Rust prevention performance analysis method, rust prevention performance analysis system, rust prevention performance management program, and computer-readable storage medium storing the rust prevention performance management program

Abstract

To highly accurately and easily determine a manufacturing process that is considered to contribute to the quality of rust prevention performance of a metal product, for the metal product manufactured through a plurality of manufacturing processes.SOLUTION: A rust prevention performance analysis method includes processes of: reading, from an SSD9, two or more types of machine learning models 69 generated in advance so as to associate each of a plurality of feature amounts P with one or more of a plurality of manufacturing processes Q; reading, from the plurality of feature amounts P, a specific feature amount Ps selected in advance; and estimating a specific process Qs estimated to contribute to rust prevention performance determination for each type of machine learning model 69 on the basis of the specific feature amount Ps and the two or more types of machine learning models 69.SELECTED DRAWING: Figure 15A

Description

[Technical Field]

[0001] The present disclosure relates to a rust prevention performance management method, a rust prevention performance management system, a rust prevention performance management program, and a computer-readable storage medium storing the rust prevention performance management program. [Background technology]

[0002] For example, Patent Document 1 discloses a manufacturing process for a product that includes a characteristic adjustment step for providing characteristic control processing conditions and a characteristic inspection step that is reached after the specific adjustment step and via at least one other step. Here, the characteristic adjustment step is provided midway through multiple steps.

[0003] According to the description in Patent Document 1, in the characteristic adjustment process, a pre-generated learning model is used to search for optimal characteristic control processing conditions from intermediate characteristics obtained in a process prior to the characteristic adjustment process.

[0004] Furthermore, the learning model described in Patent Document 1 takes intermediate characteristics and characteristic control processing conditions as inputs and outputs product characteristics obtained in the characteristic inspection process. By using such a learning model, it is possible to automatically and appropriately search for characteristic control processing conditions for obtaining desired product characteristics from the intermediate characteristics obtained each time.

[0005] On the other hand, Patent Document 2 discloses that a first discrimination step and a classification step are performed for multiple product data. The first discrimination step is a process of discriminating between good and defective products based on a first learning model with training data. The classification step is a process of grouping product data that have been discriminated as defective based on cluster analysis without training data.

[0006] Furthermore, Patent Document 2 discloses that a second discrimination step is performed after the classification step. The second discrimination step is a process of discriminating between good and defective products for multiple product data and discriminating the type of defect for each defective product based on a second learning model.

[0007] Here, the second learning model is generated by performing machine learning using, as training data, the non-defective product data discriminated in the first discrimination step and the defective product data grouped by defect type in the classification step.

[0008] According to Patent Document 2, by successively performing classification using a first learning model and cluster analysis, it becomes possible to distinguish between good and defective products, as well as to distinguish the type of defect in each defective product. Furthermore, by generating a second learning model using the classification results from the first learning model and cluster analysis as training data, it becomes possible to use the second learning model to perform two types of discrimination. [Prior art documents] [Patent documents]

[0009] [Patent Document 1] Japanese Patent Application Laid-Open No. 2002-287803 [Patent Document 2] Japanese Patent Application Publication No. 2019-204232 Summary of the Invention [Problem to be solved by the invention]

[0010] The inventors of the present application have sought a method for managing the rust prevention performance of metal products that are manufactured through multiple manufacturing processes. One possible method for managing such performance would be to start with a machine learning technique, as disclosed in Patent Documents 1 and 2.

[0011] However, the method described in Patent Document 1 can only be used in cases where a specific property adjustment process is assumed. In more general cases, there is a possibility that the desired product properties will not be achieved simply by changing the property control processing conditions in the specific property adjustment process.

[0012] When faced with such a problem, product manufacturers must consider which of the multiple processes should be adjusted. However, Patent Document 1 does not disclose or suggest such considerations.

[0013] On the other hand, Patent Document 2 discloses that the type of defect in each defective product is determined based on a machine learning model. However, even if the type of defect is determined, the process that caused the defect is not necessarily uniquely determined.

[0014] For example, in the case of a product that is manufactured through a pre-process of painting the material and a post-process of drying the painted material, even if a defect or a sign of a defect is found in the rust prevention performance of the product, it is not easy to determine whether the defect occurred in the pre-process or the post-process. Even if it were possible to determine this by inspecting such a product, it would still be inconvenient because it would require skill to make the determination.

[0015] The present disclosure has been made in consideration of these points, and its purpose is to make it possible to easily and accurately determine the manufacturing process that is thought to have contributed to the quality of the rust prevention performance of a metal product that is manufactured through multiple manufacturing processes. [Means for solving the problem]

[0016] A first aspect of the present disclosure relates to a method for analyzing rust-preventive performance, which uses a computer including a storage unit and a calculation unit to analyze the rust-preventive performance of a metal product manufactured through multiple manufacturing processes.

[0017] According to the first aspect, when one or more of the plurality of feature quantities used to determine the rust-preventive performance are designated as specific feature quantities, and one or more of the plurality of manufacturing processes are designated as specific processes that are designated as specific processes, the rust-preventive performance analysis method includes the steps of: applying a voltage while a corrosion factor is in contact with the surface of the metal product, thereby acquiring a change over time in the current caused by the voltage; determining and outputting the rust-preventive performance based on the plurality of feature quantities that characterize the waveform of the change over time; reading from the memory unit two or more machine learning models that have been generated in advance to associate each of the plurality of feature quantities with one or more of the plurality of manufacturing processes; reading the specific feature quantity from the plurality of feature quantities; and estimating the specific process for each type of machine learning model based on the specific feature quantity and the two or more machine learning models.

[0018] According to the first aspect, the rust prevention analysis method (hereinafter also simply referred to as "analysis method") uses two or more machine learning models that associate each of a plurality of feature quantities with one or more of a plurality of manufacturing processes.

[0019] Once the feature values ​​estimated to have contributed to the deterioration of the rust prevention performance, such as defects on the surface of the metal product, are determined, the manufacturing process that caused the defect can be estimated using a machine learning model. Estimation using a machine learning model can be performed easily and with high accuracy, even by non-expert workers.

[0020] Furthermore, in the case of metal products made up of many parts, it is believed that there are countless feature values ​​that can be used for analysis. In such cases, even for an experienced worker, it is not easy to identify the process that is responsible for quality issues. However, estimation using machine learning models can be performed easily and with high accuracy, even when there are many feature values.

[0021] Furthermore, rather than simply using a machine learning model, the manufacturing process is estimated using each of two or more machine learning models, which allows for more multifaceted analysis and higher accuracy. Performance analysis method.

[0022] Furthermore, according to a second aspect of the present disclosure, the two or more types of machine learning models may include a first machine learning model based on a decision tree algorithm and a second machine learning model based on a non-decision tree algorithm.

[0023] According to the second aspect, by using models constructed from different viewpoints, a more multifaceted analysis can be performed, thereby achieving a more accurate analysis.

[0024] Furthermore, according to a third aspect of the present disclosure, the calculation unit may determine the identified process based on the importance of each of the plurality of manufacturing processes corresponding to each of the plurality of feature quantities in the first machine learning model.

[0025] According to the third aspect, the importance of each manufacturing process can be calculated automatically depending on the machine learning model selected. For example, in the case of a random forest model, the importance specific to the model can be calculated. In this way, the user's input into the determination of the specific process can be reduced. This allows for more accurate analysis.

[0026] Furthermore, according to a fourth aspect of the present disclosure, the second machine learning model may be a Bayesian network in which each of the plurality of feature quantities is a child node and each of the plurality of manufacturing processes is a parent node, and the calculation unit may determine, when the Bayesian network is visualized as a directed graph structure, a manufacturing process that is connected to the specific feature quantity via one edge as the specific process.

[0027] According to the fourth aspect, the manufacturing process is determined based on the probabilistic connections (dependencies). Since a more multifaceted analysis is performed, a more accurate analysis can be realized.

[0028] Furthermore, according to a fifth aspect of the present disclosure, when the determination result based on the first machine learning model differs from the determination result based on the second machine learning model, the calculation unit may notify the estimation results of each of the first and second machine learning models and the number of samples used in generating each of the first and second machine learning models.

[0029] According to the fifth aspect, the estimation results when using each model are notified to the user without overwriting or discarding the estimation results of one model with the other. At that time, by notifying the user of the number of samples, it becomes possible to perform a more comprehensive analysis, such as an analysis of the learning status of each model.

[0030] Furthermore, according to a sixth aspect of the present disclosure, the quality judgment result may be quantified as quality data that increases or decreases depending on whether the quality is good or bad, the plurality of manufacturing processes may each be quantified to characterize the content of each manufacturing process, the memory unit may pre-store a first correlation indicating a correlation between each of the plurality of feature amounts and the quality data, and a second correlation indicating a correlation between each of the plurality of manufacturing processes and each of the plurality of feature amounts, and after determining the specified process, the calculation unit may notify the breakdown of the specified process and a control mode of the specified process to improve the quality based on the first and second correlations.

[0031] According to the sixth aspect, it is possible not only to determine the specific process but also to propose a quality improvement plan, which makes it possible for even an unskilled person to easily improve the quality of the product.

[0032] Furthermore, according to a seventh aspect of the present disclosure, the calculation unit may display, on a display unit capable of displaying the calculation results of the calculation unit, a breakdown of the specific processes and control modes of the specific processes for improving the quality, in order starting from the specific processes that contribute to the quality.

[0033] According to the seventh aspect, when proposing a quality improvement plan, it is possible to provide a highly visible display, which makes it possible to easily improve the quality of products.

[0034] Furthermore, according to an eighth aspect of the present disclosure, the calculation unit may read a performance assessment model generated by prior machine learning so as to associate the plurality of feature quantities with the degree of quality pass / fail, and determine the degree of quality pass / fail based on the plurality of feature quantities and the performance assessment model, and the calculation unit may also determine the specific feature quantity based on the importance of each of the plurality of feature quantities corresponding to the degree of quality pass / fail in the performance assessment model.

[0035] According to the eighth aspect, a specific feature is determined by a performance evaluation model, separate from a machine learning model for determining a manufacturing process. This reduces the need for user input when analyzing a manufacturing process, thereby enabling more accurate analysis.

[0036] A ninth aspect of the present disclosure relates to a rust prevention performance analysis system that includes a computer having a storage unit and a calculation unit, and analyzes the rust prevention performance of a product manufactured through a plurality of manufacturing processes.

[0037] According to the ninth aspect, when one or more of the plurality of feature quantities used to determine the rust-preventive performance are defined as specific feature quantities, and one or more of the plurality of manufacturing processes are defined as specific processes that are defined as specific manufacturing processes, the rust-preventive performance analysis system includes: a measuring device that applies a voltage while a corrosion factor is in contact with the surface of the metal product, thereby acquiring a change over time in a current caused by the voltage; a performance evaluation means that determines and outputs the rust-preventive performance based on the plurality of feature quantities that characterize the waveform of the change over time; a model reading means that reads from the memory unit two or more types of machine learning models that have been generated in advance so as to associate each of the plurality of feature quantities with one or more of the plurality of manufacturing processes; and a specific process estimation means that reads the specific feature quantities from the plurality of feature quantities and estimates the specific process for each type of machine learning model based on the specific feature quantities and the two or more machine learning models.

[0038] A tenth aspect of the present disclosure relates to a rust prevention performance analysis program that is executed by a computer having a storage unit and a calculation unit and analyzes the rust prevention performance of a product manufactured through multiple manufacturing processes.

[0039] According to the tenth aspect, when one or more of the plurality of feature quantities used to determine the rust-preventive performance are designated as specific feature quantities, and one or more of the plurality of manufacturing processes are designated as specific processes that are designated as specific processes, the rust-preventive performance analysis program causes the computer to execute the following steps: a step of having a measuring device apply a voltage while a corrosion factor is in contact with the surface of the metal product, thereby acquiring a change over time in a current caused by the voltage; a step of the calculation unit determining and outputting the rust-preventive performance based on the plurality of feature quantities that characterize the waveform of the change over time; a step of the calculation unit reading from the memory unit two or more machine learning models that have been generated in advance so as to associate each of the plurality of feature quantities with one or more of the plurality of manufacturing processes; and a step of the calculation unit reading the specific feature quantities from the plurality of feature quantities and estimating the specific process for each type of machine learning model based on the specific feature quantities and the two or more machine learning models.

[0040] An eleventh aspect of the present disclosure relates to a computer-readable storage medium that stores the rust prevention performance analysis program. [Effects of the Invention]

[0041] As described above, according to the present disclosure, it is possible to accurately and easily determine the manufacturing process that is thought to have contributed to the quality of the rust prevention performance of a metal product that is manufactured through multiple manufacturing processes. [Brief explanation of the drawings]

[0042] [Figure 1] FIG. 1 is a system diagram illustrating the configuration of a manufacturing management system. [Figure 2] FIG. 2 is a diagram illustrating the configuration of a measurement device in a manufacturing system. [Figure 3] FIG. 3 is a diagram illustrating an example of a hardware configuration of the analysis device. [Figure 4]FIG. 4 is a diagram illustrating an example of the software configuration of the analysis device. [Figure 5] FIG. 5 is a flowchart illustrating the procedure of the analysis method. [Figure 6] FIG. 6 is a diagram illustrating an example of a change in current over time and a feature quantity obtained from the waveform. [Figure 7] FIG. 7 is a flowchart illustrating the measurement process. [Figure 8] FIG. 8 is a flow chart illustrating the measurement data analysis process. [Figure 9] FIG. 9 is a flow chart illustrating the performance evaluation process. [Figure 10] FIG. 10 is a flowchart illustrating the feature selection process. [Figure 11] FIG. 11 is a flowchart illustrating a specific process estimation process. [Figure 12] FIG. 12 is a flow chart illustrating the correlation analysis process. [Figure 13] FIG. 13 is a diagram for explaining the performance determination model. [Figure 14] FIG. 14 is a diagram illustrating the third machine learning model. [Figure 15A] FIG. 15A is a diagram illustrating the first machine learning model. [Figure 15B] FIG. 15B is a diagram for explaining the specific feature amount and the specifying step. [Figure 16A] FIG. 16A is a diagram illustrating the second machine learning model. [Figure 16B] FIG. 16B is a diagram for explaining the specific feature amount and the specifying step. [Figure 17] FIG. 17 is a diagram illustrating the first correlation coefficient list. [Figure 18] FIG. 18 is a diagram illustrating the second correlation coefficient list. [Figure 19] FIG. 19 is a diagram illustrating an example of a display screen on a display. [Figure 20]FIG. 20 is a diagram illustrating an example of a display screen on a display. DETAILED DESCRIPTION OF THE INVENTION

[0043] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. Note that the following description is for illustrative purposes only.

[0044] <1. System configuration> FIG. 1 is a system diagram illustrating the configuration of a manufacturing management system S according to the present disclosure. This manufacturing management system 100 is composed of one or more systems. For example, in this embodiment, the manufacturing management system 100 includes an analysis system 101 and a manufacturing system 102. The manufacturing management system 100 is an example of the "rust prevention performance analysis system" in this embodiment.

[0045] The manufacturing system 102 includes a manufacturing device 121, a measuring device 122, and an information processing device 123. The manufacturing device 121 manufactures a product W by executing a plurality of manufacturing processes Q. The measuring device 122 inspects the product W manufactured through the plurality of manufacturing processes Q, thereby acquiring one or more feature quantities P that characterize the quality of the product W.

[0046] The analysis system 101 is configured with a computer that functions as an analysis device 1. The analysis system 101 analyzes a manufacturing process Q in a manufacturing system 102 by using the computer 1 and the quality determination result of the product W. In this embodiment, the determination of the quality of the product W is performed by the computer 1 of the analysis system 101 based on the feature P acquired by the manufacturing system 102.

[0047] Note that the inspection of the product W (in other words, the process of acquiring one or more feature quantities P) may be performed by the analysis system 101 rather than the manufacturing system 102. Also, the quality of the product W may be determined by the manufacturing system 102 rather than the analysis system 101.

[0048] In this embodiment, the product W is a metal product. The product W as a metal product includes steel materials used in various parts that constitute a vehicle. In relation to the fact that the product W is a metal product, the analysis system 101 according to this embodiment determines the rust prevention performance of the metal product as a quality factor. In this case, the metal product 201 may have an insulating layer 203 on its surface. The rust prevention performance of the metal product 201 will be described below.

[0049] In this case, the manufacturing management system 100 can be regarded as a rust prevention performance management system that manages the rust prevention performance of metal products manufactured through multiple manufacturing processes Q. Note that even when a metal product is used for the product W, it is not essential that the product W be provided with the insulating layer 203.

[0050] Also, if the product W is a metal product, the multiple manufacturing processes Q performed by the manufacturing equipment 121 of the manufacturing system 102 may include one or more processes related to the formation of the insulating layer 203, such as an electroplating process, a water washing process, a drying process, etc.

[0051] For the sake of comprehensive discussion, the multiple manufacturing processes Q are assumed to be M processes in total (M≧2). Accordingly, for example, the Mth process Q is referred to as the “Mth process Q.” M " is sometimes called.

[0052] Below, we will first explain an example of the configuration of the measuring device 122 when a metal product 201 having an insulating layer 203 on its surface is used as the product W. Then, we will explain a method for analyzing the results obtained by the measuring device 122 through an explanation of the computer 1 in the analysis system 101.

[0053] <2. Example of measurement device configuration> 2 is a diagram showing an example of the configuration of the measuring device 122 according to this embodiment. As described above, in this embodiment, the product W inspected by this measuring device 122 is a metal product 201. The metal product 201 has a base material 202 such as a steel plate, and an insulating layer 203 located on the surface of the base material 202.

[0054] Although not shown in the drawings, a chemical conversion coating may be formed on the surface of the substrate 202, and an insulating coating may be provided on the surface of the chemical conversion coating. In this case, the insulating layer 203 as in this embodiment is formed by the insulating coating.

[0055] The measuring device 122 is configured to apply a voltage to the surface of the metal product 201 as the product W, and acquire the change over time in the current caused by the voltage.

[0056] Specifically, the measuring device 122 applies a voltage to the surface of the metal product 201 while the corrosion factor 205 is in contact with the surface of the metal product 201. The corrosion factor 205 may be an electrolyte material containing a supporting electrolyte such as water or sodium chloride, and a clay mineral such as kaolinite.

[0057] Specifically, the measuring device 122 includes a container 220 , an electrode 221 , a power source 222 , a sealing material 223 , and wiring 224 .

[0058] Of these elements, the container 220 is placed on the surface of the metal product 201 (for example, the surface of the insulating layer 203) via a sealing material 223 to prevent liquid leakage. The corrosion factor 205 is contained in the container 220 and is in contact with the surface of the insulating layer 203. The shape and material of the container 220 are not particularly limited. The container 220 has a tubular shape and is made of a resin material such as acrylic resin or epoxy resin. The term "tubular shape" used here includes tubular shapes with any cross-sectional shape, such as a cylindrical shape or a polygonal cylindrical shape.

[0059] The sealing material 223 is a sheet-like sealing material made of, for example, silicone resin. When the container 220 is placed on the metal product 201, the sealing material 223 improves the adhesion between the container 220 and the insulating layer 203 and fills the gap between them. This effectively prevents the corrosion factor 205 from leaking from between the container 220 and the insulating layer 203.

[0060] The electrode 221 is used to apply a voltage between the substrate 202 and the insulating layer 203. The electrode 221 is configured so that at least its tip is embedded in the corrosion factor 205 in the container 220 and comes into contact with the corrosion factor 205. The electrode 221 may be, for example, an electrode that can be used for electrochemical measurement. The electrode 221 may be, for example, a carbon electrode or a platinum electrode.

[0061] The power supply 222 is connected to the electrode 221 and the substrate 202 via wiring 224, and applies a voltage between the electrode 221 and the substrate 202. At the same time, the power supply 222 measures the change over time in the current flowing between the electrode 221 and the substrate 202 as the voltage is applied. The application of the voltage and the measurement of the current by the power supply 222 are controlled by the information processing device 123.

[0062] Data measured by the power supply 222 (measurement data 49) is output from the power supply 222 to the information processing device 123. The information processing device 123 transmits the measurement data 49 to the analysis system 101 by wireless or wired communication.

[0063] The measurement data may be data in which the detected current value is plotted against time, or in the case of applying a gradually increasing voltage, data in which the detected current value is plotted against the applied voltage value. In addition to the detected current value, applied voltage value, and measurement time, the measurement data also includes information identifying the production lot of the metal product 201 on which the measurement was performed and the measurement position on the metal product 201.

[0064] 2 is merely an example of analyzing the rust prevention performance of a metal product (product W) having an insulating layer 203. The configuration of the measuring device 122 may vary depending on the type of product W and the type of quality to be analyzed.

[0065] Furthermore, the present disclosure does not necessarily require the measurement device 122. For example, the results of a craftsman's quality assessment may be input to the analysis system 101 together with the measurement data used by the craftsman during the inspection.

[0066] <3. Analysis System> Fig. 3 is a diagram illustrating an example of the hardware configuration of an analysis device (computer 1) according to the present disclosure, and Fig. 4 is a diagram illustrating an example of the software configuration of analysis device 1. This computer 1 is configured by a computer including a CPU 3 as a calculation unit, and a RAM 7 and an SSD 9 as storage units.

[0067] 3, the computer 1 includes a central processing unit (CPU) 3 that controls the entire analysis device 1, a read only memory (ROM) 5 that stores a boot program and the like, a random access memory (RAM) 7 that functions as a main memory, and a solid state drive (SSD) 9 that serves as a secondary storage device. Note that a hard disk drive (HDD) or the like can also be used as the secondary storage device instead of the SSD 9.

[0068] Of these elements, the CPU 3 executes various programs. The CPU 3 functions as a calculation unit in this embodiment. The RAM 7 and SSD 9 temporarily or continuously store the programs executed by the CPU 3. The RAM 7 and SSD 9 each function as a storage unit in this embodiment.

[0069] The analysis device 1 also includes a display 11, a graphics memory (Video RAM: VRAM) 13 that stores image data to be displayed on the display 11, and a keyboard 15 and a mouse 17 as man-machine interfaces. The keyboard 15 and the mouse 17 function as a reception unit that receives input from an operator. The display 11 functions as a display unit that displays a screen based on the results of calculations by the CPU 3. The computer 1 according to this embodiment can also send and receive data to and from external devices via a communication interface 21.

[0070] As shown in FIG. 4, the program memory of SSD 9 stores an operating system (OS) 19, a measurement data analysis program 291, a performance evaluation program 292, a feature selection program 293, a model reading program 294, a specific process estimation program 295, a correlation analysis program 296, and an application program 39.

[0071] The analysis method according to this embodiment is executed using the computer (analysis device 1) configured as described above and the quality judgment results of the product W, and analyzes the manufacturing process Q.

[0072] A measurement data analysis program 291, a performance evaluation program 292, a feature selection program 293, a model reading program 294, a specific process estimation program 295, and a correlation analysis program 296, which are coded to realize such analysis, constitute the analysis program 29 in this embodiment.

[0073] Here, the analysis program 29 is a program for executing the analysis method according to this embodiment, and is configured to cause the analysis device (computer 1) to execute each step constituting the analysis method. The analysis program 29 is pre-stored in a computer-readable storage medium 18. This storage medium 18 is a tangible storage medium such as a disk medium.

[0074] In the program memory of the SSD 9, each program constituting the analysis program 29 is started in response to a command input from the keyboard 15, mouse 17, etc. At that time, each program is loaded from the SSD 9 into the RAM 7 and executed by the CPU 3.

[0075] Meanwhile, measurement data 49 to be analyzed is stored in the data memory of SSD 9. The measurement data 49 is data measured by the measurement device 122 described above.

[0076] The data memory of the SSD 9 also stores one or more performance determination models 59 and two or more machine learning models 69. The performance determination models 59 and the machine learning models 69 are each machine learning models generated in advance. Details of these machine learning models will be described later.

[0077] The data memory of SSD 9 also stores first and second correlation coefficient lists 791 and 792, which are used to give specific suggestions for changes and adjustments to the manufacturing process Q. Furthermore, the data memory of SSD 9 also stores quality data 89 generated by a performance evaluation program 292. These will be described in detail later.

[0078] In addition, various data generated by each program constituting the analysis program 29 and the execution results of the application program 39 are stored in the data memory of the SSD 9 or in the RAM 7 as the main memory, as necessary.

[0079] <4. Overview of analysis method> Fig. 5 is a flowchart illustrating the procedure of the analysis method. As shown in Fig. 5, the analysis method is implemented by sequentially executing a measurement data analysis process (step S1), a performance evaluation process (step S2), a feature selection process (step S3), a model loading process (step S4), a specific process estimation process (step S5), and a correlation analysis process (step S6).

[0080] The analysis program 29 is configured to cause the computer 1 to execute these processes. That is, among these processes, the measurement data analysis process is carried out by the CPU 3 executing the measurement data analysis program 291 described above.

[0081] Similarly, the performance evaluation process is performed by CPU3 executing a performance evaluation program 292. The feature selection process is performed by CPU3 executing a feature selection program 293. The model loading process is performed by CPU3 executing a model loading program 294. The specific process estimation process is performed by CPU3 executing a specific process estimation program 295. The correlation analysis process is performed by CPU3 executing a correlation analysis program 296.

[0082] When the CPU 3 executes the specific process estimation program 295 etc., an analysis device is configured by the computer 1. That is, the computer 1 functions as an analysis device including a measurement data analysis means that executes a measurement data analysis process, a performance evaluation means that executes a performance evaluation process, a feature selection process, a model reading means that executes a model reading process, a specific process estimation means that executes a specific process estimation process, and a correlation analysis means that executes a correlation analysis process.

[0083] Below, the process (measurement process) performed prior to the analysis of the product 1 will be described with reference to FIGS. 6 and 7, and then the analysis method according to the present disclosure will be described in detail with reference to FIG. 5 again.

[0084] The measurement processing and analysis method described below are performed periodically, for example, for each production lot of product W. A production lot generally refers to a unit of production or order. In this embodiment, a production lot can be a unit separated by, for example, a shipping unit, a production unit, a production month / date, a raw material (e.g., paint) lot, a raw material change point, or a process change point (e.g., when replacing the cutting blade of the metal product 201). It is not necessary to perform the processing for each production lot. The measurement processing may be performed for each product one by one.

[0085] <5. Details of measurement process> Fig. 6 is a diagram illustrating an example of a change in current over time and a feature quantity obtained from the waveform. Fig. 7 is a flowchart illustrating a measurement process according to the present disclosure. First, as shown in step S101 of Fig. 7, the measurement device 122 extracts one or more products W for each production lot. The number of products extracted is adjusted appropriately based on the size of the production lot, the expected probability of anomaly occurrence, etc.

[0086] In the following step S102, the measurement device 122 measures the measurement data 49 to be analyzed by the measurement device 122. This measurement is performed on each product W extracted in step S101.

[0087] Specifically, in step S102, the measuring device 122 applies a voltage between the insulating layer 203 and the substrate 202, for example, with the corrosion factor 205 in contact with the surface of the insulating layer 203. The measuring device 122 measures the change over time in the current caused by the applied voltage. The positions where the measurement is performed include the main surfaces, edges, and welds of the metal product 201.

[0088] In the following step S103, the measurement device 122 converts the changes over time measured in step S103 into measurement data 49 and transmits it to the computer 1 of the analysis system 101. The transmitted measurement data 49 is stored in the RAM 7 or SSD 9 of the computer 1.

[0089] The following explanation will basically focus on the case where "measurement data 49 = change in current over time," but it can be replaced with other data as described above as appropriate.

[0090] <6. Details of analysis processing> Next, the analysis process will be described in detail. The following process is executed for each product W extracted for each production lot.

[0091] (6-1. Measurement data analysis process) Fig. 8 is a flowchart showing the details of the measurement data analysis process. When the control process proceeds to step S1 in Fig. 5, the CPU 3 executes the flow shown in Fig. 8 in order from step S201.

[0092] 8, the CPU 3 reads the measurement data 49 from the RAM 7, the SSD 9, etc. In the following step S202, the CPU 3 extracts a plurality of types of feature amounts P from the read measurement data 49.

[0093] Specifically, the CPU 3 acquires, as the plurality of feature quantities P, a plurality of parameters characterizing the change over time of the current (current resulting from the applied voltage) indicated by the measurement data 49. Here, in this embodiment, "change over time of the current" refers to "change over time of the current measured as a waveform." In other words, it refers to "current values ​​corresponding to voltage values ​​changed over time."

[0094] More specifically, the CPU 3 extracts a group of three or more parameters including the number Np of current peaks, the height Ip of each peak, and the gradient S of the current as the plurality of types of feature quantities P (see FIG. 6).

[0095] For example, if there is no abnormality or sign of an abnormality in the insulating layer 203 of the metal product 201, when a DC voltage that increases (gradually increases) over time is applied as described above, almost no current flows until the voltage (applied voltage) reaches the breakdown voltage, and once the applied voltage reaches the breakdown voltage, the current increases rapidly.

[0096] The insulating layer 203 maintains its blocking performance against corrosion factors 205 until the applied voltage reaches the breakdown voltage. As a result, almost no current flows. On the other hand, when the applied voltage reaches the breakdown voltage, the corrosion factors 205 are encouraged to penetrate the insulating layer 203, and the corrosion factors 205 may reach the surface of the substrate 202 at the weakest points of the insulating layer 203, such as points with relatively few cross-linked resin structures. As a result, the current may increase rapidly. In other words, a sudden increase in the detected current value indicates that the corrosion factors 205 have reached the surface of the substrate 202, causing the insulating layer 203 to lose its rust-preventing performance.

[0097] On the other hand, consider a case where there is a local defect in the insulating layer 203 (for example, foreign matter such as gas pins, welding spatter and slag, burrs, iron powder, or unevenness on the surface of the substrate 202), causing a localized area where the effective film thickness is small in the insulating layer 203. In this case, when a gradually increasing DC voltage is applied between the electrode 204 and the substrate 202, the current generated between them is considered to change over time as shown in FIG.

[0098] That is, if a local defect exists in the insulating layer 203, the corrosion factor 5 penetrates locally at the defect location. When the corrosion factor 205 penetrates the insulating layer 203 at a certain defect location and reaches the substrate 202, conduction occurs, causing an instantaneous increase in the current value. If a voltage equal to or greater than the voltage at which water electrolysis occurs is applied between the electrode and the substrate 202 at this point, electrochemical reactions such as water electrolysis proceed on the surface of the substrate 202 due to the conduction. As a result, the generated gas and electrolytic products accumulate in the defect, interrupting the conduction and causing a decrease in the current value. That is, if a local defect exists, the conduction and subsequent interruption at the defect location cause peaks in the waveform of the time-dependent change data of the detected current value, indicating an instantaneous increase and decrease in the current value.

[0099] When multiple defects exist, a number Np of peaks corresponding to the number of defects occurs, as shown in Figure 6. The current value at the peak, i.e., the peak height Ip, is related to the size of the conduction path through which the current flows, i.e., the size, type, and conductivity of the defect. Furthermore, since conduction occurs at a lower applied voltage in areas where the thickness of insulating layer 203 is thinner due to defects, the applied voltage that gives the peak is correlated with the film thickness of the area where the defect exists.

[0100] Furthermore, if the insulating layer 203 has a general defect (for example, a low cross-link density due to a lack of catalyst in the paint, or an oxide film on the surface of the substrate 202), corrosion factors 205 will penetrate the entire insulating layer 203 and reach the substrate 202. As a result, a gradual increase in the current value occurs in the waveform of the time-dependent change data of the detected current value. In other words, the slope S of the waveform of the time-dependent change data when the detected current value increases is related to the insulating quality (film quality) of the insulating layer 203.

[0101] The number, size, type, conductivity, film thickness, and film quality of the above defects are all factors that affect the rust-preventive performance of the insulating layer 203. Therefore, in this embodiment, the number Np of peaks in the current value, the height Ip of the peaks, and the slope S of the waveform of the change in current over time, which are related to these factors, are acquired as feature quantities P that characterize the rust-preventive performance of the insulating layer 203. Furthermore, since the likelihood of peaks occurring in the current varies depending on the position (including the main surface, edge, and weld of the metal product 201) at which the change in current over time is measured, position information indicating the position at which the change in current over time is measured may also be acquired from the measurement data as a parameter for evaluating the rust-preventive performance of the insulating layer 203.

[0102] Note that the parameters that can be used as the feature quantity P are not limited to the number of current peaks Np, the height of each peak Ip, and the current slope S. Any parameter related to the rust prevention performance of the metal product W can be used.

[0103] Additionally, as described above, when image data, audio data, text data, and mechanical or electrical data are used as the measurement data 49, it is only necessary to extract a feature amount according to the type of data.

[0104] For comprehensive discussion, the plurality of feature quantities P are assumed to be N in total (N≧2). Accordingly, the Nth feature quantity P is referred to as the “Nth feature quantity P N In the case of the metal product 201, the first feature amount P1 corresponds to the number of peaks Np, the second feature amount P2 corresponds to the peak height Ip, and the third feature amount P3 corresponds to the slope S (see FIG. 6).

[0105] Thereafter, in step S203, the CPU 3 converts each of the plurality of feature quantities P into a dimensionless risk score between 0 and 1, and stores the risk score in the RAM 7, the SSD 9, or the like.

[0106] In this embodiment, the risk score indicates the level of risk of abnormalities occurring in the quality (rust prevention performance) of the metal product 201. In other words, it can be interpreted that the higher the risk score, the more likely abnormalities will occur in the metal product 201.

[0107] The inventors of the present application attempted to determine whether the quality of the metal product 201 is good or bad based on the level of the risk score corresponding to each of the multiple feature quantities P and the level of the average value of the multiple feature quantities P. Furthermore, rather than simply dividing the quality of the metal product 201 into two categories, either "normal" or "abnormal," they attempted to achieve three or more classifications, including a gray area (an intermediate state) between white (normal) and black (abnormal), such as "there is a sign of an abnormality occurring."

[0108] However, such a judgment is not easy. Therefore, we decided to use a pre-generated machine learning model in the performance evaluation process that follows the measurement data analysis process. By using a machine learning model to make judgments, even those who are not skilled craftsmen can make judgments with a certain degree of accuracy.

[0109] For the sake of simplicity, the following description will be given assuming that each risk score is the same as the corresponding feature value P.

[0110] (6-2. Performance evaluation process) Fig. 9 is a flowchart showing the details of the performance evaluation process. When the control process proceeds to step S2 in Fig. 5, the CPU 3 executes the flow shown in Fig. 9 in order from step S301.

[0111] In the performance evaluation process, the CPU 3 determines and outputs the degree of quality (corrosion prevention performance) of the metal product 201 based on the multiple feature quantities P (more precisely, the risk scores corresponding to each feature quantity P) acquired in the measurement data analysis process. The term "degree of quality" here includes not only the two states of "normal" and "abnormal," but also the intermediate state of "there are signs of abnormality" as described above.

[0112] The quality level (quality) is quantified so that it increases or decreases depending on the quality. For example, in this embodiment, the quality exemplified as rust prevention performance is quantified as, for example, "normal = 2," "sign of abnormality present = 1," and "abnormal = 0." Hereinafter, the quantified quality level is also referred to as "quality data 89."

[0113] Specifically, first, in step S301, the CPU 3 reads a plurality of feature quantities P that have been scored as risk scores. In the specific example described above, the feature quantities P that are read here are the number of peaks Np as the first feature quantity P1, the peak height Ip as the second feature quantity P2, and the slope S as the third feature quantity P3.

[0114] In the next step S302, the CPU 3 reads the performance determination model 59 that has been generated in advance by machine learning. As shown in Fig. 13, the performance determination model 59 is a machine learning model that associates a plurality of feature quantities P with the degree of quality of the rust prevention performance (quality data 89).

[0115] In other words, the performance determination model 59 is a machine learning model that receives a plurality of feature quantities P as input and outputs an estimated value of the degree of quality of the rust prevention performance. The performance determination model 59 is generated in advance by supervised learning.

[0116] The performance assessment model 59 is trained using first training data. This first training data is composed of, for example, a plurality of feature quantities P obtained from each of the plurality of measurement data 49, and quality data 89 indicating the degree of quality of the rust prevention performance that is determined in advance when each measurement data 49 is acquired. Pairs each consisting of a plurality of feature quantities P and the associated quality data 89 are created in advance in the same number as the number of samples of the first training data.

[0117] Specifically, the performance evaluation model 59 may be a nonlinear regression model or a linear regression model. The performance evaluation model 59 as a nonlinear regression model may be a machine learning model based on a decision tree algorithm. The performance evaluation model 59 related to the decision tree may be a Random Forest (RF) model.

[0118] In the next step S303, the CPU 3 judges the pass / fail degree of the rust prevention performance based on the plurality of feature quantities P and the performance judgment model 59. Specifically, the CPU 3 inputs the plurality of feature quantities P read in step S301, i.e., the feature quantities P obtained from the product W to be analyzed, into the trained performance judgment model 59, and causes the model 59 to output an estimation result of the pass / fail degree. When quantified as described above, this estimation result is output as quality data 89.

[0119] In the following step S304, the CPU 3 stores the output result of step S303 in the RAM 7 or SSD 9. Thereafter, the control process returns from the flow in Fig. 9 to the flow in Fig. 5, and proceeds to step S3.

[0120] (6-3. Feature selection process) Fig. 10 is a flowchart showing the details of the feature selection process. When the control process proceeds to step S2 in Fig. 5, the CPU 3 executes the flow shown in Fig. 9 in order from step S301.

[0121] Hereinafter, among the multiple feature amounts P acquired during quality assessment, one or more feature amounts P estimated to contribute to the quality assessment will be referred to as specific feature amounts Ps. In the specific examples shown in Figures 2 and 6, the specific feature amounts Ps are one or more feature amounts P estimated to have contributed to the assessment of rust prevention performance.

[0122] Specifically, in the feature selection process, the CPU 3 selects a specific feature Ps from among a plurality of feature P acquired from the measurement data 49. This selection can be made based on the importance (importance) of each of the plurality of feature P corresponding to the degree of pass / fail. In the case of the specific example described above, the CPU 3 selects which of the three parameters, the number of peaks Np as the first feature P1, the peak height Ip as the second feature P2, and the slope S as the third feature P3, contributed to the determination of the rust prevention performance, including abnormalities and their precursors.

[0123] In the example of FIG. 13, the specific feature amount Ps is composed of the number of peaks Np as the first feature amount P1 and the peak height Ip as the second feature amount P2.

[0124] This determination can be made by determining the importance of each of the multiple feature amounts P when the performance determination model 59 outputs the quality data 89, and setting one or more feature amounts P in descending order of importance as the specific feature amount Ps. When the RF model M1 is used for the performance determination model 59, the contribution of each feature amount P (so-called RF importance) can be used as the importance. The importance may be determined, for example, based on the Gini coefficient of the decision tree constituting the random forest model.

[0125] Alternatively, so-called SHAP (Shapley Additive exPlanations) processing may be used. In this case, based on a plurality of feature quantities P acquired from the measurement data 49 and the performance evaluation model 59, the performance evaluation model 59 is locally approximated around the feature quantities P. The model (approximation model) generated by approximating the performance evaluation model 59 may be, for example, a simple model that easily explains the contribution of each feature quantity P. Then, based on the approximation model, the contribution of each feature quantity P to the predicted value output from the approximation model is expressed by a so-called Shapley value used in cooperative game theory, etc. The contribution may be used as the importance for selecting a specific feature quantity Ps. In addition, various techniques in image processing for two-dimensional images, such as gradient processing, may be applied.

[0126] Specifically, in step S401, the CPU 3 determines the importance of each of the plurality of feature amounts P. In the following step S402, the CPU 3 selects one or more feature amounts P in descending order of importance as specific feature amounts Ps.

[0127] Here, the importance of each feature P may be scored as a dimensionless number between 0 and 1, and if the score is equal to or greater than a predetermined value (e.g., 0.5), it may be considered to be "highly important." Furthermore, if there are multiple feature Ps with scores thus defined that are equal to or greater than a predetermined value, only the top multiple feature Ps (e.g., three feature Ps) may be selected in descending order of importance.

[0128] Furthermore, when the quality of the product W is quantified into a dimensionless number between 0 and 2 as described above, a correlation coefficient between each feature amount P and the quality data 89 can be given in addition to the importance (see FIG. 17). In the case of this embodiment, such a correlation coefficient (hereinafter also referred to as a "first correlation coefficient") is calculated in advance and stored in the data memory of the SSD 9 as a first correlation coefficient list 791 shown in FIGS. 4 and 17.

[0129] In the following step S403, the CPU 3 stores the specific feature amount Ps selected in step S402 in the RAM 7, the SSD 9, etc. Thereafter, the control process returns from the flow in Fig. 10 to the flow in Fig. 5, and proceeds to step S4.

[0130] (6-4. Model loading process) In the model loading process, the CPU 3 loads two or more (two in this embodiment) machine learning models 691, 692 from the SSD 9 serving as a storage unit. Each of the two machine learning models 691, 692 associates each of a plurality of feature quantities P with one or a plurality of manufacturing processes Q described above. As will be described in detail later, each of the plurality of manufacturing processes Q is quantified (digitized) so as to characterize the content of each manufacturing process Q.

[0131] In this embodiment, one of the two machine learning models 69 corresponds to the first machine learning model 691 illustrated in Fig. 4. The first machine learning model 691 is generated in advance by supervised learning.

[0132] The first machine learning model 691 is trained using second training data, which is composed of numerical data corresponding to each manufacturing process Q and each of a plurality of feature quantities P, which are measured in advance for each of a large number of sample data.

[0133] Here, the first machine learning model 691 is generated using a third machine learning model (shown only in FIG. 14) 693. The third machine learning model 693 is generated by using the second training data described above as teacher data, and is a machine learning model that receives numerical data corresponding to each of a plurality of manufacturing processes Q as input and outputs one feature value P.

[0134] By generating the third machine learning model 693, it is possible to determine one or more manufacturing processes Q that contributed to the output of one feature P from among the multiple manufacturing processes Q. This selection can be made based on the importance of each of the multiple manufacturing processes Q. Specifically, the importance of each of the multiple manufacturing processes Q is determined, and one or more manufacturing processes Q are selected in descending order of importance.

[0135] Specifically, the third machine learning model 693 according to this embodiment is a model based on a decision tree algorithm. The third machine learning model related to the decision tree is, for example, an RF model. When an RF model is used, the contribution of each manufacturing process Q (so-called RF importance) can be used as the importance. The importance may be determined based on, for example, the Gini coefficient of the decision tree constituting the random forest model.

[0136] Alternatively, so-called SHAP (Shapley Additive exPlanations) processing may be used. In this case, based on numerical data quantifying each of the multiple manufacturing processes Q and a third machine learning model 69, the third machine learning model 69 is locally approximated around the numerical data corresponding to each manufacturing process Q. A model (second approximation model) generated by approximating the third machine learning model 69 may be, for example, a simple model that easily explains the contribution of the numerical data corresponding to each manufacturing process Q. Then, based on the second approximation model, the contribution of the numerical data corresponding to each manufacturing process Q to the predicted value output from the second approximation model is expressed using a so-called Shapley value, which is used in cooperative game theory, etc. The contribution may be used as the importance for selecting one or more manufacturing processes Q.

[0137] Then, similarly, the manufacturing processes Q with high importance are determined for the other feature quantities P. By performing this for all feature quantities P, it is possible to generate a first machine learning model 691 in which each of the multiple feature quantities P is associated with one or more of the multiple manufacturing processes Q, as shown in FIG.

[0138] Here, the importance of each manufacturing process Q may be scored as a dimensionless number between 0 and 1, and if the score is equal to or greater than a predetermined value (e.g., 0.5), it may be considered to be "highly important." Furthermore, if there are multiple manufacturing processes Q with scores thus defined that are equal to or greater than a predetermined value, only the top three may be selected in descending order of importance.

[0139] 15A, the first machine learning model 691 associates each feature amount P with one or more manufacturing processes Q. Each square in FIG. 15A is associated with the importance of each manufacturing process Q for each feature amount P. The first machine learning model 691 can be considered as a model that uses each feature amount P as an input and one or more manufacturing processes Q as an output.

[0140] The manufacturing process Q output from the first machine learning model 691 is a process that has a relatively high importance for the feature value P input to the model 691. In the example of FIG. 15A, the first machine learning model 691 determines the number of peaks as the feature value P between the first process Q1 and the Mth process Q M and for the peak height as the feature amount P, the Mth step QM is output.

[0141] Furthermore, when the quality of product W is quantified as a dimensionless number between 0 and 2 as described above, a correlation coefficient can be given between each manufacturing process Q and each feature quantity P, in addition to the importance (see Figure 18). Note that for manufacturing processes Q that can only be quantified as categorical data, the correlation coefficient is left undefined, as in the case of the third process Q3 in Figure 18.

[0142] In this embodiment, such correlation coefficients (hereinafter also referred to as "second correlation coefficients") are calculated in advance and stored in the data memory of the SSD 9 as a second correlation coefficient list 792 shown in FIGS.

[0143] The other of the two machine learning models 69 is a Bayesian network in which each of the plurality of feature quantities P is a child node and each of the plurality of manufacturing processes Q is a parent node. This Bayesian network can be visualized as a directed graph structure as shown in FIG. 16A, for example. This model corresponds to the second machine learning model 692 illustrated in FIG. 4. The second machine learning model 692 is generated in advance by unsupervised learning.

[0144] The second machine learning model 692 is trained using third training data. This third training data is composed of waveforms of current changes over time corresponding to each feature amount P, which are measured in advance for each of the L sample data (measurement data for learning, not for testing), and the details of each of the multiple manufacturing processes Q (quantified numerical data).

[0145] Specifically, the second machine learning model 692 may be a model based on a non-decision tree algorithm. The second machine learning model 692 related to the non-decision tree may have a graph structure determined by so-called graph structured analysis (hereinafter referred to as "GSA").

[0146] GSA is a big data analysis method proposed by the inventors of the present application that combines probability theory (Bayesian estimation) and graph theory. In this embodiment, GSA is used to determine the graph structure. However, it is not necessary to use GSA to determine the graph structure. For details of GSA, please refer to JP 2021-111063 A.

[0147] In the example of Figure 16A, N feature quantities P and M manufacturing processes Q are nodes, and the dependency between one feature quantity P and one or more manufacturing processes Q is graphed as an edge.

[0148] A probability distribution function corresponding to a Bayesian network is usually expressed by multiplying multiple conditional probabilities. For example, in FIG. 16A, the multiple conditional probabilities include a probability distribution function with the first process Q1 and the second process Q2 as conditions and the first feature P1 as a variable. The former variable is visualized as a child node, and the latter condition is visualized as a first-level parent node. This suggests that the first feature P1 is more strongly dependent on the first process Q1 and the second process Q2 than on the other manufacturing processes Q.

[0149] The graph structure of Fig. 16A can be replaced with a map such as that shown in Fig. 16B. A check mark in each square in Fig. 16B indicates that one feature (e.g., the number of peaks) P is connected to one or more manufacturing processes (e.g., the first process and the Mth process) Q via an edge. A square without a check mark indicates that no such connection exists.

[0150] When a feature quantity P is input, the second machine learning model 692 outputs one or more manufacturing processes Q that are connected via an edge to the feature quantity P. The manufacturing processes Q output from the second machine learning model 692 are processes that have a relatively strong dependency on the feature quantity P input to the second machine learning model 692.

[0151] Once the first and second machine learning models 691, 692 are loaded, the control process proceeds from step S4 to step S5 of FIG.

[0152] (6-5. Specific process estimation process) Fig. 11 is a flowchart showing the details of the specific process estimation process. When the control process proceeds to step S5 in Fig. 5, the CPU 3 executes the flow shown in Fig. 11 in order from step S501.

[0153] Hereinafter, one or more manufacturing processes Q that are estimated to contribute to the assessment of quality among the multiple manufacturing processes Q are referred to as specific processes Qs. In the specific examples shown in Figures 2 and 4, the specific processes Qs are one or more manufacturing processes Q that are estimated to have contributed to the assessment of rust-preventive performance.

[0154] First, in step S501, the CPU 3 reads a specific feature quantity Ps that has been selected in advance from among a plurality of feature quantities P in a feature quantity selection process.

[0155] In the following steps S502 to S504, the CPU 3 determines a specific process Qs from among the multiple manufacturing processes Q. This determination is made based on the specific feature amount Ps and two or more types of machine learning models 69, for each type of the machine learning models 69.

[0156] Specifically, in step S502, the CPU 3 inputs the specific feature quantity Ps to the first machine learning model 691. The first machine learning model 691 outputs a manufacturing process Q that has a high contribution to the specific feature quantity Ps, that is, a specific process Qs. For example, as shown in FIG. 15B, assume that the "number of peaks" is selected as the specific feature quantity Ps. In this case, the specific process Qs corresponds to the first and second processes in the figure.

[0157] In the following step S503, the CPU 3 inputs the specific feature quantity Ps to the second machine learning model 692. The second machine learning model 692 outputs a manufacturing process Q that has a strong dependency on the specific feature quantity Ps, that is, a specific process Qs. For example, assume that the second feature quantity P2 in FIG. 16A is selected as the specific feature quantity Ps. The specific process Qs in this case corresponds to the second process in the same figure.

[0158] In the following step S503, the CPU 3 inputs the specific feature amount Ps to the second machine learning model 692. The second machine learning model 692 outputs a manufacturing process Q having a strong dependency on the specific feature amount Ps, that is, a specific process Qs.

[0159] In the following step S504, the CPU 3 stores the specific process Qs output for each model in the RAM 7, the SSD 9, etc. Thereafter, the control process returns from the flow in Fig. 11 to the flow in Fig. 5, and proceeds to step S5.

[0160] (6-6. Correlation analysis process) Fig. 12 is a flowchart showing the details of the correlation analysis process. When the control process proceeds to step S6 in Fig. 5, the CPU 3 executes the flow shown in Fig. 12 in order from step S601.

[0161] In this correlation analysis process, the CPU 3 notifies the details of the specific process Qs and the control mode of the specific process Qs for improving quality based on the first and second correlation coefficient lists 791, 792.

[0162] Specifically, the CPU 3 displays the breakdown of the specific processes Qs and the control modes of the specific processes Qs for improving quality on the display 11 as a display unit in the order of the specific processes Qs that contribute to the quality. The processing related to the display order of the specific processes Qs is performed when the specific process Qs is composed of multiple manufacturing processes Q.

[0163] Specifically, in step S601, the CPU 3 reads one or more specific processes Qs and specific feature quantities Ps associated with each specific process Qs.

[0164] In the following step S602, the CPU 3 refers to the second correlation coefficient list 792 to read out the correlation coefficients (second correlation coefficients) of the specific feature amounts Ps corresponding to each specific process Qs.

[0165] Also in step S602, the CPU 3 selects, for each specific process Qs, specific feature quantities Ps whose absolute value of the second correlation coefficient is equal to or greater than a second predetermined value (e.g., 0.2). If there are no specific feature quantities Ps that satisfy this condition, the CPU 3 displays a message to that effect on the display 11.

[0166] In the next step S603, the CPU 3 generates a union of the specific feature amounts Ps selected for each specific process Qs, and regards the specific feature amounts Ps included in the union as the feature amounts P to be controlled.

[0167] In the following step S604, the CPU 3 reads out the correlation coefficients (first correlation coefficients) of the quality data 89 corresponding to the specific feature quantities Ps constituting the union by referring to the first correlation coefficient list 791. This reading is performed for each specific process Qs.

[0168] In the same step S604, the CPU 3 selects, from among the specific feature quantities Ps constituting the union, a specific feature quantity Ps whose absolute value of the correlation coefficient (first correlation coefficient) read immediately before is equal to or greater than a third predetermined value (e.g., 0.2). If there is no specific feature quantity Ps that satisfies this condition, the CPU 3 displays that fact on the display 11.

[0169] In the next step S605, the CPU 3 lists the specific feature amounts Ps selected in step S604 in association with the corresponding specific processes Qs. As a result, one or more specific feature amounts Ps that are strongly correlated with the quality data 89 are listed for each specific process Qs.

[0170] In the next step S606, the CPU 3 multiplies the first correlation coefficient and the second correlation coefficient for each specific feature quantity Ps listed in step S607 for each specific process Qs. Hereinafter, this multiplied value will be referred to as an "effect index."

[0171] For one specific process Qs, an effect index is calculated for each specific feature Ps associated with that process. For example, if there are two specific feature Ps associated with one specific process Qs, two effect indexes are also calculated.

[0172] The larger the effect index, the larger the change in the quality data 89 when the corresponding specific process Qs is changed. By paying attention to the positive or negative sign of the multiplied value, it is possible to determine the control mode of the specific process Qs in order to improve the quality data 89.

[0173] As described above, the effect index is calculated for each specific feature amount Ps for each specific process Qs. Therefore, in step S607, the CPU 3 adds up the effect indexes calculated for each specific feature amount Ps. A total value of the effect indexes is calculated for each specific process Qs.

[0174] Thereafter, in the following step S608, the CPU 3 displays the specific processes Qs and their control modes on the display 11 in descending order of the absolute value of the effect index.

[0175] Note that the estimation result of the specific process Qs by the first machine learning model 691 and the estimation result of the specific process Qs by the second machine learning model 692 may differ from each other.

[0176] 12, the CPU 3 individually notifies the estimation results of the first and second machine learning models 691 and 692 and the number of samples used in generating the first and second machine learning models 691 and 692. This notification may be performed, for example, by displaying them on the display 11.

[0177] Here, the details of the control mode are set in advance for each manufacturing process Q. Each manufacturing process Q may include one or more control factors, and when displaying the data in step S608, the adjustment direction of the control factors may be displayed.

[0178] As an example, suppose the first process Q1 is an electrodeposition coating process. In this case, the control factors of the first process Q1 include the electrical conductivity (μS / cm) during electrodeposition coating, the paint temperature, the voltage value during electrodeposition, the ion concentration (MEQ) of the electrodeposition solution, the acid concentration of the electrodeposition solution, the distribution of the electrodeposition solution, the amount and type of solvent in the electrodeposition solution, the presence or absence of foreign matter in the electrodeposition solution, and the takt time.

[0179] The above-mentioned numerical data quantified corresponding to each manufacturing process Q may be the control target of these control factors. When the control process Q includes multiple control factors, the numerical data obtained by quantifying each control process Q may be numerical data corresponding to each control factor.

[0180] As another example, suppose the first process Q1 is a drying process using a drying oven. In this case, the control factors of the first process Q1 include the amount of moisture in the drying oven, the set temperature of the drying oven, the baking time in the drying oven, the capacity of the drying oven, etc.

[0181] The RAM 7 serving as a storage unit stores in advance the association between increasing or decreasing each of the control factors described above and increasing or decreasing the numerical data corresponding to the specific process Qs. By increasing or decreasing one or more control factors based on the stored contents, it becomes possible to control the specific process Qs in a direction that improves quality.

[0182] Each control factor constituting each manufacturing process Q may be regarded as an independent manufacturing process Q. For example, in an electrodeposition coating process, the electrical conductivity may be regarded as one manufacturing process Q, and the paint temperature may be regarded as another manufacturing process Q.

[0183] In other words, instead of directly estimating the specific process Qs, the CPU 3 may estimate the control factors that make up the specific process Qs based on the specific feature Ps and two or more types of machine learning models, for each type of machine learning model.

[0184] (6-7. Display example) 19 and 20 are diagrams illustrating display screens Sc and Sc' on the display 11. For example, when the processing shown in FIG. 11 is completed, the CPU 3 displays on the display 11 the specific feature Ps that contributed to the selection of the specific process Qs, the specific process Qs selected by the first machine learning model 691, and the specific process Qs selected by the second machine learning model 692. See boxed section C1 in FIG. 19 for the specific feature Ps, boxed section C2 for the specific process Qs related to the first machine learning model 691, and boxed section C3 for the specific process Qs related to the second machine learning model 692.

[0185] Next, the CPU 3 generates a union of the specific process Qs associated with the first machine learning model 691 and the specific process Qs associated with the second machine learning model 692 (see boxed portion C4). As shown in FIG. 20, which will be described later, the CPU 3 determines whether the union is an empty set.

[0186] The CPU 3 also calculates the aforementioned effect index for each of the specific processes Qs included in the union. In the illustrated example, for each of the specific processes Qs, "process a," "process b," and "process c," the related specific feature values ​​Ps are selected and grouped. For each specific process Qs, an effect index is calculated for each specific feature value Ps. The calculated effect indexes are summed for each specific process Qs.

[0187] The CPU 3 displays the specific processes Qs and their control modes (e.g., "increase," "decrease," etc.) on the display 11 in descending order of the calculated effect index (see boxed section C5). If the specific process Qs includes multiple control factors, a display may be made for each control factor (e.g., a display indicating "increase paint temperature"). Furthermore, if the specific feature quantity Ps corresponding to the specific process Qs is, for example, a categorical value and it is not possible to set a correlation coefficient, and therefore "increase," "decrease," etc., it may be displayed as "unpredictable," as in the example "process c."

[0188] Furthermore, as shown in boxed areas C12 and C13 in Figure 20, if the specific process Qs differs between the two models 691, 692, a message indicating this (e.g., "Mismatch") may be displayed on screen Sc', and the number of samples for each model 691, 692 may also be displayed (see boxed area C14).

[0189] 7. Significance of the analysis method As described above, the analysis method according to the embodiment uses two or more machine learning models 691, 692 that associate each of a plurality of feature quantities P with one or more of a plurality of manufacturing processes Q (see FIGS. 15, 16A, and 16B). Here, consider the appearance of a product W as an example of quality.

[0190] In this example, once the feature value P estimated to have contributed to the deterioration of appearance, such as the presence or absence of a scratch and its location, is determined, the manufacturing process Q that caused the scratch can be estimated using the machine learning model 69. Estimation using the machine learning model 69 can be performed easily and with high accuracy, even by non-expert workers.

[0191] Furthermore, in the case of a product W consisting of many parts, such as a car body, it is believed that there are an infinite number of feature values ​​P that can be used as candidates for analysis. In such cases, even if a skilled worker is involved, it is not easy to identify the process that is responsible for the quality. Estimation using the machine learning model 69 can be performed easily and with high accuracy, even when there are many feature values.

[0192] Furthermore, rather than simply using machine learning model 69, the manufacturing process Q is estimated using each of two or more machine learning models 691 and 692. This allows for more multifaceted analysis and more accurate analysis.

[0193] Furthermore, as illustrated in Figures 14 and 15A, the two or more types of machine learning models 69 include a first machine learning model 691 based on a decision tree algorithm and a second machine learning model 692 based on a non-decision tree algorithm.

[0194] By using these combinations, a more multifaceted analysis can be performed, which makes it possible to achieve a more accurate analysis.

[0195] Furthermore, the importance of each manufacturing process Q can be calculated automatically depending on the machine learning model 69 selected. For example, in the case of a random forest model as in this embodiment, the importance specific to the model can be calculated. In this way, it is possible to reduce the possibility of the user's intervention in determining the specific process Qs. This allows for more accurate analysis.

[0196] 16A is used to select the second machine learning model 692, the manufacturing process Q is determined based on probabilistic connections (dependencies). This results in a more multifaceted analysis, making it possible to achieve a more accurate analysis.

[0197] Furthermore, even if the estimation results from two types of machine learning models 69 differ from each other, one estimation result is not overwritten or discarded by the other, and the estimation results when using each machine learning model 69 are notified to the user. At that time, by notifying the number of samples as well, it becomes possible to perform a more comprehensive analysis, such as an analysis of the learning status of each model.

[0198] 12, the specific process Qs is not simply determined, but a quality improvement plan is also proposed, making it possible for even an unskilled worker to easily improve the quality of the product W.

[0199] 12, when proposing a quality improvement plan, a highly visible display can be provided, which makes it possible to easily improve the quality of the product W.

[0200] Furthermore, the quality determination results used in the present disclosure are not limited to the output from the machine learning model 69. By having a worker perform the quality determination and then having the computer 1 perform the subsequent processes as described above, it is possible to achieve division of labor at the manufacturing site. This can improve the ease of use of the analysis method.

[0201] Furthermore, the feature P used in the present disclosure is not limited to electrical data. For example, the analysis method according to the present disclosure can be executed based on text data written by a worker (e.g., text describing the condition of a scratch, etc.). This can improve the usability of the analysis method.

[0202] 5, the specific feature Ps is determined by the performance evaluation model 59, separately from the machine learning model 69 for determining the manufacturing process Q. This reduces the possibility of the user's intervention in the analysis of the manufacturing process Q. This allows for a more accurate analysis to be achieved.

[0203] As described with reference to FIG. 2 etc., the present disclosure is particularly effective for analyzing the rust prevention performance of the metal product 201.

[0204] <8. Other embodiments> In the above embodiment, the output data (quality data 89) of the performance evaluation program 292 is used as the quality determination result, but the present disclosure is not limited to such a configuration.

[0205] That is, in the flow shown in FIG. 5, at least the measurement data analysis process (step S1) and the performance evaluation process (step S2) may be omitted.

[0206] The "quality judgment result" in the present disclosure may be, instead of or in addition to the output from a pre-generated machine learning model, a judgment result by a factory worker or a judgment result by rule-based judgment using multiple feature quantities P as input. For example, when using the judgment result by a factory worker, as described above, the configuration of machine learning model 69 can be reconfigured so that the text data recorded by the factory worker is used as input.

[0207] The feature value P used in this process may be, for example, the index (color, shape, texture, etc. of the product) used by a worker when he or she judges the quality based on experience. By quantifying these indexes, the process similar to that of the above embodiment can be performed.

[0208] In addition, in the above embodiment, a machine learning model based on a decision tree algorithm is exemplified as an example of a nonlinear regression model, but the present disclosure is not limited to such a configuration. Instead of the decision tree algorithm, a so-called support vector machine may also be used.

[0209] Furthermore, although the above embodiment has shown an example in which the analysis device is configured by one computer 1, the present disclosure is not limited to this example. The analysis method and analysis program 29 according to the present disclosure may be executed using multiple computers 1, such as by having a first computer execute processing related to the specific process estimation process and a second computer execute processing related to the correlation analysis process. Furthermore, the computer 1 in the present disclosure also includes parallel computers such as supercomputers and PC clusters.

[0210] Furthermore, the screen on which various information can be displayed is not limited to the display screen on the display 11 of the computer 1. The graph structure may be displayed on a screen prepared separately from the computer 1. [Explanation of symbols]

[0211] 100 Manufacturing management system (rust prevention performance analysis system) 122 Inspection equipment 1. Computer 3 CPU (arithmetic unit) 7 RAM (memory section) 9 SSD (storage unit) 11 Display (display unit) 15 Keyboard (reception area) 17 Mouse (Reception) 18 Storage medium 29 Analysis Program 291 Measurement Data Analysis Program 292 Performance Evaluation Program 293 Feature Selection Program 294 Model Loading Program 295 Specific Process Estimation Program 296 Correlation Analysis Program 49 Measurement Data 59 Performance Evaluation Model 69 Machine Learning Models 691 First Machine Learning Model 692 Second Machine Learning Model 89 Quality Data P feature Ps specific feature Q Manufacturing process Qs specific process W Products 201 Metal products 203 Insulating layer

Claims

1. A rust prevention performance analysis method for analyzing the rust prevention performance of a metal product manufactured through a plurality of manufacturing processes by using a computer having a storage unit and a calculation unit, Among the plurality of feature quantities used for determining the rust-preventive performance, one or more feature quantities that are estimated to contribute to the determination of the rust-preventive performance are defined as specific feature quantities, and among the plurality of manufacturing processes, one or more manufacturing processes that are estimated to contribute to the determination of the rust-preventive performance are defined as specific processes. the calculation unit applies a voltage to the surface of the metal product in a state where a corrosion factor is in contact with the surface of the metal product, and acquires a change over time in a current caused by the voltage; the calculation unit determines and outputs the rust-preventive performance based on a plurality of feature amounts that characterize the waveform of the change over time; the calculation unit reads from the storage unit two or more types of machine learning models that have been generated in advance so as to associate each of the plurality of feature amounts with one or more of the plurality of manufacturing processes; reading the specific feature from among the plurality of feature amounts; The calculation unit estimates the identifying step for each type of machine learning model based on the specific feature and the two or more types of machine learning models. A rust prevention performance analysis method characterized by:

2. 2. The rust prevention performance analysis method according to claim 1, The two or more machine learning models are a first machine learning model based on a decision tree algorithm; a second machine learning model based on a non-decision tree algorithm. A rust prevention performance analysis method characterized by:

3. 3. The rust prevention performance analysis method according to claim 2, The calculation unit determines the specified process based on the importance of each of the plurality of manufacturing processes corresponding to each of the plurality of feature amounts in the first machine learning model. A rust prevention performance analysis method characterized by:

4. 3. The rust prevention performance analysis method according to claim 2, the second machine learning model is a Bayesian network in which each of the plurality of feature quantities is a child node and each of the plurality of manufacturing processes is a parent node; The calculation unit determines, as the specified process, a manufacturing process that is connected to the specific feature via one edge when the Bayesian network is visualized as a directed graph structure. A rust prevention performance analysis method characterized by:

5. 3. The rust prevention performance analysis method according to claim 2, When the determination result by the first machine learning model and the determination result by the second machine learning model differ, Estimation results of the first and second machine learning models; and and the number of samples used in generating each of the first and second machine learning models. A rust prevention performance analysis method characterized by:

6. 2. The rust prevention performance analysis method according to claim 1, The quality judgment result is quantified as quality data that increases or decreases depending on whether the quality is good or bad, The plurality of manufacturing processes are quantified so as to characterize the content of each manufacturing process, The storage unit a first correlation indicating a correlation between each of the plurality of feature amounts and the quality data; a second correlation indicating a correlation between each of the plurality of manufacturing processes and each of the plurality of feature quantities; After determining the specifying step, the calculation unit calculates, based on the first and second correlations, A breakdown of the specific steps; and notifying the control mode of the specific process for improving the quality. A rust prevention performance analysis method characterized by:

7. 7. The rust prevention performance analysis method according to claim 6, The calculation unit includes a display unit capable of displaying the calculation results of the calculation unit. A breakdown of the specific steps; and a control mode of the specific process for improving the quality, in order of the specific process contributing to the quality. A rust prevention performance analysis method characterized by:

8. 2. The rust prevention performance analysis method according to claim 1, The calculation unit reading a performance determination model generated in advance by machine learning so as to associate the plurality of feature amounts with the degree of quality; determining the degree of pass / fail based on the plurality of feature amounts and the performance determination model; The calculation unit also determines the specific feature quantity based on the importance of each of the plurality of feature quantities corresponding to the degree of pass / fail in the performance judgment model. A rust prevention performance analysis method characterized by:

9. A rust prevention performance analysis system that includes a computer having a storage unit and a calculation unit, and analyzes the rust prevention performance of a product manufactured through a plurality of manufacturing processes, Among the plurality of feature quantities used for determining the rust-preventive performance, one or more feature quantities that are estimated to contribute to the determination of the rust-preventive performance are defined as specific feature quantities, and among the plurality of manufacturing processes, one or more manufacturing processes that are estimated to contribute to the determination of the rust-preventive performance are defined as specific processes. a measuring device that applies a voltage to the surface of the metal product while the corrosion factor is in contact with the surface of the metal product, and acquires a change over time in the current caused by the voltage; a performance evaluation means for determining and outputting the rust-preventing performance based on a plurality of feature quantities that characterize the waveform of the change over time; a model reading means for reading from the storage unit two or more types of machine learning models that have been generated in advance so as to associate each of the plurality of feature quantities with one or more of the plurality of manufacturing processes; and a specific process estimation means for reading the specific feature from among the plurality of feature amounts and estimating the specific process for each type of machine learning model based on the specific feature and the two or more types of machine learning models. A rust prevention performance analysis system characterized by:

10. A rust prevention performance analysis program executed by a computer having a storage unit and a calculation unit, which analyzes the rust prevention performance of a product manufactured through a plurality of manufacturing processes, Among the plurality of feature quantities used for determining the rust-preventive performance, one or more feature quantities that are estimated to contribute to the determination of the rust-preventive performance are defined as specific feature quantities, and among the plurality of manufacturing processes, one or more manufacturing processes that are estimated to contribute to the determination of the rust-preventive performance are defined as specific processes. The computer, a step in which a measuring device applies a voltage to the surface of the metal product while the corrosion factor is in contact with the surface, thereby acquiring a change over time in a current caused by the voltage; the calculation unit determining and outputting the rust prevention performance based on a plurality of feature quantities that characterize the waveform of the change over time; the calculation unit reading from the storage unit two or more types of machine learning models that have been generated in advance so as to associate each of the plurality of feature amounts with one or more of the plurality of manufacturing processes; the calculation unit reads the specific feature from among the plurality of feature, and estimates the identifying process for each type of machine learning model based on the specific feature and the two or more types of machine learning models. A rust prevention performance analysis program characterized by:

11. The rust prevention performance analysis program according to claim 10 is stored. A computer-readable storage medium comprising:

Images