JP2021149727A - Factor analysis apparatus and factor analysis method - Google Patents

Abstract

To specify, with satisfactory accuracy, a factor which generates unsatisfactory process in a facility.SOLUTION: A factor analysis apparatus 10 according to the present invention for specifying a factor generating unsatisfactory process in a production facility 1 has a data acquiring unit 14 for acquiring time-series data of a plurality of parameters indicative of an operation state of the production facility 1 detected during a monitor period including a time when unsatisfactory process is generated in the production facility 1, a feature amount generating unit 15 for generating a plurality of feature amounts as representative values of the plurality of parameters based on the acquired time-series data, a reference determination unit 16 for determining a determination reference for use in determination as to whether or not the operation state of the production facility 1 is abnormal based on the plurality of generated feature amounts, an abnormality degree calculating unit 17 for calculating an abnormality degree indicating frequency of abnormality of the operation state of the production facility 1 based on the determined determination reference, and a factor specifying unit 18 for specifying a feature amount causing generation of unsatisfactory process in the production facility 1, from among the plurality of feature amounts based on the calculated abnormality degree.SELECTED DRAWING: Figure 3

Description

The present invention relates to a factor analyzer and a factor analysis method for analyzing factors that cause processing defects in equipment.

Conventionally, there have been known devices that detect abnormalities that occur in equipment including various machines and devices (see, for example, Patent Document 1). In the apparatus described in Patent Document 1, feature quantities are generated from the state values obtained by monitoring the equipment, virtual feature quantities are generated by statistical processing, and the equipment is generated based on these feature quantities. Generate an anomaly detection model to detect anomalies.

JP-A-2019-159902

However, in the apparatus described in Patent Document 1, since an abnormality detection model is generated based on a virtual feature amount before a processing defect actually occurs in the equipment, it is necessary to accurately identify the cause of the processing defect in the equipment. Is difficult.

One aspect of the present invention is a factor analyzer for identifying a factor that causes a processing defect in the equipment, and shows an operating state of the equipment detected during a monitoring period including the time when the processing defect occurs in the equipment. A data acquisition unit that acquires time-series data of a plurality of parameters, a feature amount generation unit that generates a plurality of feature amounts as representative values of a plurality of parameters based on the time-series data acquired by the data acquisition unit, and a feature amount generation unit. Based on a plurality of feature quantities generated by the feature quantity generation unit, the criterion determination unit that determines the criterion for determining whether or not the operating state of the equipment is abnormal and the criterion determined by the criterion determination unit Based on the anomaly degree calculation unit that calculates the anomaly degree indicating the frequency of abnormalities in the operating state of the equipment, and the feature amount generation unit, a plurality of feature quantities generated by the feature amount generation unit based on the anomaly degree calculated by the abnormality degree calculation unit. It is provided with a factor specifying unit that identifies a feature amount that causes processing defects in the equipment.

Another aspect of the present invention is a factor analysis method for identifying a factor that causes a processing defect in the equipment, and determines the operating state of the equipment detected during the monitoring period including the time when the processing defect occurs in the equipment. A data acquisition step for acquiring time-series data of a plurality of parameters shown, and a feature quantity generation step for generating a plurality of feature quantities as representative values of a plurality of parameters based on the time-series data acquired in the data acquisition step. , A criterion determination step for determining a criterion for determining whether or not the operating state of the equipment is abnormal based on a plurality of feature quantities generated in the feature quantity generation step, and a criterion determined in the criterion determination step. A plurality of features generated in the feature amount generation step based on the abnormality degree calculation step for calculating the abnormality degree indicating the frequency of abnormality in the operating state of the equipment and the abnormality degree calculated in the abnormality degree calculation step. Includes a factor identification step that identifies the feature quantity that causes processing defects in the equipment from the quantity.

According to the present invention, it is possible to accurately identify the cause of processing defects in equipment.

The schematic diagram which shows typically an example of the production process to which the factor analysis apparatus which concerns on embodiment of this invention is applied. The figure which shows an example of the time series data which shows the operating state of the production equipment of FIG. The figure which shows an example of the time series data of FIG. 2A which is managed for each processing period. The block diagram which shows the whole structure of the factor analysis system including the factor analysis apparatus which concerns on embodiment of this invention. The figure which shows an example of the judgment result based on the judgment standard determined by the reference determination part of FIG. A partially enlarged view of FIG. 4A. The figure which shows an example of the abnormality degree calculated by the abnormality degree calculation part of FIG. A partially enlarged view of FIG. 5A. The figure which shows an example of the abnormality degree calculated by the abnormality degree calculation part of FIG. 3 when the setting of the ν value and the σ value of the kernel function is changed. The figure for demonstrating the setting of the threshold value by the threshold value setting part of FIG. The figure which shows an example of the result of the regression analysis by the contribution calculation part of FIG. The figure which shows an example of the contribution degree calculated by the contribution degree calculation part of FIG. The flowchart which shows an example of the process executed by the factor analysis apparatus which concerns on embodiment of this invention.

Hereinafter, embodiments of the present invention will be described with reference to FIGS. 1 to 10. FIG. 1 is a schematic view schematically showing an example of a production process to which the factor analyzer according to the embodiment of the present invention is applied, and shows a laser blanking step of cutting a sheet-shaped steel plate from a coil material R by laser cutting. show. As shown in FIG. 1, in the laser blanking process, production equipment 1 such as a laser processing machine that cuts a steel plate flowing on a conveyor C at a constant speed and processes it into a plurality of workpieces W1, W2, ..., Wn. And an inspection device 2 for inspecting the processing state of the work Wn by the production equipment 1 are arranged.

The production equipment 1 and the inspection device 2 are connected to an equipment controller 3 such as a control panel, and when the inspection device 2 detects the occurrence of a processing defect such as a cutting defect of the work Wn, the processing defect is detected via the equipment controller 3. Is notified, and the production equipment 1 is stopped as necessary. The production equipment 1 is provided with a plurality of sensors that continuously detect a plurality of parameters indicating the operating state of the production equipment 1, such as the focal length of the laser and the transfer speed of the conveyor C.

2A and 2B are diagrams for explaining time-series data showing an operating state of the production equipment 1. FIG. 2A is a diagram showing an example of time-series data of a plurality of parameters indicating the operating state of the production equipment 1. As shown in FIG. 2A, the detection results of a plurality of (for example, 7) parameters X1 to X7 indicating the operating state of the production equipment 1 are when each of n (for example, 5000) workpieces Wn is processed. It is managed as time series data for each processing period Tn.

As shown in FIG. 2A, the machining period Tn of each work Wn is associated with the detection result of machining defects by the inspection device 2. For example, when the inspection device 2 detects a machining defect of the 5000th work W5000, it is assumed that the machining defect has occurred in the operating state of the production equipment 1 in the machining period T5000, and the machining period T5000 is managed as the time when the machining defect occurs. NS.

The time series data shown in FIG. 2A can also be managed for each of n processing periods Tn. For example, as shown in FIG. 2B, the detection result of processing defects by the inspection device 2 can be managed as time-series data for each of n (for example, 5,000) processing periods Tn. Similarly, the average value and standard deviation of the parameters X1 to X7 for each machining period Tn can be calculated as a feature amount representing each machining period Tn and managed as time series data for each machining period Tn.

In this way, by managing the feature amount indicating the operating state of the production equipment 1 as time-series data for each processing period Tn, the operating state of the production equipment 1 when a processing defect occurs as shown by a broken line in FIG. 2B can be grasped. can do. However, as shown in FIG. 2B, since the frequency of processing defects of the work Wn is extremely low (for example, 5 out of 5,000 pieces), it is difficult to identify the cause of the processing defects in the production equipment 1. .. Therefore, in the present embodiment, the factor analysis apparatus is configured as follows so that the cause of the processing defect in the production equipment 1 can be accurately identified.

FIG. 3 is a block diagram showing an overall configuration of a factor analysis system 100 including a factor analysis device (hereinafter, device) 10 according to an embodiment of the present invention. As shown in FIG. 3, the factor analysis system 100 includes a production facility 1, an inspection device 2, and a device 10.

The device 10 includes a CPU 11, a memory 12 such as a ROM and a RAM, and a computer having an I / O and other peripheral circuits, and also has an input / output unit 13 such as a keyboard, a mouse, a touch panel, and a liquid crystal display. .. The CPU 11 functions as a data acquisition unit 14, a feature amount generation unit 15, a reference determination unit 16, an abnormality degree calculation unit 17, a factor identification unit 18, and a data output unit 19.

The data acquisition unit 14 acquires the detection result of the processing defect by the inspection device 2 and the time series data of the parameters X1 to X7 in the monitoring period including the time when the processing defect occurs in the production equipment 1. For example, as shown in FIG. 2B, time-series data of parameters X1 to X7 in the machining periods T1 to T5000 is acquired as a monitoring period when a machining defect occurs in 5 out of 5,000 work Wns. The time-series data of the parameters X1 to X7 acquired by the data acquisition unit 14 is stored in the memory 12.

The feature amount generation unit 15 generates the feature amount X (n) for each processing period Tn based on the time series data of the parameters X1 to X7 acquired by the data acquisition unit 14. Specifically, first, the time series data of each parameter X1 to X7 is divided for each machining period Tn, and a plurality of (for example, 7) data sets are generated for each machining period Tn.

Next, as shown in FIG. 2A, each data set is divided into division periods S1 to S8 in which each processing period Tn is divided (for example, divided into 8), and a plurality of data sets (for example, 7 × 8 pieces) are divided for each processing period Tn. Generate a dataset of. Further, for each data set, a plurality of (for example, 7) representative values such as an average value and a standard deviation are calculated, and a plurality of (for example, 7 × 8 × 7) features are generated for each processing period Tn. ..

In other words, the feature amount generation unit 15 generates a multidimensional (for example, 7 × 8 × 7 dimension) feature amount X (n) for each processing period Tn. That is, a multidimensional (for example, 7 × 8 × 7 dimensional) feature whose component is a plurality of (for example, 7 × 8 × 7) features such as the feature amount X1_S1_AVE which is the average value of the division period S1 of the parameter X1. Generate the quantity X (n). The feature amount X (n) generated by the feature amount generation unit 15 is stored in the memory 12.

The reference determination unit 16 determines a determination criterion for determining whether or not the operating state of the production equipment 1 is abnormal, based on the multidimensional feature amount X (n) generated by the feature amount generation unit 15. For example, a determination for determining whether or not the operating state of the production equipment 1 is abnormal by using a learning module such as OCSVM (One-Class Support Vector Machine) suitable for abnormality detection in a multidimensional data set. Generate a model. That is, as shown in FIG. 2B, a data set consisting of 5,000 feature quantities X (1) to X (5000) in which 5 are abnormal and 4995 are normal is learned, and the operating state of the production equipment 1 is abnormal. A judgment model for judging whether or not is is generated.

More specifically, in the reference determination unit 16, whether each feature amount X (n) is abnormal in the multidimensional space corresponding to the dimension of the feature amount X (n) (for example, 7 × 8 × 7 dimension). The identification boundary Th for identifying whether it is normal is set. When learning by the OCSVM learning module, the reference determination unit 16 sets the ν value and the kernel function k (including parameters such as the σ value). The ν value defines the position of the identification boundary Th with respect to the origin O in the multidimensional space, and the larger the ν value, the larger the feature amount X (n) determined to be abnormal. Further, when the kernel function k is, for example, an RBF kernel (The Gaussian RBF kernel; radial basis function), the σ value of the kernel function k defines the shape of the discrimination boundary Th. The kernel function k is not limited to the RBF kernel function.

In other words, using the identification boundary Th, the combination of a plurality of features constituting the multidimensional feature X (n) corresponding to a specific machining period Tn is statistically deviated from the other machining periods. Then, it is determined whether or not the operating state of the production equipment 1 is abnormal. That is, it is determined whether or not the operating state of the production equipment 1 is abnormal, including the case where no processing defect actually occurs. Further, the ratio of the processing period Tn determined to be abnormal is adjusted from the processing periods T1 to T5000 corresponding to the feature quantities X (1) to X (5000) by using the ν value.

4A and 4B are diagrams showing an example of the determination result based on the determination criteria determined by the reference determination unit 16, and determine the value of the feature amount X1_S1_AVE (n) which is one component of the feature amount X (n). If the result is abnormal, it is shown in a white plot, and if the judgment result is abnormal, it is shown in a black plot. As shown in FIGS. 4A and 4B, the feature amount X (n) is abnormal other than when processing defects actually occur in the production equipment 1 five times (shown by broken lines in FIGS. 4A and 4B). Is determined to be. The determination criteria (ν value and σ value of the kernel function k) determined by the reference determination unit 16 are stored in the memory 12.

The anomaly degree calculation unit 17 sets the determination result based on the determination criteria determined by the reference determination unit 16 as “1” in the case of abnormality and “0” in the case of normal, and integrates them by exponential moving average or the like, and accumulates them in a time series. The degree of abnormality p (n) in consideration of sex is calculated. That is, the abnormality degree p (n) indicating the frequency at which the feature amount X (n) for each processing period Tn is determined to be abnormal is calculated based on the determination criteria determined by the reference determination unit 16. The abnormality degree p (n) calculated by the abnormality degree calculation unit 17 is stored in the memory 12. The degree of abnormality p (n) may be calculated by a weighted moving average or the like instead of the exponential moving average.

5A and 5B are diagrams showing an example of the abnormality degree p (n) calculated by the abnormality degree calculation unit 17. As shown by the broken lines in FIGS. 5A and 5B, it is determined that the feature amount X (n) for each processing period Tn is abnormal in response to the occurrence of processing defects five times when the processing defects actually occur in the production equipment 1. The degree of abnormality p (n) indicating the frequency of this is increasing. By paying attention to the period in which the abnormality degree p (n) increases in response to the actual occurrence of processing defects, it is possible to identify the cause of the occurrence of processing defects in the production equipment 1 without being affected by noise. ..

The data output unit 19 generates and outputs a display control signal for displaying a chart of the abnormality degree p (n) for each processing period Tn as shown in FIGS. 5A and 5B on the input / output unit 13 of a liquid crystal display or the like. .. FIG. 6 is a diagram showing an example of the abnormality degree p (n) calculated by the abnormality degree calculation unit 17 when the ν value set by the reference determination unit 16 and the σ value of the kernel function k are changed. The chart of the abnormality degree p (n) displayed in the input / output unit 13 is shown.

As shown in FIG. 6, when the ν value and the σ value of the kernel function k change, the judgment result of the feature amount X (n) for each identification boundary Th and the processing period Tn changes, so that the judgment result is integrated. The shape of the chart at degrees p (n) changes. In the example of FIG. 6, in the chart of the degree of abnormality p (n) when the ν value is set to 0.10 and the σ value of the kernel function k is set to 0.1, the peak corresponding to the actual occurrence of machining defects shown by the broken line. The shape of is sharpened by being separated from the noise component. The reference determination unit 16 compares the chart of the anomaly degree p (n) when the ν value and the σ value of the kernel function k are changed with those at the time of the actual machining defect occurrence, and for example, pattern matching with a predetermined peak shape. Determines the ν value and the σ value of the kernel function k.

The factor identification unit 18 is produced based on the abnormality degree p (n) calculated by the abnormality degree calculation unit 17 based on the judgment criteria (ν value and the σ value of the kernel function k) determined by the reference determination unit 16. Identify the feature amount that causes processing defects in the equipment 1. That is, the feature amount that causes the abnormality degree p (n) to increase is specified from the plurality of feature amounts constituting the feature amount X (n). The factor identification unit 18 has a threshold value setting unit 181, a period setting unit 182, and a contribution calculation unit 183 as functional configurations.

FIG. 7 is a diagram for explaining the setting of the threshold value p0 by the threshold value setting unit 181. The threshold value setting unit 181 sets the threshold value p0 of the abnormality degree p (n) when the operating state of the production equipment 1 shifts from normal to abnormal. Specifically, the threshold value setting unit 181 compares the chart of the abnormality degree p (n) with the time when the actual processing defect occurs, and shows the peak of the abnormality degree p (n) corresponding to the actual occurrence of the processing defect shown by the broken line. The threshold p0 is set so that the maximum value is separated from the noise component. For example, the minimum value among the maximum values of the plurality of peaks of the degree of abnormality p (n) is set as the threshold value p0.

The period setting unit 182 sets the periods t1 to t7 immediately before the operating state of the production equipment 1 shifts from normal to abnormal based on the threshold value p0 set by the threshold value setting unit 181. In other words, the periods t1 to t7 in which the degree of abnormality p (n) increases beyond the threshold value p0 are set. More specifically, the period setting unit 182 sets the periods t1 to t7 from the threshold value p0 immediately before the peak to the maximum value of the peak for all peaks whose maximum value exceeds the threshold value p0. Alternatively, the processing period Tn of several times (for example, 5 times) immediately before reaching the maximum value of each peak may be set as the periods t1 to t7.

The contribution calculation unit 183 calculates the contribution Rc to the abnormality degree p (n) in the periods t1 to t7 set by the period setting unit 182 for each component of the feature amount X (n). Specifically, the contribution calculation unit 183 extracts the feature amounts X (n) of the periods t1 to t7 from all the feature amounts X (1) to X (5000) in the monitoring period, and the feature amount X (n). Regression analysis (multiple regression analysis) such as Lasso regression is performed for each component of.

FIG. 8 is a diagram showing an example of the result of the regression analysis by the contribution calculation unit 183, and shows the correlation between the predicted abnormality degree p (n) <^> obtained by the regression analysis and the actual abnormality degree p (n). Show the relationship. The regression coefficient (standard partial regression coefficient) of each component of the feature amount X (n) is calculated corresponding to the prediction abnormality degree p (n) <^> as shown in FIG. The contribution calculation unit 183 calculates the absolute value of the regression coefficient obtained by the regression analysis as the contribution Rc of each component of the feature amount X (n). Instead of the Lasso regression, the contribution Rc may be calculated by a random forest, XGBost, or the like.

FIG. 9 is a diagram showing an example of the contribution degree Rc of each component of the feature amount X (n) calculated by the contribution degree calculation unit 183, and shows the top 10 components having a high contribution degree Rc. Since the feature amount X (n) of the period t1 to t7 (FIG. 7) in which the abnormality degree p (n) increases beyond the threshold value p0 is extracted and regression analysis is performed to calculate the contribution degree Rc of each component, the abnormality degree is calculated. It is possible to identify a component having a high contribution Rc to an increase in p (n). That is, it is possible to identify a component that causes a processing defect in the production equipment 1 from a plurality of components constituting the feature amount X (n). Further, by regenerating (re-learning) the judgment model using the feature amount X (n) narrowed down to only the upper components having a high contribution Rc as shown in FIG. 9, overfitting can be prevented and the judgment model can be used. Generalization performance can be improved.

FIG. 10 is a flowchart showing an example of processing executed by the device 10, and shows an example of processing executed by the CPU 11 according to a program stored in the memory 12 in advance. The process shown in the flowchart of FIG. 10 is started, for example, when time-series data indicating the operating state of the production equipment 1 is input from the production equipment 1 and the inspection device 2 to the device 10.

As shown in FIG. 10, first, in step S1, time-series data of the monitoring period input from the production equipment 1 and the inspection device 2 is acquired. Next, in step S2, a multidimensional feature amount X (n) for each machining period Tn is generated based on the time series data acquired in step S1. Next, in step S3, a determination model for determining whether or not the operating state of the production equipment 1 is abnormal is generated by using the feature amount X (n) for each machining period Tn generated in step S2 as learning data. ..

Next, in step S4, using the determination model determined in step S3, an abnormality determination is made for the feature amount X (n) generated in step S2 for each processing period Tn. Next, in step S5, the results of the abnormality determination in step S4 are integrated to calculate the abnormality degree p (n). Next, in step S6, it is determined whether or not the shape of the chart of the abnormality degree p (n) calculated in step S5 is appropriate.

If denied in step S6, the process returns to step S3 to change the setting of the determination model. On the other hand, if affirmed in step S6, the process proceeds to step S7, and the chart of the abnormality degree p (n) calculated in step S5 is compared with the actual processing defect occurrence to set the threshold value p0 of the abnormality degree p (n). do. Then, in step S8, the target period of the regression analysis is set based on the threshold value p0 set in step S7.

Next, in step S9, regression analysis is performed for each component of the feature amount X (n) in the target period set in step S8, and the contribution Rc to the abnormality degree p (n) in the range from the threshold value p0 to the maximum value is calculated. do. Next, in step S10, the determination model is regenerated using the feature amount X (n) narrowed down to only the higher-order components having a high contribution Rc calculated in step S9.

In this way, since the feature amount X (n) is generated and learned based on the time-series data of the monitoring period including the time when the actual processing defect occurs, a judgment model reflecting the actual data can be generated. (Steps S1 to S3). When the result of abnormality judgment (discrete value) by the judgment model is integrated and quantified as the degree of abnormality p (n) (continuous value) indicating the frequency of judgment of abnormality, the actual frequency of machining defects is extremely low. However, the period during which the degree of abnormality p (n) increases can be specified (steps S4 to S8). Since the regression analysis is performed by extracting the data of the period in which the abnormality degree p (n) is equal to or higher than the threshold value p0 and increases, the cause of the processing defect in the production equipment 1 is accurately identified without being affected by noise. Can be done (step S9). Further, by re-learning the feature amount X (n) narrowed down to only the components that cause abnormalities, it is possible to prevent over-learning and regenerate a determination model with improved generalization performance (step S10).

According to the embodiment of the present invention, the following effects can be obtained.
(1) The apparatus 10 identifies a factor that causes a processing defect in the production equipment 1. The device 10 acquires time-series data of a plurality of parameters X1 to X7 indicating the operating state of the production equipment 1 detected during the monitoring period T1 to T5000 including the time when the processing defect occurs in the production equipment 1. A feature amount that generates a feature amount X (n) containing a plurality of feature amounts as representative values of a plurality of parameters X1 to X7 based on the time-series data acquired by the acquisition unit 14 and the data acquisition unit 14. The generation unit 15 and the reference determination unit 16 for determining the determination criteria for determining whether or not the operating state of the production equipment 1 is abnormal based on the feature amount X (n) generated by the feature amount generation unit 15. , Calculated by the abnormality degree calculation unit 17 and the abnormality degree calculation unit 17 that calculate the abnormality degree p (n) indicating the frequency of the abnormality of the operating state of the production equipment 1 based on the judgment criteria determined by the reference determination unit 16. Based on the obtained abnormality degree p (n), a component that causes a processing defect in the production equipment 1 is specified from a plurality of components of the feature amount X (n) generated by the feature amount generation unit 15. A factor identification unit 18 is provided (FIG. 3).

By quantifying the result of abnormality judgment by the judgment model that reflects the actual data as the frequency, even if the actual frequency of processing defects is extremely low, the production equipment 1 will not be affected by noise and processing defects will occur. It is possible to accurately identify the factors that occur. In addition, by narrowing down to the specified factors and re-learning, it is possible to prevent overfitting and regenerate a judgment model with improved generalization performance.

(2) The factor identification unit 18 is in a normal operating state of the production facility 1 based on the time-series data acquired by the data acquisition unit 14 and the abnormality degree p (n) calculated by the abnormality degree calculation unit 17. The operating state of the production equipment 1 changes from normal to abnormal based on the threshold setting unit 181 that sets the threshold p0 of the abnormality degree p (n) when shifting from normal to abnormal and the threshold p0 set by the threshold setting unit 181. The period t1 set by the period setting unit 182 for each of the plurality of components of the feature amount X (n) generated by the feature amount generation unit 15 and the period setting unit 182 that sets the periods t1 to t7 immediately before the transition. It has a contribution calculation unit 183 for calculating the contribution Rc to the abnormality degree p (n) in ~ t7 (FIG. 3).

That is, by specifying the period in which the frequency of determination of abnormality increases by the determination model and performing regression analysis for such a period, it is possible to identify the cause of the abnormality without being affected by noise.

In the above, the present invention has been described as a factor analysis apparatus, but the present invention can also be used as a factor analysis method. That is, the factor analysis method is a factor analysis method for identifying the factors that cause processing defects in the production equipment 1, and is the production equipment detected during the monitoring period including the time when the processing defects occur in the production equipment 1. Based on the data acquisition step S1 for acquiring the time-series data of a plurality of parameters indicating the operating state of 1 and the time-series data acquired in the data acquisition step S1, a plurality of feature quantities as representative values of the plurality of parameters are obtained. Based on the feature amount generation step S2 to be generated and the plurality of feature amounts generated in the feature amount generation step S2, a criterion determination step for determining a determination criterion for determining whether or not the operating state of the production equipment 1 is abnormal. S3, in the abnormality degree calculation step S5 for calculating the abnormality degree p (n) indicating the frequency of the abnormality of the operating state of the production equipment 1 based on the judgment criteria determined in the reference determination step S3, and in the abnormality degree calculation step S5. Based on the calculated anomaly degree p (n), a factor specifying step S6 for specifying a feature amount that causes a processing defect in the production equipment 1 from among a plurality of feature amounts generated in the feature amount generation step S2. Includes ~ S9 (FIG. 10).

The above description is merely an example, and the present invention is not limited to the above-described embodiments and modifications as long as the features of the present invention are not impaired. It is also possible to arbitrarily combine one or a plurality of the above-described embodiments and the modified examples, and it is also possible to combine the modified examples.

1 Production equipment, 2 Inspection equipment, 3 Equipment controller, 10 Factor analysis equipment (equipment), 11 CPU, 12 Memory, 13 Input / output unit, 14 Data acquisition unit, 15 Feature quantity generation unit, 16 Reference determination unit, 17 Abnormality Calculation unit, 18 factor identification unit, 19 data output unit, 100 factor analysis system, 181 threshold setting unit, 182 period setting unit, 183 contribution calculation unit, Wn work

Claims (3)

A factor analyzer that identifies the factors that cause processing defects in equipment.
A data acquisition unit that acquires time-series data of a plurality of parameters indicating the operating status of the equipment detected during the monitoring period including the time when the processing defect occurs in the equipment.
Based on the time-series data acquired by the data acquisition unit, a feature amount generation unit that generates a plurality of feature amounts as representative values of the plurality of parameters, and a feature amount generation unit.
A reference determination unit that determines a determination criterion for determining whether or not the operating state of the equipment is abnormal based on a plurality of feature quantities generated by the feature amount generation unit.
An abnormality degree calculation unit that calculates an abnormality degree indicating the frequency of abnormalities in the operating state of the equipment based on the judgment criteria determined by the reference determination unit, and an abnormality degree calculation unit.
Factor identification unit that identifies the feature amount that causes processing defects in the equipment from among the plurality of feature amounts generated by the feature amount generation unit based on the abnormality degree calculated by the abnormality degree calculation unit. A factor analyzer characterized by being equipped with. In the factor analyzer according to claim 1,
The factor identification part
Based on the time-series data acquired by the data acquisition unit and the abnormality degree calculated by the abnormality degree calculation unit, a threshold value of the abnormality degree when the operating state of the equipment shifts from normal to abnormal is set. Threshold setting unit and
Based on the threshold value set by the threshold value setting unit, a period setting unit that sets a period immediately before the operating state of the equipment changes from normal to abnormal, and a period setting unit.
Factor analysis characterized by having a contribution calculation unit for calculating the contribution to the degree of abnormality in the period set by the period setting unit for each of the plurality of features generated by the feature amount generation unit. Device. It is a factor analysis method that identifies the factors that cause processing defects in equipment.
A data acquisition step for acquiring time-series data of a plurality of parameters indicating the operating status of the equipment detected during the monitoring period including the occurrence of a processing defect in the equipment, and a data acquisition step.
A feature amount generation step that generates a plurality of feature amounts as representative values of the plurality of parameters based on the time series data acquired in the data acquisition step, and a feature amount generation step.
Based on the plurality of feature quantities generated in the feature quantity generation step, a criterion determination step for determining a determination criterion for determining whether or not the operating state of the equipment is abnormal, and a criterion determination step.
Based on the judgment criteria determined in the reference determination step, the abnormality degree calculation step for calculating the abnormality degree indicating the frequency of the abnormality in the operating state of the equipment, and the abnormality degree calculation step.
A factor identification step for identifying a feature amount that causes a processing defect in the equipment from a plurality of feature amounts generated in the feature amount generation step based on the abnormality degree calculated in the abnormality degree calculation step. A factor analysis method characterized by including.

Images