value included in the partial data, or a value calculated from the value included in the partial data, as the plurality of features.

8. The determination system according to any one of claims 5 to 7, wherein the one or more of the aforementioned partial data are extracted using the maximum or minimum value of strain before and after the time when the upper mold clamps the material as a reference point.

9. The determination system according to claim 8, wherein the feature calculation unit detects the reference point based on the waveform of the waveform data.

10. The determination system according to any one of claims 1 to 9, further comprising an output unit that outputs a result indicating whether the operation of the production equipment corresponding to the waveform data determined by the determination unit is normal or abnormal.

11. The determination system according to claim 10, wherein the output unit outputs the result as an image.

12. The determination system according to claim 10 or 11, further comprising a storage unit that stores the result output by the output unit, the waveform data, and the plurality of feature quantities calculated from the waveform data.

13. A determination method comprising: an acquisition step of acquiring waveform data from a sensor installed in a mold press machine, which is a production facility; a feature calculation step of calculating a plurality of feature quantities from the waveform data; and a determination step of determining whether the operation of the production facility corresponding to the waveform data is normal or abnormal (a waste buildup) based on an abnormality score calculated from the plurality of feature quantities of the waveform data using a trained model generated based on a plurality of reference waveform data acquired from the sensor.

14. The determination method according to claim 13, further comprising a learning step of generating the trained model based on a plurality of features of each of the plurality of reference waveform data.

15. A program that causes one or more processors to execute the determination method described in claim 13 or 14.

Images