JP7290484B2 - Anomaly detection device, anomaly detection system, and anomaly detection method - Google Patents

Description

The present invention relates to an anomaly detection device, an anomaly detection system, and an anomaly detection method.

2. Description of the Related Art Conventionally, a manufacturing line for manufacturing semiconductor products or the like has a plurality of processes, and each process is equipped with a processing apparatus for processing a workpiece (for example, a semiconductor wafer). For example, in the etching process, a plasma etching device is installed as a processing device.

In order to control the quality of products, it is ideal to inspect whether the amount of processing (e.g., the amount of etching) performed on the workpiece is within a predetermined range immediately after processing by the processing equipment. However, it is difficult to perform such an inspection immediately after processing.

Therefore, it is known to perform product quality control by monitoring the state (temperature, pressure, etc.) during processing of the processing equipment related to the processing conditions (process recipe) of the workpiece (for example, patent documents 1). Specifically, it is known that a processing apparatus incorporates a state measuring device for measuring the state of the processing device, and detects an abnormality of the processing device based on various measurement data measured by the state measuring device. .

Moreover, in detecting an abnormality in a processing apparatus, it is necessary to set a threshold for identifying whether or not measurement data is normal. This measurement data depends on the installation location of the processing apparatus, the surrounding environment, and processing conditions. Therefore, in abnormality detection, it is necessary to acquire measurement data during processing of a plurality of workpieces that have been normally processed for each processing device and processing condition, and set a threshold value based on the degree of variation in the measurement data (for example, , see Patent Document 2).

JP 2006-173373 A JP 2018-41217 A

However, in the method described in Patent Document 2, it is necessary to acquire measurement data during machining of a plurality of workpieces in order to set the threshold. In this way, immediately after a new processing device is introduced or a processing condition is changed, there is no measurement data for multiple workpieces processed by the processing device or the processing conditions. Since it is necessary to set a threshold value, the threshold value may not be set to an appropriate value, and the accuracy of anomaly detection may decrease. In addition, as the number of processing devices in a production line and the number of products to be manufactured increase, the number of targets for anomaly detection increases, and manual setting of thresholds may become unrealistic. .

For this reason, when starting mass production of products with new processing equipment or processing conditions different from existing products, at the beginning of mass production, the accuracy of abnormality detection is low and thresholds are not set. There is a possibility that an abnormality in the device will be overlooked without being detected, or that an abnormality will be detected excessively.

SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and aims to enable automatic abnormality detection before the start of mass production and to improve the accuracy of abnormality detection at the beginning of mass production.

The disclosed technology is an anomaly detection device that detects an anomaly based on measurement data measured with respect to a measurement parameter related to a state during processing of a processing device included in a manufacturing line, and includes a plurality of manufactured under first processing conditions. Based on a plurality of measurement data for existing products and measurement data for one prototype manufactured under second processing conditions different from the first processing conditions, obtained when mass-producing products under the second processing conditions an anomaly detection model generation unit for generating an anomaly detection model based on the plurality of measurement data estimated by the mass production measurement data estimation unit; and an anomaly detection execution unit that detects an anomaly of the processing apparatus based on measurement data obtained when the product is mass-produced under the second processing condition and the anomaly detection model. It is a detection device.

According to the present invention, it is possible to automatically detect an abnormality before the start of mass production, and to improve the accuracy of abnormality detection at the beginning of the start of mass production.

It is a figure which shows an example of schematic structure of an abnormality detection system. It is a figure which shows an example of the hardware constitutions of the computer which comprises an anomaly detection system. It is a figure which shows an example of the functional structure of an abnormality detection apparatus. It is a figure which illustrates measurement data. It is a figure explaining the estimation method of the measurement data at the time of mass production by the measurement data estimation part at the time of mass production. It is a figure explaining the estimation method of the measurement data at the time of mass production by the measurement data estimation part at the time of mass production. It is a figure which illustrates the abnormality detection model produced|generated by the abnormality detection model production|generation part. It is a flowchart which shows the flow of a process of an abnormality detection apparatus. 4 is a flowchart showing the flow of model generation processing; 4 is a flowchart showing the flow of abnormality determination processing; It is a figure which shows the example of presentation of the contribution of a measurement parameter with respect to anomaly detection. It is a figure which shows an example of a functional structure of a process quality contribution calculation apparatus. It is a figure which shows the example of presentation of the contribution of a manufacturing process with respect to quality.

An embodiment of the present invention will be described below with reference to the accompanying drawings. In addition, regarding the descriptions of the specifications and drawings according to each embodiment, components having substantially the same functional configuration are denoted by the same reference numerals, thereby omitting overlapping descriptions.

An abnormality detection system 1 according to the present embodiment is a system that detects an abnormality in a processing device provided in each processing step that constitutes a manufacturing line, and calculates the degree of contribution of each processing step to product quality.

First, a schematic configuration of the anomaly detection system 1 will be described. FIG. 1 is a diagram showing an example of a schematic configuration of an anomaly detection system 1. As shown in FIG. An anomaly detection system 1 is applied to a production line 2 . The production line 2 includes a plurality of processing steps, and each processing step is provided with a processing device 20 . The manufacturing line 2 also includes an inspection process as a final process, and an inspection device 30 is provided in the inspection process. For example, the manufacturing line 2 is a semiconductor manufacturing line that processes a semiconductor wafer as a workpiece W in each processing step.

The processing device 20 is a device that processes the workpiece W based on processing conditions. A processing apparatus 20 of the same or different type is provided for each processing step. Examples of the processing device 20 include an etching device, an exposure device, a diffusion furnace, an ion implantation device, and the like. The workpiece W is processed in the order of processing step 1, processing step 2, .

Each processing device 20 has a processing condition setting section 21 and a state measurement section 22 . The processing condition setting unit 21 enables setting of processing conditions for processing the workpiece W by the processing device 20 . For example, when the processing device 20 is an etching device, processing conditions include temperature, pressure, etching time, and the like.

The state measuring unit 22 is a measuring device that measures the state of the processing device 20 while the processing device 20 is processing the workpiece W. As shown in FIG. The measurement data acquired by the state measurement unit 22 is data dependent on processing conditions. The state measuring unit 22 includes a plurality of sensors that measure various measurement parameters related to the state of the processing device 20 . For example, a temperature sensor that measures temperature as a measurement parameter, a pressure sensor that measures pressure as a measurement parameter, and an acceleration sensor that measures vibration as a measurement parameter are included. The state measurement unit 22 outputs measurement data representing the state of the processing device 20 during processing to the abnormality detection system 1 .

The inspection device 30 is an inspection device for inspecting the quality of the workpiece W that has been processed by the processing steps 1 to p. For example, the inspection device 30 is a semiconductor tester that measures various electrical characteristics of products formed on a semiconductor wafer as the workpiece W. As shown in FIG. The inspection device 30 outputs quality data representing product quality to the abnormality detection system 1 .

The abnormality detection system 1 is connected to each processing device 20 included in the production line 2 via a wired or wireless network. The anomaly detection system 1 has an anomaly detection device 200 and a process quality contribution degree calculation device 300 . The anomaly detection device 200 and the process quality contribution degree calculation device 300 are configured by a computer such as a PC (Personal Computer) or a server.

Next, the hardware configuration of the abnormality detection system 1 will be described. FIG. 2 is a diagram showing an example of the hardware configuration of the computer 100 that configures the anomaly detection system 1. As shown in FIG.

The anomaly detection device 200 and the process quality contribution calculation device 300 included in the anomaly detection system 1 may be composed of separate computers 100 or may be composed of one computer 100 .

The computer 100 includes a CPU (Central Processing Unit) 101, a ROM (Read Only Memory) 102, a RAM (Random Access Memory) 103, a HDD (Hard Disk Drive) 104, an input device 105, a display device 106, A communication interface 107 , a connection interface 108 and a bus 109 are provided.

The CPU 101 implements the functions of the anomaly detection system 1 by controlling each part by executing a program. A program executed by the CPU 101 can be recorded on any computer-readable recording medium such as a CD (Compact Disk), DVD, flash memory, or the like.

The ROM 102 stores programs executed by the CPU 101 and various data. A RAM 103 provides a work area for the CPU 101 . The HDD 104 stores programs executed by the CPU 101 and various data. The computer 100 may include an SSD (Solid State Drive) in addition to or instead of the HDD 104 .

The input device 105 inputs information into the device according to a user's operation. The input device 105 is a touch pad, keyboard, mouse, or hardware buttons, but is not limited thereto.

The display device 106 displays a screen according to the user's operation. The display device 106 is a liquid crystal display or an organic EL (Electro Luminescence) display, but is not limited to this. Computer 100 may include a touch panel that integrates a touch pad and display device 106 .

A communication interface 107 connects the computer 100 to a network such as the Internet or a LAN (Local Area Network). The anomaly detection system 1 communicates with external devices on the network via the communication interface 107 . A connection interface 108 connects the anomaly detection system 1 and the production line 2 .

A bus 109 connects the CPU 101, ROM 102, RAM 103, HDD 104, input device 105, display device 106, communication interface 107, and connection interface 108 to each other.

Next, the functional configuration of the abnormality detection device 200 will be described. FIG. 3 is a diagram showing an example of the functional configuration of the abnormality detection device 200. As shown in FIG.

The anomaly detection device 200 includes an existing product measurement data storage unit 201, a measurement data acquisition unit 202, a prototype measurement data storage unit 203, a mass production measurement data storage unit 204, a mass production measurement data estimation unit 205, an anomaly A detection model generation unit 206, an anomaly detection model storage unit 207, an anomaly detection execution unit 208, a regression model generation unit (first regression model generation unit) 210, a regression model storage unit 211, and a contribution degree calculation unit (first 1 contribution calculation unit) 212 , an output unit (first output unit) 213 , and a model update trigger unit 214 . The anomaly detection execution unit 208 includes an anomaly degree calculation unit 209a and a determination unit 209b.

The mass-production measurement data estimation unit 205, the abnormality detection model generation unit 206, the abnormality detection execution unit 208, the regression model generation unit 210, and the contribution degree calculation unit 212 are implemented by the CPU 101 executing programs.

Since the abnormality detection device 200 performs processing for each processing step (each processing device 20), the processing of the abnormality detection device 200 in one processing step will be described below.

The existing product measurement data storage unit 201 stores a plurality of measurement data (existing product measurement data) acquired for existing products. An existing product is a product manufactured from a plurality of workpieces W processed in the manufacturing line 2 based on predetermined processing conditions (first processing conditions). Here, the existing product is, for example, one lot (25 wafers) of semiconductor wafers, and measurement data acquired during processing of each semiconductor wafer is stored in the existing product measurement data storage unit 201 .

The measurement data acquisition unit 202 acquires measurement data from the state measurement unit 22 of each processing device 20 when a prototype or mass-produced product is processed by the processing device 20 . Here, the prototype is a workpiece W (for example, one semiconductor wafer). A mass-produced product means a plurality of workpieces W (for example, one lot of semiconductor wafers) that are processed under the same processing conditions as the prototype after manufacturing the prototype.

The prototype measurement data storage unit 203 stores the measurement data (prototype measurement data) measured by the state measurement unit 22 when the prototype is processed by the processing device 20 . For example, the prototype measurement data storage unit 203 stores measurement data during processing when one semiconductor wafer is processed as a prototype.

The mass-production measurement data storage unit 204 stores measurement data (mass-production measurement data) measured by the state measurement unit 22 when mass-produced products are processed by the processing apparatus 20 . For example, the mass-production measurement data storage unit 204 stores measurement data during processing when semiconductor wafers for a unit quantity N (for example, N=25) are processed as mass-produced products.

FIG. 4 is a diagram illustrating measurement data. FIG. 4A illustrates “existing product measurement data” stored in the existing product measurement data storage unit 201 . FIG. 4B illustrates the “prototype measurement data” stored in the prototype measurement data storage unit 203 .

The existing product measurement data includes temperature, pressure, and vibration measurement values X ₁ to X ₃ , and has N sets of these measurement data. The prototype measurement data includes temperature, pressure, and vibration measurement values X ₁ to X ₃ . There is only one set of measurement data at the time of prototyping.

The mass-production measurement data is the same as the existing product measurement data, and includes temperature, pressure, and vibration measurement values X ₁ to X ₃ , and has N sets of measurement data.

Returning to FIG. 3 , the mass-production measurement data estimation unit 205 stores the existing product measurement data (N sets) stored in the existing product measurement data storage unit 201 and the prototype measurement data stored in the trial production measurement data storage unit 203 . Based on (1 set), the measurement data assumed to be obtained when the product is mass-produced under the second processing conditions (mass-production measurement data) is estimated. For example, the mass-production measurement data estimation unit 205 obtains the difference between the averaged data obtained by averaging the existing product measurement data and the prototype measurement data, and based on this difference, each of the existing product measurement data (N sets) is calculated. By correcting, mass-production measurement data (N sets) is estimated when products are mass-produced under the second processing conditions.

5 and 6 are diagrams for explaining a method of estimating mass-production measurement data by the mass-production measurement data estimation unit 205. FIG. As shown in FIGS. 5 and 6, the measurement data is waveform data including temporal variations in measured values. 5 and 6 show only the waveform data of one of the plurality of measurement parameters (temperature, pressure, vibration). The waveform data has a data length (period) corresponding to the processing time included in the processing conditions.

As shown in FIG. 5A, first, the mass-production measurement data estimation unit 205 reads the existing product measurement data (N sets) from the existing product measurement data storage unit 201, and averages the measured values to obtain the average Generates customized data. Here, the existing product measurement data and averaged data have the same data length T1.

Next, the mass-production measurement data estimation unit 205 reads the prototype measurement data from the prototype measurement data storage unit 203 . Prototype measurement data may differ in processing time and data length from existing product measurement data due to differences in processing conditions. As shown in FIG. 5B, the data length of the prototype measurement data is T2. The figure illustrates the case of T2<T1.

The mass-production measurement data estimation unit 205 uses the dynamic time warping method to obtain a warping path between the averaged data and the prototype measurement data, as shown in FIG. 5B. Then, as shown in FIG. 5C, the mass-production measurement data estimating unit 205 expands or contracts the time series of the averaged data so that the data length becomes equal based on the determined warping path, thereby generating a waveform Align them (matching process). Specifically, the mass-production measurement data estimation unit 205 expands or contracts the time series of the averaged data so that the points connected by the warping path have the same time.

Next, the mass-production measurement data estimation unit 205 obtains an offset value between the aligned averaged data and the prototype measurement data. Specifically, as shown in FIG. 6A, mass-production measurement data estimator 205 obtains a difference (offset value) between data at each time. This offset value corresponds to a value caused by a difference in machining conditions.

Then, as shown in FIG. 6B, the mass-production measurement data estimating unit 205 applies the dynamics shown in FIGS. 5B and 5C to each waveform data included in the existing product measurement data. Position alignment is performed by the time warping method, and offset correction is performed based on the offset value. As a result, mass-production measurement data (N sets) having the same data length T2 as the prototype measurement data and corrected for the offset due to the difference in processing conditions is generated.

Returning to FIG. 3 , the anomaly detection model generation unit 206 performs mass production measurement data estimated by the mass production measurement data estimation unit 205 on the basis of N+1 sets of measurement data obtained by combining the mass production measurement data and the prototype measurement data. An anomaly detection model for detecting an anomaly of the processing device 20 is automatically generated. Specifically, the anomaly detection model generation unit 206 calculates an anomaly detection model using a machine learning algorithm such as Isolation Forest.

The anomaly detection model storage unit 207 stores an anomaly detection model generated by the anomaly detection model generation unit 206 .

FIG. 7 is a diagram illustrating an anomaly detection model generated by the anomaly detection model generation unit 206. As shown in FIG. FIG. 7 shows an anomaly detection model in a two-dimensional space with temperature and pressure as parameters. An anomaly detection model is expressed as a function that outputs an anomaly degree (for example, a degree of deviation from an average) with respect to input measurement data. Note that when there are three or more types of measurement parameters, the anomaly detection model is represented by a function of three or more dimensions.

Note that the anomaly detection model generation unit 206 may generate an anomaly detection model by any method other than the isolation forest, as long as it is possible to obtain a function that outputs the degree of anomaly.

Further, the anomaly detection model generation unit 206 detects an anomaly based on mass-production measurement data stored in the mass-production measurement data storage unit 204 during mass production, not limited to the mass-production measurement data estimated by the mass-production measurement data estimation unit 205 . Allows to generate a detection model.

The anomaly detection model generation unit 206 can receive a trigger signal from the model update trigger unit 214 at any timing. By outputting a trigger signal to the anomaly detection model generation unit 206, the model update trigger unit 214 replaces the measurement data used for anomaly detection model generation with the mass production measurement estimated by the mass production measurement data estimation unit 205. The data is switched to the mass-production measurement data stored in the mass-production measurement data storage unit 204, and the mass-production measurement data used for generating the abnormality detection model is updated.

The model update trigger unit 214 transmits a trigger signal to the anomaly detection model generation unit 206 based on an input signal from the outside. The anomaly detection model generation unit 206 reads mass-production measurement data from the mass-production measurement data storage unit 204 at the timing of receiving a trigger signal input from the model update trigger unit 214, calculates an anomaly detection model, and calculates an anomaly detection model storage unit. 207.

Returning to FIG. 3, the anomaly degree calculation unit 209a included in the anomaly detection execution unit 208 calculates the anomaly degree of the measurement data obtained during mass production based on the anomaly detection model stored in the anomaly detection model generation unit 206. do. For example, the degree-of-abnormality calculation unit 209a calculates the degree of abnormality by inputting mass-production measurement data into a function representing an anomaly detection model. The degree-of-abnormality calculation unit 209a calculates the degree of abnormality for each of a plurality of mass-production measurement data.

The determination unit 209b determines whether there is an abnormality by comparing the degree of abnormality calculated by the degree-of-abnormality calculation unit 209a with a predetermined threshold value. Specifically, when the degree of abnormality is greater than the threshold, the determination unit 209b determines that there is an abnormality, and treats the degree of abnormality as an outlier to the contribution calculation unit 212, the output unit 213, etc., which will be described later. Output. If the degree of abnormality is equal to or less than the threshold, the determination unit 209b determines that there is no abnormality, and outputs the degree of abnormality as a “normal value” to the contribution calculation unit 212, the output unit 213, and the like.

The regression model generation unit 210 generates N+1 sets of measurement data, which is a combination of the mass-production measurement data estimated by the mass-production measurement data estimation unit 205 and the prototype measurement data, and the N+1 sets of measurement data calculated by the abnormality degree calculation unit 209a. Generate a regression model that represents the relationship between a set of measurement data and multiple degrees of anomaly. For example, the regression model generation unit 210 generates a regression model by using multiple measurement parameters as explanatory variables, the degree of abnormality as an objective variable, and performing multiple regression analysis using the following equation (1).

α= _a1X1 + _a2X2 + _{a3X3 +} b ( _{1 )}
Here, a ₁ , a ₂ and a ₃ are partial regression coefficients. b is a constant term.

Note that the regression model generation unit 210 may generate a regression model by a method such as a decision tree or a random forest, without being limited to multiple regression.

The regression model storage unit 211 stores the regression model generated by the regression model generation unit 210 .

Further, the regression model generation unit 210 generates a regression model based on mass-production measurement data stored in the mass-production measurement data storage unit 204 during mass production, not limited to the mass-production measurement data estimated by the mass-production measurement data estimation unit 205. It is possible to generate

The regression model generation unit 210 can receive a trigger signal from the model update trigger unit 214 at arbitrary timing. By outputting a trigger signal to the regression model generation unit 210, the model update trigger unit 214 changes the measurement data used for generating the regression model from the mass production measurement data estimated by the mass production measurement data estimation unit 205. Switching to the mass-production measurement data stored in the mass-production measurement data storage unit 204 and updating of the mass-production measurement data used to generate the regression model are executed.

The model update trigger unit 214 transmits a trigger signal to the regression model generation unit 210 based on an input signal from the outside. The regression model generating unit 210 reads mass-production measurement data from the mass-production measurement data storage unit 204 at the timing of receiving a trigger signal input from the model update trigger unit 214 , calculates an abnormality detection model, and stores the data in the regression model storage unit 211 . Write.

When the anomaly detection model generation unit 206 generates a regression model by regression analysis (for example, when Mahalanobis distance or autoregressive model is used as an anomaly detection model), the process of the regression model generation unit 210 is omitted. , the regression model generated by the anomaly detection model generation unit 206 can be stored in the regression model storage unit 211 .

The contribution calculation unit 212 calculates the contribution of each measurement parameter (explanatory variable) to the degree of abnormality (objective variable) determined to be abnormal by the determination unit 209b, based on the regression model stored in the regression model storage unit 211. . Specifically, the contribution degree calculation unit 212 has a normal data storage unit 212a. If it is determined as such, the mass-production measurement data is stored as normal data. Then, when the anomaly detection execution unit 208 detects an anomaly, that is, when the determination unit 209b determines that the anomaly degree is an “outlier”, the contribution degree calculation unit 212 calculates the mass production measurement data and the normal data. , and by applying this to the regression model, the degree of importance of the explanatory variable is calculated as the degree of contribution.

In response to the abnormality detection by the abnormality detection execution unit 208, the output unit 213 outputs a signal indicating that there is a possibility that the processing apparatus 20 has an abnormality to the display device 106 or the like for display. Further, in response to the contribution degree calculation unit 212 having calculated the degree of contribution, the output unit 213 outputs a signal representing the degree of contribution of each measurement parameter to the abnormality detection to the display device 106 or the like for display.

Next, the flow of processing of the abnormality detection device 200 will be described. FIG. 8 is a flow chart showing the flow of processing of the anomaly detection device 200. As shown in FIG.

First, the abnormality detection device 200 determines whether measurement data is acquired by the measurement data acquisition unit 202 (step S10). When the measurement data is acquired by the measurement data acquisition unit 202 (step S10: YES), model generation processing (step S11) and abnormality determination processing (step S12) are executed.

When the model generation process and the abnormality determination process are completed, the abnormality detection device 200 determines whether or not an end signal has been input from the outside (step S13). If the end signal has not been input (step S13: NO), the process returns to step S10, and if the end signal has been input (step S13: YES), the process ends.

FIG. 9A is a flow chart showing the flow of model generation processing executed in step S11. In this model generation process, first, the abnormality detection device 200 determines whether or not the measurement data acquired by the measurement data acquisition unit 202 is prototype measurement data (step S20). If the obtained measurement data is the prototype measurement data (step S20: YES), the mass production measurement data estimation unit 205 obtains the existing product measurement data from the existing product measurement data storage unit 201 (step S21). , mass-production measurement data is estimated based on the existing product measurement data and the prototype measurement data (step S22).

Next, the abnormality detection model generation unit 206 generates an abnormality detection model based on the mass-production measurement data and the prototype measurement data estimated by the mass-production measurement data estimation unit 205 (step S23). Here, the anomaly detection model generated by the anomaly detection model generation unit 206 is stored in the anomaly detection model storage unit 207 .

Next, the anomaly degree calculation unit 209 a included in the anomaly detection execution unit 208 stores the anomaly degrees of the mass-production measurement data and the prototype measurement data estimated by the mass-production measurement data estimation unit 205 in the anomaly detection model storage unit 207 . calculated based on the anomaly detection model (step S24). Further, the regression model generation unit 210 generates a regression model representing the relationship between the mass-production measurement data and the prototype measurement data estimated by the mass-production measurement data estimation unit 205 and the degree of abnormality calculated by the degree-of-abnormality calculation unit 209a. Generate (step S25). Here, the regression model generated by the regression model generation unit 210 is stored in the regression model storage unit 211. FIG. Thereafter, the anomaly detection device 200 terminates the model generation process and returns the process to the main flow of FIG.

On the other hand, in step S20, when it is determined that the measurement data acquired by the measurement data acquisition unit 202 is not the measurement data for prototype but the measurement data for mass production (step S20: NO), the model update trigger unit 214 transmits a trigger signal, and it is determined whether or not the anomaly detection model generation unit 206 and the regression model generation unit 210 have received it (step S26). If the trigger signal has not been received (step S26: NO), the abnormality detection device 200 terminates the model generation process and returns the process to the main flow of FIG.

On the other hand, when the trigger signal is received (step S26: YES), the abnormality detection model generation unit 206 acquires mass-production measurement data from the mass-production measurement data storage unit 204 (step S27). An anomaly detection model is generated based on the measurement data (step S23). Here, the anomaly detection model generated by the anomaly detection model generation unit 206 is stored in the anomaly detection model storage unit 207 .

After that, the abnormality degree calculation unit 209a calculates the abnormality degree of the mass-production measurement data based on the abnormality detection model (step S24). Further, the regression model generation unit 210 generates a regression model representing the relationship between the measurement data during mass production and the degree of abnormality calculated by the degree of abnormality calculation unit 209a (step S25). Here, the regression model generated by the regression model generation unit 210 is stored in the regression model storage unit 211. FIG. Then, the anomaly detection device 200 terminates the model generation process and returns the process to the main flow of FIG.

FIG. 9B is a flow chart showing the flow of the abnormality determination process executed in step S12. In this abnormality determination process, first, the abnormality detection device 200 determines whether the abnormality detection model generated by the abnormality detection model generation unit 206 is stored in the abnormality detection model storage unit 207, that is, whether the abnormality detection model exists. is determined (step S30). If the abnormality detection model does not exist (step S30: NO), the abnormality detection device 200 terminates the abnormality determination process and returns the process to the main flow of FIG.

If an anomaly detection model exists (step S30: YES), the anomaly degree calculator 209a calculates the anomaly degree of the measurement data acquired in step S10 based on the anomaly detection model (step S31). Then, the determination unit 209b determines whether or not there is an abnormality by comparing the calculated degree of abnormality with a threshold (step S32). When the determination unit 209b determines that there is an abnormality (step S32: YES), the output unit 213 notifies (abnormality notification) that there is a possibility that an abnormality has occurred in the processing apparatus 20 (step S33).

When the determination unit 209b determines that there is no abnormality (step S32: NO), the mass-production measurement data is stored as normal data in the normal data storage unit 212a.

Then, the contribution calculation unit 212 calculates the contribution of each measurement parameter to the abnormality determined in step S32 (step S34), and the output unit 213 presents the contribution to the user (step S35). Thereafter, the anomaly detection device 200 terminates the model generation process and returns the process to the main flow of FIG.

The above processing is executed for each of the processing steps 1 to p.

FIG. 10 is a diagram showing a presentation example of the degree of contribution of measurement parameters to anomaly detection. As shown in FIG. 10, the output unit 213 presents the contribution by causing the display device 106 or the like to display the contribution for each measurement parameter. FIG. 10 shows that the measurement parameter that contributes the most to the degree of anomaly is "pressure".

Next, the functional configuration of the process quality contribution calculation device 300 will be described. FIG. 11 is a diagram showing an example of the functional configuration of the process quality contribution calculation device 300. As shown in FIG.

The process quality contribution degree calculation device 300 includes a quality data acquisition unit 301, a measurement data acquisition unit 302, a regression model generation unit (second regression model generation unit) 303, a regression model storage unit 304, and a contribution degree calculation unit ( It has a second contribution calculation unit) 305 and an output unit (second output unit) 306 . The regression model generation unit 303 and the contribution degree calculation unit 305 are implemented by the CPU 101 executing programs.

During mass production, the process quality contribution calculation device 300 executes processing after the N workpieces W are processed by the production line 2 and inspected.

The quality data acquisition unit 301 acquires product quality data Q obtained by inspecting each workpiece W from the inspection device 30 .

The measurement data acquisition unit 302 acquires mass production measurement data from the state measurement unit 22 of each processing apparatus 20 in the processing steps 1 to p. Note that the measurement data acquisition unit 302 may acquire the mass production measurement data from the mass production measurement data storage unit 204 in the abnormality detection device 200 . The mass-production measurement data may relate to one measurement parameter out of a plurality of measurement parameters, or may be obtained by integrating measurement data of a plurality of measurement parameters. In the following, mass-production measurement data of machining steps 1 to p are expressed as Z ₁ to Z _p .

The regression model generation unit 303 generates a regression model representing the relationship between the quality data Q and mass production measurement data Z ₁ to Z _p . For example, the regression model generation unit 303 uses mass production measurement data Z ₁ to Z _p as explanatory variables, quality data Q as an objective variable, and performs multiple regression analysis using the following equation (2) to generate a regression model. Generate.

Q= _c1Z1 + _c2Z2 + _... + _{cqZq +} d... ₍ 2)
Here, c ₁ to c _q are partial regression coefficients. d is a constant term.

Note that the regression model generation unit 303 may generate a regression model by a method such as a decision tree, random forest, or the like, without being limited to multiple regression.

The regression model storage unit 304 stores the regression model generated by the regression model generation unit 303 .

The contribution calculation unit 305 calculates the importance of the mass production measurement data Z ₁ to Z _p (explanatory variables) with respect to the quality data Q (objective variable) as a contribution based on the regression model stored in the regression model storage unit 304. do.

In response to the calculation of the degree of contribution by the degree-of-contribution calculation unit 305, the output unit 306 outputs a signal representing the degree of contribution to the display device 106 or the like for display.

FIG. 12 is a diagram showing a presentation example of the degree of contribution of the manufacturing process to quality. As shown in FIG. 12, the output unit 306 presents the degree of contribution by causing the display device 106 or the like to display the degree of contribution for each processing step. FIG. 12 shows that the processing step that contributes the most to the quality is "processing step 2".

As described above, according to the present embodiment, the anomaly detection system 1 mass-produces products under the second processing conditions based on existing product measurement data (N sets) and prototype measurement data (1 set). It generates measurement data that is estimated to be obtained at the time of mass production (measurement data during mass production), and automatically generates an anomaly detection model based on the estimated measurement data during mass production. This eliminates the need to do so, and improves the accuracy of abnormality detection at the beginning of mass production.

Further, according to the present embodiment, the abnormality detection system 1 obtains the contribution of the measurement parameter to the abnormality detection and presents it to the user. The causative measurement parameter can be easily identified.

Further, according to the present embodiment, the anomaly detection system 1 obtains the degree of contribution of the manufacturing process to the quality and presents it to the user. can be easily identified.

Although preferred embodiments of the present invention have been described in detail above, the present invention is not limited to the above-described embodiments, and various modifications and substitutions can be made to the above-described embodiments without departing from the scope of the present invention. can be added.

1 Abnormality detection system 2 Production line 20 Processing device 22 State measurement unit 30 Inspection device 200 Abnormality detection device 201 Existing product measurement data storage unit 202 Measurement data acquisition unit 203 Prototype measurement data storage unit 204 Mass production measurement data storage unit 205 Mass production Measurement data estimation unit 206 Abnormality detection model generation unit 207 Abnormality detection model storage unit 208 Abnormality detection execution unit 209a Abnormality degree calculation unit 209b Judgment unit 210 Regression model generation unit (first regression model generation unit)
211 regression model storage unit 212 contribution calculation unit (first contribution calculation unit)
213 output unit (first output unit)
214 model update trigger unit 300 process quality contribution calculation device 301 quality data acquisition unit 302 measurement data acquisition unit 303 regression model generation unit (second regression model generation unit)
304 regression model storage unit 305 contribution calculation unit (second contribution calculation unit)
306 output unit (second output unit)

Claims (10)

An anomaly detection device that detects an anomaly based on measurement data measured for measurement parameters related to the state during processing of a processing device included in a production line,
Based on multiple measurement data for multiple existing products and measurement data for one prototype, multiple measurements assumed to be obtained when the product is mass-produced using the same processing conditions and the same processing equipment as the one prototype a mass-production measurement data estimation unit for estimating measurement data;
an anomaly detection model generation unit that generates an anomaly detection model based on a plurality of pieces of measurement data estimated by the mass production measurement data estimation unit;
An anomaly detection execution unit that detects an anomaly of the processing device based on the measurement data obtained when the product is mass-produced with the same processing conditions and the same processing device as the one prototype, and the anomaly detection model;
An abnormality detection device characterized by comprising: The processing conditions for manufacturing the plurality of existing products and the processing conditions for manufacturing the one prototype are different processing conditions,
2. The abnormality detection device according to claim 1, wherein a processing device for manufacturing the plurality of existing products and a processing device for manufacturing the one prototype are different processing devices. The processing conditions for manufacturing the plurality of existing products and the processing conditions for manufacturing the one prototype are different processing conditions,
2. The abnormality detection device according to claim 1, wherein a processing device for manufacturing the plurality of existing products and a processing device for manufacturing the one prototype are the same processing device. The processing conditions for manufacturing the plurality of existing products and the processing conditions for manufacturing the one prototype are the same processing conditions,
2. The abnormality detection device according to claim 1, wherein a processing device for manufacturing the plurality of existing products and a processing device for manufacturing the one prototype are different processing devices. The processing conditions for manufacturing the plurality of existing products and the processing conditions for manufacturing the one prototype are the same processing conditions,
2. The abnormality detection device according to claim 1, wherein a processing device for manufacturing the plurality of existing products and a processing device for manufacturing the one prototype are the same processing device. The measurement data is waveform data,
The mass-production measurement data estimating unit uses a matching process based on the dynamic time expansion/contraction method to estimate a plurality of measurement data assumed to be obtained when the product is mass-produced using the same processing conditions and the same processing equipment as the one prototype. 5. The anomaly detection device according to any one of claims 1 to 4, wherein the estimation is performed. The anomaly detection execution unit
Based on the anomaly detection model, an anomaly degree calculation unit that calculates the anomaly degree of measurement data obtained when the product is mass-produced with the same processing conditions and the same processing equipment as the one prototype;
A determination unit that compares the degree of abnormality calculated by the degree of abnormality calculation unit with a predetermined threshold value to determine the presence or absence of an abnormality;
The abnormality detection device according to any one of claims 1 to 6, characterized by comprising: There are multiple measurement parameters,
A first regression model generation unit that generates a regression model representing the relationship between the measurement data obtained when mass-producing the product with the same processing conditions and the same processing equipment as the one prototype, and the degree of abnormality;
A first contribution degree calculation unit that calculates the degree of contribution of each measurement parameter to the degree of abnormality based on the regression model generated by the first regression model generation unit;
a first output unit that presents the contribution calculated by the first contribution calculation unit;
The abnormality detection device according to claim 7, characterized by comprising: The abnormality detection device according to any one of claims 1 to 8;
a process quality contribution calculation device that calculates the contribution of each processing process to quality based on quality data obtained by inspecting products processed by a plurality of processing processes included in the production line;
An anomaly detection system comprising
a second regression model generation unit that generates a regression model representing the relationship between measurement data obtained for each processing step and the quality data;
a second contribution calculation unit that calculates the contribution of each processing step to the quality based on the regression model generated by the second regression model generation unit;
a second output unit that presents the contribution calculated by the second contribution calculation unit;
An anomaly detection system comprising: An anomaly detection method for detecting an anomaly based on measurement data measured for a measurement parameter related to a state during processing of a processing device included in a production line,
Based on multiple measurement data for multiple existing products and measurement data for one prototype, multiple measurements assumed to be obtained when the product is mass-produced using the same processing conditions and the same processing equipment as the one prototype a mass-production measurement data estimation step of estimating measurement data;
an anomaly detection model generation step for generating an anomaly detection model based on a plurality of measurement data estimated by the mass production measurement data estimation step;
An anomaly detection execution step of detecting an anomaly of the processing device based on the measurement data obtained when the product is mass-produced using the same processing conditions and the same processing device as the one prototype, and the anomaly detection model;
An anomaly detection method, comprising:

Images