Learning apparatus and methods, evaluation apparatus, systems and methods, recording media

Abstract

This invention provides a learning device, an evaluation device, an evaluation system, a learning method, an evaluation method, and a recording medium. The learning device includes: a correspondence receiving unit that receives correspondences between the quality evaluations of each product of an object process and the quality evaluations of each downstream product produced in a downstream process using the products of the object process; a learning processing unit that uses at least one production parameter relating to the production of each product of the object process and the quality evaluations of each downstream product produced using the products of the object process to generate a predictive model that infers the quality evaluation of the downstream products based on the at least one production parameter; a calculation unit that calculates a model evaluation based on at least one of the accuracy or complexity of the predictive model; and a model evaluation sending unit that sends the model evaluation calculated by the calculation unit to the evaluation device, which evaluates at least one upstream process using model evaluations for each of at least one upstream process that is upstream of the downstream process.

Description

Technical Field

[0001] This invention relates to learning devices, evaluation devices, evaluation systems, learning methods, evaluation methods, and recording media. Background Technology

[0002] Patent document 1 describes "an analytical method for determining the factors that hinder product performance and thereby stabilizing product performance in a manufacturing process".

[0003] Existing technical documents

[0004] Patent Document 1: Japanese Patent Publication No. 2016-177794 Summary of the Invention

[0005] In a first aspect of the present invention, a learning apparatus is provided. The learning apparatus may include a correspondence receiving unit that receives correspondences between the quality evaluations of each product of an object process and the quality evaluations of each downstream product produced in a downstream process using the products of the object process. The learning apparatus may include a learning processing unit that, using at least one production parameter relating to the production of each product of the object process and the quality evaluations of each downstream product produced using the products of the object process, generates a predictive model that infers the quality evaluations of the downstream products based on at least one production parameter. The learning apparatus may include a calculation unit that calculates a model evaluation based on at least one of the accuracy or complexity of the predictive model. The learning apparatus may include a model evaluation sending unit that sends the model evaluation calculated by the calculation unit to an evaluation device that evaluates at least one upstream process using model evaluations for each of at least one upstream process that is upstream of the downstream process.

[0006] The learning device may include an improvement request receiving unit that receives improvement request information sent by an evaluation device, which determines the target process to be improved based on a model evaluation for each of at least one upstream process and sends the improvement request information accordingly.

[0007] The learning device may include a parameter selection unit that selects a production parameter to be adjusted from at least one production parameter in the target process based on received improvement request information.

[0008] The learning device may include a quality evaluation prediction unit that predicts the quality of downstream products produced in downstream processes using each product of the target process, after adjusting the production parameters selected by the parameter selection unit.

[0009] In a second aspect of the invention, an evaluation apparatus is provided. The evaluation apparatus may include a model evaluation receiving unit that, for each of at least one upstream process, receives a model evaluation based on at least one of the accuracy or complexity of the inference model, obtained from a learning apparatus that generates an inference model based on at least one production parameter relating to the production of each upstream product of that upstream process and the quality evaluation of each downstream product produced using the upstream products of that upstream process, and through learning. The evaluation apparatus may also include a process evaluation unit that evaluates at least one upstream process based on the model evaluation for each of the at least one upstream process.

[0010] The process evaluation department can determine that an upstream process for which a model evaluation has been provided, in at least one upstream process, is required to be improved if the accuracy of the predicted model is greater than the benchmark value or the complexity is less than the benchmark value.

[0011] The evaluation device may include an information output unit that outputs improvement request information for upstream processes that are determined to require improvement.

[0012] The evaluation apparatus may include a correspondence acquisition unit that, for each process of at least one upstream process, acquires a correspondence between each product supplied from the upstream process and each product supplied to the downstream process. The evaluation apparatus may include a correspondence generation unit that, using the correspondences acquired by the correspondence acquisition unit, generates a correspondence relationship for quality evaluation between each product of at least one upstream process and each downstream product. The evaluation apparatus may include a correspondence transmission unit that transmits the correspondence relationship to a learning device.

[0013] The evaluation device may include a start determination unit that determines, based on the quality evaluation of at least one downstream product, whether to begin the evaluation of at least one upstream process by the process evaluation unit.

[0014] A third aspect of the invention provides an evaluation system. The evaluation system may include a learning device that takes at least one upstream process as at least one target process. The evaluation system may include an evaluation device.

[0015] A fourth aspect of the present invention provides a learning method. The learning method may include a learning device receiving a correspondence between the quality evaluations of each product of an object process and the quality evaluations of each downstream product produced in a downstream process using the products of the object process. The learning method may include the learning device using at least one production parameter relating to the production of each product of the object process and the quality evaluations of each downstream product produced using the products of the object process to generate a predictive model that infers the quality evaluations of the downstream products based on at least one production parameter. The learning method may include the learning device calculating a model evaluation based on at least one of the accuracy or complexity of the predictive model. The learning method may include the learning device sending the calculated model evaluation to an evaluation device that evaluates at least one upstream process using model evaluations for each of at least one upstream process that is upstream of the downstream process.

[0016] In a fifth aspect of the invention, a recording medium is provided that records a learning program executed by a computer. The computer can function as a correspondence receiving unit by executing the learning program, receiving correspondences between the quality evaluations of each product of an object process and each downstream product produced in a downstream process using the products of the object process. The computer can also function as a learning processing unit by executing the learning program, using at least one production parameter relating to the production of each product of the object process and the quality evaluations of each downstream product produced using the products of the object process, to generate a predictive model that infers the quality evaluation of the downstream products based on at least one production parameter. Furthermore, the computer can function as a calculation unit by executing the learning program, calculating a model evaluation based on at least one of the accuracy or complexity of the predictive model. Finally, the computer can function as a model evaluation sending unit by executing the learning program, sending the model evaluation calculated by the calculation unit to an evaluation device, which evaluates at least one upstream process using model evaluations for each of at least one upstream process that is upstream of the downstream process.

[0017] A sixth aspect of the present invention provides an evaluation method. The evaluation method may include an evaluation device, for each of at least one upstream process, receiving a model evaluation based on at least one production parameter relating to the production of each upstream product of that upstream process and the quality evaluation of each downstream product produced using that upstream product, and a learning device that generates a predictive model based on the at least one production parameter to predict the quality evaluation of the downstream products, and receiving a model evaluation based on at least one of the accuracy or complexity of the predictive model. The evaluation method may include the evaluation device evaluating at least one upstream process based on the model evaluation for each of the at least one upstream process.

[0018] In a seventh aspect of the invention, a recording medium storing an evaluation program executed by a computer is provided. The computer can function as a model evaluation receiving unit by executing the evaluation program. This model evaluation receiving unit, for each of at least one upstream process, receives a model evaluation based on at least one of the accuracy or complexity of the inference model, obtained from a learning device that generates an inference model based on at least one production parameter relating to the production of each upstream product of that upstream process and a quality evaluation of each downstream product produced using that upstream product, and through learning. The computer can also function as a process evaluation unit by executing the evaluation program, evaluating at least one upstream process based on the model evaluation for each of the at least one upstream process.

[0019] Furthermore, the above summary of the invention does not list all the features of the invention. In addition, sub-combinations of these feature groups can also constitute inventions. Attached Figure Description

[0020] Figure 1 The evaluation system 10 of this embodiment is represented together with the supply chain including multiple manufacturers 20a to 1.

[0021] Figure 2 This describes the configuration of the learning device 100 in this embodiment.

[0022] Figure 3 This represents an example of production data stored in the production data storage unit 215 of this embodiment.

[0023] Figure 4 This describes the operation flow of the learning device 100 in this embodiment.

[0024] Figure 5 This describes the configuration of the evaluation device 110 in this embodiment.

[0025] Figure 6 This describes the operation flow of the evaluation device 110 in this embodiment.

[0026] Figure 7 An example illustrating the correspondence between the products produced in each process step and the final product.

[0027] Figure 8 Examples of computer 2200 that can implement the present invention in whole or in part are shown.

[0028] Explanation of reference numerals in the attached figures

[0029] 10 Evaluation System, 20a-1 Manufacturers, 30 Equipment, 100d-1 Learning Device, 105a-c Product Evaluation Device, 110 Evaluation Device, 210 Production Management Department, 215 Production Data Storage Department, 220 Correspondence Sending Department, 225 Correspondence Receiving Department, 235 Learning Processing Department, 240 Predictive Model Storage Department, 245 Calculation Department, 250 Model Evaluation Sending Department, 255 Improvement Request Receiving Department, 260 Parameter Selection Department, 265 Quality Evaluation Prediction Department, 500 Correspondence Acquisition Department, 510 Correspondence Generation Department, 520 Correspondence Data Storage unit, 530 Correspondence transmission unit, 550 Start determination unit, 560 Model evaluation receiving unit, 570 Process evaluation unit, 580 Information output unit, 2200 Computer, 2201 DVD-ROM, 2210 Main controller, 2212 CPU, 2214 RAM, 2216 Graphics controller, 2218 Display device, 2220 Input / output controller, 2222 Communication interface, 2224 Hard disk drive, 2226 DVD-ROM drive, 2230 ROM, 2240 Input / output chip, 2242 Keyboard. Detailed Implementation

[0030] The present invention will now be described through embodiments thereof; however, these embodiments do not limit the invention as defined in the claims. Furthermore, not all combinations of features described in the embodiments are necessary for the solution of the invention.

[0031] Figure 1 The evaluation system 10 of this embodiment is represented together with a supply chain including multiple manufacturers 20a-1. For example, there is a common process, namely the supply chain, from production to sale of tangible goods such as automobiles, electrical appliances, industrial equipment, clothing, and food. This is referred to as a supply chain. The supply chain includes multiple production steps (also referred to as "steps") from raw materials to finished products. As an example, the supply chain shown in this figure includes raw material manufacturers 20j-1, raw material manufacturers 20g-i, component manufacturers 20d-f, and product manufacturers 20a-c (collectively referred to as "manufacturers 20").

[0032] Raw material manufacturers 20j to 1 use, for example, crude oil or natural gas as starting materials to produce petroleum products and other raw materials, and supply them to raw material manufacturers 20g to 1. In the example shown in the diagram, raw material manufacturer 20j supplies raw materials to raw material manufacturer 20g, raw material manufacturer 20k supplies raw materials to raw material manufacturers 20g to 20h, and raw material manufacturer 20l supplies raw materials to raw material manufacturer 20i.

[0033] Raw material manufacturers 20g to 20i use petroleum products, which are also products of raw material manufacturers 20j to 20l, as raw materials to produce resin materials and other raw materials, and supply them to component manufacturers 20d to 20f. In the example shown in the diagram, raw material manufacturer 20g supplies raw materials to component manufacturers 20e to 20f, raw material manufacturer 20h supplies raw materials to component manufacturer 20d, and raw material manufacturer 20i supplies raw materials to component manufacturer 20f.

[0034] Component manufacturers 20d to 20f use resin materials and other materials, which are the same as those used in the products of raw material manufacturers 20g to 20i, as raw materials to produce electronic components and other parts, and supply them to product manufacturers 20a to 20c. In the example shown in the figure, component manufacturer 20d supplies components to product manufacturer 20a, component manufacturer 20e supplies components to product manufacturers 20a to 20b, and component manufacturer 20f supplies components to both product manufacturers 20a and 20c.

[0035] Product manufacturers 20a-c assemble electronic components and other parts that are used in the products of component manufacturers 20d-f to produce finished products. Thus, in the supply chain, multiple manufacturers 20, such as raw material manufacturers 20j-l, raw material manufacturers 20g-i, component manufacturers 20d-f, and product manufacturers 20a-c, become participants from upstream to downstream, realizing the supply chain.

[0036] Here, "production process (process)" refers to the entire operation of manufacturer 20, which processes products received from upstream sources to produce its own product. Manufacturer 20 may use a production line located in a factory to perform manufacturing operations equivalent to a "production process," or it may include multiple operation processes that further subdivide such a production line itself. In addition, in this embodiment, one "production process" refers to the entire manufacturing operation undertaken by manufacturer 20, but a "production process" may correspond to only a part of the manufacturing operation undertaken by one manufacturer 20, or it may include a series of manufacturing operations undertaken by two or more manufacturers 20.

[0037] In such a supply chain, the quality of products in downstream production processes can generally be affected by the quality of products in upstream production processes. That is, for example, the quality of the finished product (product) of product manufacturer 20b can be affected by the quality of the components (product) of component manufacturer 20e, which is a supply source. Similarly, the quality of the components (product) of component manufacturer 20e can be affected by the quality of the raw materials (product) of raw material manufacturer 20g, which is a supply source. Likewise, the quality of the raw materials (product) of raw material manufacturer 20g can be affected by the quality of the raw materials (products) of raw material manufacturers 20j to 20k, which are supply sources. In this embodiment, production within the interconnected supply chain that may produce such quality issues can be considered as the evaluation object.

[0038] In recent years, supply chains have become more complex. As illustrated in this diagram, component manufacturer 20e frequently supplies components to competing product manufacturers 20a and 20b. Therefore, if product manufacturer 20a, in order to improve the quality of the final product or to reduce its cost, provides detailed technical information to component manufacturer 20e regarding quality improvement or cost reduction of components, this technical information may flow to competing product manufacturer 20b, or it may contribute to improving product quality or reducing costs for competing product manufacturer 20b. Therefore, product manufacturer 20a desires to minimize the information provided to component manufacturer 20e and to improve the quality of components produced by component manufacturer 20e or reduce costs.

[0039] The same relationship occurs between component manufacturers 20d-f and raw material manufacturers 20g-i, and between raw material manufacturers 20g-i and raw material manufacturers 20j-l, further upstream in the supply chain. As a result, upstream manufacturers 20d-l arbitrarily make process improvements based on imagination or assumptions, relying on limited information from the downstream side of the supply chain. This generates a large amount of wasted labor hours in the supply chain as a whole, and progress in improvement is slow.

[0040] Therefore, the evaluation system 10 of this embodiment provides an evaluation environment for suppressing the information that each manufacturer 20 should provide and improving the processes in each manufacturer 20. The evaluation system 10 includes: one or more product evaluation devices 105a-c (also referred to as "product evaluation device 105") corresponding to the final process (the most downstream process) in the supply chain, one or more learning devices 100d-l (also referred to as "learning device 100") corresponding to each process in the supply chain other than the final process, and an evaluation device 110.

[0041] Each of the plurality of product evaluation devices 105 is provided corresponding to the final process of manufacturing the final product (the most downstream product). That is, in this embodiment, each of the plurality of product evaluation devices 105 is provided at each of the product manufacturers 20 that have the final process of manufacturing the final product.

[0042] The product evaluation device 105 evaluates the quality of each final product. Furthermore, the product evaluation device 105 supplies the evaluation device 110 with corresponding quality evaluations of each product from the direct upstream side and the final products produced using those products. Additionally, in this embodiment, the evaluation system 10 conveniently considers the process for which the product quality evaluation is obtained as the "final process," and uses the quality evaluation of the final process to evaluate processes upstream of the final process. "Final process" refers to the downstream process in the supply chain that is the object of both the product quality evaluation and the process evaluation based on the evaluation system 10. Therefore, the final products of product manufacturers 20a-c may not necessarily be finished products delivered to users and usable by them; they may be used further downstream in the supply chain where the evaluation based on the evaluation system 10 is conducted (i.e., downstream of manufacturers 20a-c) as components of other products.

[0043] In the aforementioned sense, the process that obtains a product quality evaluation is not necessarily the final or most downstream process in the overall supply chain. Therefore, the most downstream process within the scope of product quality evaluation and process evaluation based on evaluation system 10, i.e., the process that obtains a product quality evaluation, is simply referred to as a "downstream process," and the product of the "downstream process" is also referred to as a "downstream product." Furthermore, within the scope of product quality evaluation and process evaluation based on evaluation system 10, processes upstream of the "downstream process" are also referred to as "upstream processes," and the product of the "upstream process" is also referred to as an "upstream product."

[0044] Here, "upstream" means, unless it is expressed as "direct upstream," for example, not limited to the process of raw material manufacturer 20h relative to the process of component manufacturer 20d, but also includes processes that are upstream of the process of raw material manufacturer 20k relative to the process of component manufacturer 20d, which are separated by at least one other process. Similarly, "downstream of a certain process" means, unless it is expressed as "direct downstream of a certain process," it is not limited to direct downstream, but also includes processes that are downstream of the process separated by at least one other process.

[0045] The learning device 100 is provided corresponding to each of at least one upstream process that is the object of process evaluation based on the evaluation system 10. In this embodiment, for ease of explanation, the evaluation system 10 takes multiple upstream processes of multiple manufacturers 20d to 1 as the objects of process evaluation. Therefore, in this embodiment, each of the multiple learning devices 100 is provided corresponding to each process that is upstream of the downstream process (final process). That is, in this embodiment, each of the multiple learning devices 100 is provided at each of the manufacturers 20d to 1 that have upstream processes that are upstream of the downstream processes.

[0046] Furthermore, the evaluation system 10 can also include one or more downstream processes as the objects of process evaluation. In this case, the product evaluation device 105 may also include a learning device 100 for downstream processes as part of it, so that downstream processes can also be included as the objects of process evaluation.

[0047] For each process being processed, the learning device 100 provides the evaluation device 110 with the correspondence between each product supplied from the direct upstream process of the target process and each product of the target process. Furthermore, the learning device 100 receives from the evaluation device 110 the correspondence between the quality evaluations of each product of the target process and each downstream product. Using one or more production parameters used in the production of each product in the target process, and the quality evaluations of each downstream product produced using the products of the target process, the learning device 100 generates a predictive model that infers the quality evaluation of downstream products based on one or more production parameters. The learning device 100 then calculates the model evaluation based on the learned predictive model and sends it to the evaluation device 110.

[0048] Evaluation device 110 is connected to each of the plurality of product evaluation devices 105 and each of the plurality of learning devices 100. Evaluation device 110 receives from each product evaluation device 105 the correspondence between the quality evaluations of each product directly upstream of the downstream process and the quality evaluations of each downstream product. Furthermore, for each process in manufacturer 20d-1 (except the most upstream process), evaluation device 110 receives from each learning device 100 the correspondence between the products supplied from the upstream process and the products of the target process. Evaluation device 110 uses these correspondences to generate a correspondence between the quality evaluations of each product of each upstream process and the quality evaluations of each downstream product produced using the products of each upstream process. Furthermore, evaluation device 110 sends the correspondence between the quality evaluations of each product of each process and the quality evaluations of each downstream product to the learning device 100 corresponding to each process.

[0049] Furthermore, the evaluation device 110 evaluates at least one process that is the object of process evaluation using a model for each of the at least one process that is the object of process evaluation. Also, based on a determination that a certain process should be improved, the evaluation device 110 sends an improvement request message to the learning device 100 corresponding to that process.

[0050] In summary, the quality of the final product can be significantly influenced by specific production parameters used in each process from upstream to downstream. When a particular production parameter in a certain process has a significant impact on the quality of the downstream product, the learning device 100 corresponding to that process can generate a sound predictive model that infers the quality evaluation of the downstream product based on the production parameters of that process. Conversely, when any production parameter in a certain process has almost no impact on the quality of the downstream product, even if the learning device 100 corresponding to that process generates a predictive model that infers the quality evaluation of the downstream product based on the production parameters of that process, it will not be a sound predictive model that explains the quality of the downstream product based on the production parameters of that process.

[0051] The evaluation system 10 utilizes this property, and the learning device 100 corresponding to each process generates a predictive model for each process, which infers the quality evaluation of the downstream product based on at least one production parameter. The evaluation device 110 requests improvement for the process corresponding to the learning device 100 that can generate a good predictive model. Thus, the evaluation system 10 is able to request improvement for processes that use production parameters that have a significant impact on the quality of the downstream product.

[0052] According to the evaluation system 10 described above, each learning device 100 sends a model evaluation to the evaluation device 110, which includes the correspondence between the products supplied from the upstream side to the target process and the products of the target process, as well as the generated speculative model. However, it does not need to send the production parameters used in the target process to the evaluation device 110. The evaluation device 110 can evaluate each process using the model evaluation received from each learning device 100 without receiving the production parameters used in each learning device 100. Therefore, the evaluation system 10 can provide an environment in which manufacturers 20 can improve the target processes that affect the quality of the final product without disclosing production parameters.

[0053] Figure 2 This describes the configuration of the learning device 100 in this embodiment. The learning device 100 can be a portable computer such as a PC (personal computer), workstation, server computer, or general-purpose computer installed on the corresponding manufacturer 20, a tablet computer, or a smartphone, or it can be a computer system connected to multiple computers. Such a computer system is also a computer in a broad sense. The learning device 100 can be implemented by executing a learning program on such a computer. Furthermore, the learning device 100 can also be implemented within a computer using one or more executable virtual computer environments. Alternatively, the learning device 100 can be a dedicated computer designed to perform various processes related to the target process, or it can be dedicated hardware implemented using dedicated circuitry.

[0054] Furthermore, the learning device 100 can also be implemented via a cloud computing system. In this case, the learning device 100 may not disclose one or more production parameters used in the target process to other manufacturers 20 responsible for other processes.

[0055] The learning device 100 includes: a production management unit 210, a production data storage unit 215, a correspondence sending unit 220, a correspondence receiving unit 225, a learning processing unit 235, a prediction model storage unit 240, a calculation unit 245, a model evaluation sending unit 250, an improvement request receiving unit 255, a parameter selection unit 260, and a quality evaluation prediction unit 265.

[0056] The production management department 210 is connected to the equipment 30 responsible for production in the target process that is the object of the learning device 100, and manages the production of the products of the equipment 30. The equipment 30 is installed in, for example, an industrial plant such as a chemical plant, a plant that manages and controls the wellhead and its surroundings of a gas field or oil field, or a manufacturing plant for industrial products. For example, the equipment 30 may include a control device such as a distributed control system (DCS), and one or more processing devices that are controlled by such a control device to produce products, and the production management department 210 may be connected to such a control device.

[0057] The production management unit 210 obtains from the equipment 30 the correspondence between each product of materials or components supplied to the target process from the upstream side and each product of the target process supplied downstream, and stores this correspondence in the production data storage unit 215. Here, the downstream process learning device 100 can obtain from the equipment 30 the correspondence between each product of materials or components supplied to the target process from the upstream side and the quality evaluation of the downstream products produced using each product, and store this correspondence in the production data storage unit 215. Alternatively, the production management unit 210 can obtain these correspondences by receiving input from operators or others managing the target process, instead of obtaining them from the equipment 30.

[0058] Furthermore, the production management department 210 obtains one or more production parameters from the equipment 30 regarding the production of each product in the target process. Here, "product" can be a single product such as a component or product, or a product of a specific management unit such as a predetermined quantity (e.g., 1 cubic meter) or a predetermined number (e.g., 1 batch). In this embodiment, for ease of explanation, "product" refers to 1 batch of products.

[0059] The production data storage unit 215 is connected to the production management unit 210. The production data storage unit 215 stores the corresponding data and one or more production parameters obtained by the production management unit 210.

[0060] The corresponding sending unit 220 is connected to the production data storage unit 215. The corresponding sending unit 220 sends the correspondence regarding the object process of the learning device 100 stored in the production data storage unit 215 to the evaluation device 110.

[0061] The correspondence receiving unit 225 receives from the evaluation device 110 the correspondence between the quality evaluations of each product in the target process and each downstream product produced in the downstream process using the products in the target process. Alternatively, if a learning device 100 is also provided in the final process, the learning device 100 in the final process may not have a correspondence receiving unit 225, and can obtain the correspondence between the quality evaluations of each product (materials or components, etc.) in the target process and each downstream product stored in the production data storage unit 215.

[0062] The learning processing unit 235 is connected to the production data storage unit 215 and the correspondence receiving unit 225. Using the correspondence between the quality evaluations of each product in the target process and each downstream product, the learning processing unit 235 correlates one or more production parameters stored in the production data storage unit 215 regarding the production of each product in the target process with the quality evaluations of each downstream product. Furthermore, using one or more production parameters regarding the production of each product in the target process, and using the quality evaluations of each downstream product produced by each product in the target process, the learning processing unit 235 generates a predictive model that infers the quality evaluation of downstream products based on one or more production parameters.

[0063] The prediction model storage unit 240 is connected to the learning processing unit 235. The prediction model storage unit 240 stores the prediction models generated by the learning processing unit 235.

[0064] The calculation unit 245 is connected to the prediction model storage unit 240. The calculation unit 245 calculates a model evaluation of the prediction model generated by the learning processing unit 235 and stored in the prediction model storage unit 240. The model evaluation sending unit 250 is connected to the calculation unit 245. The model evaluation sending unit 250 sends the model evaluation calculated by the calculation unit 245 to the evaluation device 110. The evaluation device 110 receives the model evaluation and evaluates at least one upstream process using model evaluations for each of at least one upstream process that is upstream of the downstream process that provides the product quality evaluation. Based on the model evaluations for each of the at least one upstream process, the evaluation device 110 determines that the process should be improved and sends improvement request information to the learning device 100 corresponding to the process. In addition, when at least one downstream process is also the object of process evaluation, the evaluation device 110 may also use model evaluations for each process including at least one upstream process and at least one downstream process that are the objects of process evaluation to evaluate each upstream process and send improvement request information to the learning device 100 corresponding to the process that is the object of improvement.

[0065] When the evaluation device 110 sends an improvement request for the target process, the improvement request receiving unit 255 receives the improvement request. The parameter selection unit 260 is connected to the prediction model storage unit 240 and the improvement request receiving unit 255. Based on the received improvement request information sent to the learning device 100, the parameter selection unit 260 selects the production parameter to be adjusted from one or more production parameters in the target process.

[0066] The quality evaluation prediction unit 265 is connected to the prediction model storage unit 240 and the parameter selection unit 260. The quality evaluation prediction unit 265 predicts the quality evaluation of downstream products produced in downstream processes using each product of the target process, after adjusting the production parameters selected by the parameter selection unit 260.

[0067] Figure 3 This represents an example of production data stored in the production data storage unit 215 of this embodiment. The production data includes the correspondence between each product of the target process and the product of the closest upstream process used in the production of the product of the target process, as well as one or more production parameters related to the production of the product of the target process.

[0068] In this embodiment, production data is stored for each of multiple entries: product batch identification information (product batch ID) for identifying the product (batch) of the target process; supply source batch identification information (supply source batch ID) and supply source identification information (supply source ID) for identifying each of one or more products supplied to the target process from the nearest upstream process; and one or more production parameters used in the production of the product in the target process.

[0069] This example illustrates production data stored in the production data storage unit 215 of the learning device 100d located in manufacturer 20d. In this example, using upstream products with batch ID "G0000001" supplied by manufacturer 20 (e.g., manufacturer 20g) directly upstream of supplier ID "G", upstream products with batch ID "H0000001" supplied by manufacturer 20 (e.g., manufacturer 20h) directly upstream of supplier ID "H", and upstream products with batch ID "I0000001" supplied by manufacturer 20 (e.g., manufacturer 20i) directly upstream of supplier ID "I", the product with batch ID "D0000001" in the target process is produced through a production process based on production parameters stored in the "Production Parameters" column. That is, the product with batch ID "D0000001" is a component manufactured using raw materials from manufacturer 20g identified by "G0000001", raw materials from manufacturer 20h identified by "H0000001", and raw materials from manufacturer 20i identified by "I0000001".

[0070] Two or more products in an object process can be produced using the same products (e.g., products from the same batch) in an upstream process, while one product in an object process (e.g., a batch of products) can be produced using different products from an upstream process. In the example shown in the diagram, of the product (batch) represented by "D0000001", a portion is produced using a group of upstream product (batch) represented by "G0000001", "H0000001", and "I0000001", and the remaining portion is produced using a group of upstream product (batch) represented by "G0000001", "H0000001", and "I0000002".

[0071] Figure 4 The operation flow of the learning device 100 in this embodiment is shown. In step S400, the production management unit 210 records production data in the production data storage unit 215 based on the production of each product in the target process, of which the learning device 100 is the target. This production data includes at least one production parameter related to the production of each product in the target process, and the correspondence between each upstream product (materials or components, etc., supplied from the upstream side of the target process) and each product in the target process. Here, the learning device 100 in the downstream process may also record the correspondence between the quality evaluation of each upstream product (materials or components, etc., of the target process) and each product in the target process in the production data storage unit 215.

[0072] In S410, the correspondence sending unit 220 sends the correspondence of the target process of the learning device 100 stored in the production data storage unit 215 to the evaluation device 110. In S425, the correspondence receiving unit 225 receives from the evaluation device 110 the correspondence between the quality evaluation of each product of the target process and each downstream product in the downstream process.

[0073] In S440, the learning processing unit 235 uses at least one production parameter related to the production of each product in the target process, and the quality evaluation of each downstream product produced using each product in the target process, to generate a predictive model that infers the quality evaluation of downstream products based on at least one production parameter. Here, at least a portion of the production parameter can be any control parameter that can be set in the equipment 30, such as the processing apparatus used in the target process, including raw material allocation, reaction temperature, reaction time, and pressure. For example, it can be measurement data measured by sensor devices such as pressure gauges, flow meters, and temperature sensors installed in the factory. Furthermore, the production parameter is not limited to control parameters; it can also be any parameter related to the four elements of production known as the "4Ms": "Material," "Machine," "Method," and "Man."

[0074] The learning processing unit 235 can use parameters obtained directly from the target process as production parameters, or it can process the parameters obtained directly from the target process and use them as production parameters. For example, the learning processing unit 235 can also use general characteristic quantities of the parameters (such as characteristic quantities representing the average or variance of the trend of parameter change), characteristic quantities specific to the target process (such as parameter changes or fluctuations at the beginning of the process), or factors that affect the quality of the product of the target process as defined by the manufacturer 20 responsible for the target process.

[0075] The quality evaluation of downstream products can be quantified as a score from 0 to 100. Alternatively, the quality evaluation of downstream products can also be expressed as a rating system such as A, B, or C.

[0076] Once learned, the predictive model, given one or more production parameters used in product manufacturing, outputs a predicted value for the quality assessment of downstream products produced using those parameters. The predictive model, as an example, utilizes various machine learning algorithms, including neural networks, support vector machines (SVMs), random forests, gradient boosting, or logistic regression, and learns in a way that predicts appropriate outputs for the provided inputs.

[0077] The learning processing unit 235 uses a combination of learning input data and learning labels as learning data. The learning input data includes at least one production parameter related to the production of each product in the target process, and the learning labels include quality evaluations of each downstream product produced using each product in the target process. The learning processing unit 235 updates the learning target parameters of the inference model to reduce the error between the inferred quality evaluation value and the learning labels. The inferred quality evaluation value is the output of the inference model when the learning input data from each sample in the learning data is input into the inference model. For example, when using a neural network as the inference model, the learning processing unit 235 uses the error between the output value of the neural network based on each input sample and the label, and adjusts the weights and biases of each neuron in the neural network through methods such as backpropagation.

[0078] Here, the quality of downstream products may be affected by the production parameters used in each process from the upstream to the downstream. However, the learning processing unit 235 generates a predictive model that does not refer to production parameters related to the target process among the multiple processes constituting the supply chain, but only predicts the quality evaluation of the downstream products based on at least one production parameter related to the target process.

[0079] The learning processing unit 235 can use methods such as mimic modeling to transform the speculative model generated in the above manner into another speculative model, and then use the transformed speculative model in subsequent steps. Mimic modeling is a method of transforming a machine learning model into a simpler machine learning model using a simpler machine learning approach, and it is effective when the input data (multiple production parameters in this embodiment) is complex (when the quantity is large and they are interdependent).

[0080] In S450, the calculation unit 245 calculates a model evaluation of the inferred model generated by the learning processing unit 235. The model evaluation calculated by the calculation unit 245 can be based on at least one of the accuracy or complexity of the inferred model.

[0081] When using the accuracy of the inference model in model evaluation, the calculation unit 245 calculates an index representing the accuracy of the inference model. This index represents the correct answer rate or output error of the inference model relative to the learning data, indicating the degree to which the inference model can accurately predict the quality evaluation or the degree of error in predicting the quality evaluation. The calculation unit 245 can use the accuracy of the inference model itself as the model evaluation, or it can use the accuracy of the inference model as a factor in the model evaluation.

[0082] The calculation unit 245 can also calculate the accuracy of the inference model through cross-validation. That is, the learning processing unit 235 can generate the inference model by performing learning processing using a portion of the learning data, and the calculation unit 245 can evaluate the accuracy of the inference model using the other portion of the learning data. Furthermore, in this embodiment, the accuracy of the inference model is a value where a larger value indicates a higher correct answer rate or a smaller output error. Alternatively, the accuracy of the inference model could also be a value where a smaller value indicates a higher correct answer rate or a smaller output error.

[0083] The computation unit 245 can also use the complexity of the inferential model in model evaluation. The inventors of this application have discovered a rule of thumb that can explain the relationship between various production parameters and the quality of the final product in a relatively simple way. This is because if a certain production parameter affects the quality of the final product, the quality of the downstream product varies according to that production parameter, independent of the values ​​of other production parameters. The inferential model is conceived to be simplistically designed to infer the quality of the downstream product based solely on the production parameter that affects the quality of the downstream product among all production parameters.

[0084] Therefore, the calculation unit 245 can also use the complexity of the inferred model to calculate the model evaluation. The calculation unit 245 can use indicators such as Radmach complexity, VC dimension, or generalization error limit, which represent the degree of complexity of a formula used to explain the relationship between input and output, as the complexity of the inferred model. The calculation unit 245 can use the complexity of the inferred model itself as the model evaluation, or it can use the complexity of the inferred model as a factor in the model evaluation. Furthermore, in this embodiment, a higher value of the inferred model complexity indicates a more complex inferred model and a lower model evaluation. Alternatively, a lower value of the inferred model complexity could also indicate a more complex inferred model.

[0085] In this embodiment, the calculation unit 245 increases the model evaluation value in at least one of the following situations: the prediction model has high accuracy due to a high correct answer rate, or the prediction model is simple due to low complexity. Conversely, it decreases the model evaluation value in at least one of the following situations: the prediction model has low accuracy due to a low correct answer rate, or the prediction model is complex due to high complexity. Alternatively, the calculation unit 245 may generate a smaller model evaluation value when the prediction model has high accuracy or when the prediction model is simple.

[0086] The model evaluation generated in the above manner is an indicator of the evaluation of the learned predictive model itself, but it can also be used as an indicator of the extent to which the production parameters of the target process affect the quality of the final product. In S450, the model evaluation sending unit 250 sends this model evaluation calculated by the calculation unit 245 to the evaluation device 110.

[0087] In S460, the improvement request receiving unit 255 determines whether an improvement request information has been received from the evaluation device 110. If no improvement request information is received, the learning device 100 advances the process to S400, adds the newly produced product, and performs the processes from S400 to S450. If an improvement request information is received, the learning device 100 advances the process to S470.

[0088] In S470, the parameter selection unit 260 selects at least one production parameter from one or more production parameters in the target process that should be adjusted according to the improvement request. Here, if the factor parameters that are factors contributing to quality changes among the one or more production parameters can be directly extracted from the speculative model generated in S440, the parameter selection unit 260 can directly extract the factor parameters from the speculative model. For example, if the speculative model uses sparse learning and the relationship between input and output is clear, or if the speculative model uses a neural network and multiple production parameters can be arranged in order of their impact on quality according to the weights assigned to the propagation paths from each input layer to each output layer, the parameter selection unit 260 can directly extract the factor parameters from the speculative model.

[0089] When factor parameters cannot be directly extracted, the parameter selection unit 260 uses an inference model to generate pseudo-data that is widely distributed within a space mapping multiple production parameters. Alternatively, the parameter selection unit 260 can use a Generative Adversarial Network (GAN) to generate pseudo-data containing multiple production parameters and quality evaluation. Then, using the generated pseudo-data, the parameter selection unit 260 learns and generates a machine learning model capable of factor inference such as sparse learning to extract factor parameters that influence quality from the multiple production parameters.

[0090] The parameter selection unit 260 can extract production parameters whose impact on quality exceeds a predetermined threshold as factor parameters. Alternatively, the parameter selection unit 260 can also sequentially extract a predetermined number of production parameters as factor parameters, starting with the production parameters with the largest impact on quality.

[0091] In S480, after adjusting the production parameters selected as factor parameters, the quality evaluation prediction unit 265 predicts how the quality evaluation of each product of the target process will change in the downstream products produced in the downstream process. The quality evaluation prediction unit 265 can generate a scatter plot (one-dimensional or multi-dimensional) of one or more factor parameters, which represents the pass / fail status of downstream products or the boundaries of the downstream product's grade.

[0092] Furthermore, the quality evaluation prediction unit 265 can also determine how much to shift the factor parameters in order to improve the quality of downstream products. For example, the quality evaluation prediction unit 265 can determine how to change the factor parameters so that the quality evaluation of downstream products produced using those factor parameters can exceed the boundary of the level in the scatter plot and further improve the level. In addition, the quality evaluation prediction unit 265 can also use the prediction model stored in the prediction model storage unit 240 to predict the quality evaluation of downstream products when the factor parameters are changed.

[0093] The quality evaluation prediction unit 265 can generate the aforementioned scatter plot or a report containing at least one of the displacement amounts of factor parameters used to improve quality, and provide it to the user of the learning device 100. Thus, the learning device 100 can input instructions from the user who has referred to the report to change the displacement amount of the value of the factor parameter supplied to the device 30. Furthermore, the quality evaluation prediction unit 265 can also perform the processing of changing the displacement amount of the value of the factor parameter used by the device 30 via the production management unit 210.

[0094] Furthermore, in S460, when a specific downstream product becomes the object of an improvement request, the improvement request receiving unit 255 can also receive improvement request information, which includes the product used in the target process for that specific downstream product and the quality evaluation of that downstream product. In this case, in S480, the parameter selection unit 260 can also determine how much to shift the factor parameter in order to improve the quality of that specific downstream product.

[0095] According to the learning device 100 described above, by sending the correspondence between each product supplied from the upstream side to the target process and each product of the target process to the evaluation device 110, each learning device 100 can receive the correspondence between the quality evaluation of the target process's products and the downstream products. Furthermore, each learning device 100 can independently calculate a model evaluation, which serves as an indicator of the degree to which the production parameters of the target process affect the quality of the downstream products, independent of learning devices 100 corresponding to other processes. Therefore, each learning device 100 can achieve the system environment of evaluating a process without disclosing the production parameters to other learning devices 100 or the evaluation device 110.

[0096] Figure 5This describes the configuration of the evaluation device 110 in this embodiment. The evaluation device 110 can be a portable computer such as a PC (personal computer), workstation, server computer, or general-purpose computer, tablet computer, or smartphone, or it can be a computer system connected to multiple computers. Such a computer system is also a computer in a broad sense. The evaluation device 110 can be implemented by executing an evaluation program on such a computer. Furthermore, the evaluation device 110 can be implemented by one or more virtual computer environments that can be executed within the computer, or it can be implemented by a cloud computing system. Alternatively, the evaluation device 110 can be a dedicated computer designed for evaluating the target processes of each learning device 100, or it can be dedicated hardware implemented by dedicated circuitry.

[0097] The evaluation device 110 includes: a correspondence acquisition unit 500, a correspondence generation unit 510, a correspondence data storage unit 520, a correspondence transmission unit 530, a start determination unit 550, a model evaluation receiving unit 560, a process evaluation unit 570, and an information output unit 580. The correspondence acquisition unit 500 receives, for one or more upstream processes included in the supply chain, correspondences from one or more learning devices 100 between products supplied from directly upstream processes and products supplied to directly downstream processes. Furthermore, for one or more downstream processes, the correspondence acquisition unit 500 receives, from one or more product evaluation devices 105, correspondences of quality evaluations between products supplied from directly upstream processes and each downstream product.

[0098] The correspondence generation unit 510 is connected to the correspondence acquisition unit 500. Using the correspondences acquired by the correspondence acquisition unit 500, the correspondence generation unit 510 generates correspondences for quality evaluation between products from one or more upstream processes and products from one downstream process. The correspondence data storage unit 520 is connected to the correspondence generation unit 510. The correspondence data storage unit 520 stores the correspondence data generated by the correspondence generation unit 510, i.e., the correspondence data. The correspondence sending unit 530 is connected to the correspondence data storage unit 520. The correspondence sending unit 530 sends the correspondences generated by the correspondence generation unit 510 regarding products from each upstream process to the learning device 100 corresponding to each upstream process.

[0099] The start determination unit 550 is connected to the correspondence data storage unit 520. The start determination unit 550 receives quality evaluations of at least one downstream product from the correspondence data storage unit 520, and based on the quality evaluations of at least one downstream product, determines whether to start the evaluation of at least one upstream process by the process evaluation unit 570.

[0100] The model evaluation receiving unit 560 receives a model evaluation of the inference model from the learning device 100 for each of at least one upstream process. The learning device 100 generates an inference model that infers the quality evaluation of the downstream product based on at least one production parameter in that process through learning. Here, if a learning device 100 is also provided in each of the downstream processes, the model evaluation receiving unit 560 can also receive a model evaluation of the inference model from the learning device 100 of the downstream process.

[0101] The process evaluation unit 570 is connected to the start determination unit 550 and the model evaluation receiving unit 560. Upon receiving an instruction from the start determination unit 550 to begin the evaluation of at least one upstream process, the process evaluation unit 570 evaluates at least one upstream process based on model evaluations for each of the at least one upstream process. Here, if model evaluations of the inferred model are also received from the learning device 100 of the downstream process, the process evaluation unit 570 may evaluate at least one upstream process and the downstream process based not only on model evaluations for each of the at least one upstream process but also on model evaluations for the downstream process.

[0102] The information output unit 580 is connected to the process evaluation unit 570. When there is an upstream or downstream process that is determined to need improvement, the information output unit 580 sends the improvement request information of the process that is determined to need improvement to the learning device 100 corresponding to the process that is determined to need improvement.

[0103] Figure 6 This describes the operation flow of the evaluation device 110 in this embodiment. In S600, the correspondence acquisition unit 500 receives from one or more learning devices 100 the correspondence between each product supplied from the upstream process in each target process and each product supplied to the downstream process. Here, the correspondence acquisition unit 500 receives from the learning device 100 corresponding to the downstream process the correspondence between each product supplied from the upstream process in the downstream process and the quality evaluation of each downstream product.

[0104] In S610, the correspondence generation unit 510 uses the correspondences obtained by the correspondence acquisition unit 500 to generate a correspondence for the quality evaluation of each product in one or more upstream processes and each product in one downstream process. (This is followed by...) Figure 7 The method for generating correspondences in the correspondence generation unit 510 will be explained in connection with this.

[0105] In S620, the correspondence sending unit 530 sends the correspondence generated by the correspondence generation unit 510 to the learning device 100 corresponding to each upstream process. Alternatively, the correspondence sending unit 530 may extract and send the correspondences related to downstream products manufactured using the product from each upstream process from all the correspondences for the learning device 100 corresponding to each upstream process.

[0106] In S640, the start determination unit 550 determines whether to start process evaluation. In this embodiment, the start determination unit 550 determines whether to start process evaluation based on the quality evaluation of at least one downstream product obtained by the correspondence acquisition unit 500 and stored in the correspondence relationship data storage unit 520. The start determination unit 550 may, for each manufacturer 20a to c, start process evaluation associated with that downstream process based on the quality evaluation of downstream products produced at the closest time, the average of the quality evaluations of two or more downstream products produced within the closest predetermined period (e.g., 1 hour), the average of the quality evaluations of a predetermined number of downstream products produced recently, or an index value corresponding to the quality evaluations of other downstream products being less than a threshold.

[0107] Furthermore, the start determination unit 550 can also initiate process evaluation if the defect rate of recently produced downstream products exceeds a threshold. Alternatively, instead of evaluation device 110, product evaluation device 105 can also determine whether to initiate process evaluation, and product evaluation device 105 can also accept instruction input to initiate process evaluation.

[0108] Without initiating process evaluation, the evaluation device 110 can advance the process to S600 and continue collecting information such as obtaining new corresponding data (S600). Alternatively, the evaluation device 110 can perform the processes of S600 to S620 after determining that process evaluation has begun (i.e., if "yes" is set in S640).

[0109] When a process evaluation begins, the model evaluation receiving unit 560 receives model evaluations of the inference model from the learning device 100 for each of at least one upstream process and one downstream process that is the object of the process evaluation. The learning device 100 generates an inference model that infers the quality evaluation of the downstream product based on at least one production parameter in that process. In S660, the process evaluation unit 570 evaluates each process that is the object of the process evaluation based on the model evaluations for each process that is the object of the process evaluation.

[0110] Here, the process evaluation unit 570 determines that processes in at least one upstream and downstream process whose model evaluation values ​​are greater than the benchmark value should be improved. Thus, the process evaluation unit 570 can identify processes whose production parameters have a significant impact on the quality of the final product as processes that should be improved. The benchmark value used to determine whether a process should be improved can be, for example, an absolute value indicating that improvement is needed when "model evaluation > 1.0", or a relative value indicating that improvement is needed when "the model evaluation of a certain process > the sixth largest model evaluation" (i.e., when the model evaluation is up to the fifth largest).

[0111] When using model evaluation based on the accuracy of the speculative model, the process evaluation unit 570 can determine that an upstream or downstream process should be improved if a model evaluation showing that the accuracy of the speculative model is greater than the benchmark value. Furthermore, when using model evaluation based on the complexity of the speculative model, the process evaluation unit 570 can determine that an upstream or downstream process should be improved if a model evaluation showing that the complexity of the speculative model is less than the benchmark value.

[0112] In S670, if there is no process to be improved ("No" in the figure), the evaluation device 110 advances the process to S600 to continue collecting information such as obtaining new corresponding information (S600). In S670, if there is a process to be improved ("Yes" in the figure), in S680, the information output unit 580 outputs improvement request information for at least one of the upstream or downstream processes determined to be to be improved to the learning device 100 located in the process to be improved. Here, if a specific downstream product becomes the object of an improvement request, the information output unit 580 may also send improvement request information, which includes the product of the target process used in that specific downstream product and a quality evaluation of that downstream product. Then, the evaluation device 110 advances the process to S600.

[0113] According to the evaluation device 110 described above, it can receive from each learning device 100 the correspondence between each product supplied from the upstream side for each process and the products of that process, and generate a correspondence between the products of each process and the downstream products, including the products of each upstream process that has passed through one or more other processes up to the downstream process. Thus, even if it is not a process directly upstream of the downstream process, the evaluation device 110 can determine in what quality evaluation each product of each process will ultimately be used in the production of the downstream product. Furthermore, the evaluation device 110 can use the model evaluation received from each learning device 100 to determine the processes that should be improved, without needing to obtain the production parameters of each process.

[0114] Figure 7This diagram illustrates an example of the correspondence between products produced at each stage and the final product. In this example, raw material manufacturer 20j produces product J0000001. Correspondingly, learning device 100j stores production data in production data storage unit 215. Figure 3 The "Product Batch ID" field is set to "J0000001", the "Supply Source Batch ID" and "Supply Source ID" fields are left blank, and the "Production Parameters" field contains the values ​​of the production parameters used in the production of product J0000001. Similarly, raw material manufacturer 20k produces product K0000001, and raw material manufacturer 20l produces product L0000001.

[0115] Raw material manufacturer 20g uses products J0000001, K0000001, and L0000001 as raw materials to produce product G0000001. Correspondingly, learning device 100g stores production data in production data storage unit 215, and this production data will... Figure 3 The "Product Batch ID" field is designated as "G0000001", the "Supply Source Batch ID" field as "J0000001, K0000001, L0000001", the "Supply Source ID" field as "J, K, L", and the "Production Parameters" field as the values ​​of the production parameters used in the production of product G0000001. Raw material manufacturer 20h and raw material manufacturer 20i also use the necessary raw materials to produce products H0000001 and I0000001, respectively.

[0116] Component manufacturer 20d uses products G0000001, H0000001, and L0000001 as raw materials to produce product D0000001. Correspondingly, learning device 100d stores production data in production data storage unit 215, and this production data will... Figure 3 The "Product Batch ID" field is designated as "D0000001", the "Supply Source Batch ID" field as "G0000001, H0000001, I0000001", the "Supply Source ID" field as "G, H, I", and the "Production Parameters" field as the values ​​of the production parameters used in the production of product D0000001. Component manufacturer 20e and component manufacturer 20f also use the necessary raw materials to produce products E0000001 and F0000001, respectively.

[0117] Product manufacturer 20a uses products D0000001, E0000001, and F0000001 as components to produce product A0000001. Correspondingly, product evaluation device 105a stores production data, which will... Figure 3The “Product Batch ID” field is set to “A0000001”, the “Supply Source Batch ID” field is set to “D0000001, E0000001, F0000001”, the “Supply Source ID” field is set to “D, E, F”, and the “Production Parameters” field is set to the values ​​of the production parameters used in the production of product A0000001.

[0118] In the example shown in the figure, the learning device 100g sends the corresponding products J0000001, K0000001, and L0000001 from the upstream side to the evaluation device 110, along with the product G0000001 supplied to the downstream side. The learning device 100d sends the corresponding products G0000001, H0000001, and I0000001 from the upstream side to the evaluation device 110, along with the product D0000001 supplied to the downstream side. The product evaluation device 105a sends the corresponding products D0000001, E0000001, and F0000001 from the upstream side to the evaluation device 110, along with the quality evaluation of product A0000001.

[0119] The learning devices 100d-f and the product evaluation device 105a receive correspondences of products from each process. The correspondence generation unit 510 of the evaluation device 110 generates a correspondence between the quality evaluations of the products in each process and the products in the downstream process. In the example shown in this figure, the correspondence received from the learning device 100a indicates that product D0000001 was used in the production of product A0000001. Therefore, the correspondence generation unit 510 determines that the quality evaluation of product D0000001 corresponds to that of product A0000001.

[0120] Furthermore, the correspondence received from the learning device 100a indicates that product D0000001 was used in the production of product A0000001, and the correspondence received from the learning device 100d indicates that product G0000001 was used in the production of product D0000001. Therefore, the correspondence generation unit 510 determines that the quality evaluation of product G0000001 corresponds to that of product A0000001.

[0121] Therefore, the correspondence generation unit 510 further determines, for products such as D0000001 used in the production of the final product A0000001, the upstream products such as G0000001 used in the production of the final product D0000001, and by tracing further upstream, can generate a correspondence that products D0000001 to L0000001 in each upstream process are products used in the production of the final product A0000001.

[0122] Correspondence generation unit 510 can also generate Figure 7The graphical structure shown uses each product in each process as a node and the correspondence between products as edges. Furthermore, when there is a path from the node corresponding to product X in a certain upstream process along one or more edges to the node corresponding to the final product Y, the correspondence generation unit 510 can generate a correspondence between product X and the final product Y for production.

[0123] In addition to the functions described above, the evaluation device 110 may also include the function of detecting whether the quality evaluation of downstream products has increased or the amount of increase in the quality evaluation after the learning device 100, which has received improvement request information, has improved at least one factor parameter. For example, the evaluation device 110 may use the average quality evaluation of downstream products or the yield rate, etc., to detect whether the quality evaluation has increased. Furthermore, the evaluation device 110 may also issue incentives for the quality improvement of downstream products accompanying the factor parameter improvement, such as by assigning points or encrypting assets, based on whether the quality evaluation of downstream products has increased compared to before the improvement or whether the increase in the quality evaluation exceeds a threshold.

[0124] Furthermore, the evaluation system 10 described above can be used not only for improving the quality of the final product, but also for reducing the production cost of the final product. In the case of reducing the production cost of the final product, the process evaluation unit 570 of the evaluation device 110 can determine that processes whose model evaluation is smaller than the benchmark value should be improved. Thus, the process evaluation unit 570 can determine that processes whose production parameters will not have a significant impact on the quality of the final product are processes for which changes would not be problematic in order to reduce production costs.

[0125] Correspondingly, the parameter selection unit 260 of the learning device 100 selects at least one production parameter that can be changed to reduce production costs from one or more production parameters in the target process according to the improvement request. The parameter selection unit 260 and... Figure 4 The operation shown in S470 is the opposite. Among one or more production parameters, a production parameter whose impact on quality is less than a predetermined threshold is selected as a changeable production parameter. When the selected changeable production parameter is adjusted, the quality evaluation prediction unit 265 predicts how the quality evaluation of each product used in the target process will change in the downstream products produced in the downstream processes, and to what extent the production cost of the target process products can be reduced. The quality evaluation prediction unit 265 can change the selected changeable production parameter to suppress the reduction in the quality evaluation of downstream products to less than the threshold, and to make the cost reduction effect exceed the threshold.

[0126] Various embodiments of the present invention can be described with reference to flowcharts and block diagrams, where a module may represent (1) a stage of the process of performing an operation or (2) a part of a device that performs the operation. Specific stages and parts may be implemented by dedicated circuitry, programmable circuitry supplied together with computer-readable instructions stored on a computer-readable medium, and / or a processor supplied together with computer-readable instructions stored on a computer-readable medium. Dedicated circuitry may include digital and / or analog hardware circuitry, and may also include integrated circuits (ICs) and / or discrete circuitry. Programmable circuitry may include reconfigurable hardware circuitry, including logic AND, logic OR, logic XOR, logic NAND, logic NOR and other logic operations, flip-flops, registers, field-programmable gate arrays (FPGAs), programmable logic arrays (PLAs), and other memory elements.

[0127] Computer-readable media can include any tangible device capable of storing instructions executable by a suitable device. Consequently, a computer-readable medium having instructions stored therein includes an article containing instructions executable by means of a flowchart or block diagram. Examples of computer-readable media include: electronic storage media, magnetic storage media, optical storage media, electromagnetic storage media, semiconductor storage media, etc. More specific examples of computer-readable media include: floppy disks, magnetic disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), electrically erasable programmable read-only memory (EEPROM), static random access memory (SRAM), optical disc read-only memory (CD-ROM), digital multi-purpose disc (DVD), Blu-ray disc, memory stick, integrated circuit card, etc.

[0128] Computer-readable instructions include any one of source code and object code described by any combination of one or more programming languages, including assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or existing procedural programming languages ​​such as Smalltalk (registered trademark), JAVA (registered trademark), C++, and "C" or similar programming languages.

[0129] Computer-readable instructions can be provided via a local area network (LAN), wide area network (WAN) such as the Internet, to the processor or programmable circuit of a programmable data processing device such as a general-purpose computer, a special-purpose computer, or other computers, and are executed in order to create means for performing the operations specified by a flowchart or block diagram. Examples of processors include: computer processors, processing units, microprocessors, digital signal processors, controllers, microcontrollers, etc.

[0130] Figure 8 Examples of computer 2200 that can implement the present invention wholly or partially are shown. Through programs installed on computer 2200, computer 2200 can perform operations associated with an apparatus or one or more parts of that apparatus as an embodiment of the present invention, or execute that operation or those one or more parts, and / or computer 2200 can execute processes or stages of embodiments of the present invention. To enable computer 2200 to perform specific operations associated with several or all of the modules in the flowcharts and block diagrams described in this specification, such programs can be executed by CPU 2212.

[0131] The computer 2200 of this embodiment includes a CPU 2212, RAM 2214, a graphics controller 2216, and a display device 2218, which are interconnected via a main controller 2210. The computer 2200 also includes input / output units such as a communication interface 2222, a hard disk drive 2224, a DVD-ROM drive 2226, and an IC card drive, which are connected to the main controller 2210 via an input / output controller 2220. The computer also includes conventional input / output units such as a ROM 2230 and a keyboard 2242, which are connected to the input / output controller 2220 via an input / output chip 2240.

[0132] CPU 2212 operates according to the program stored in ROM 2230 and RAM 2214, thereby controlling each unit. Graphics controller 2216 acquires image data generated by CPU 2212 from frame buffers or other storage provided in RAM 2214 or from its own storage, and displays the image data on display device 2218.

[0133] Communication interface 2222 enables communication with other electronic devices via a network. Hard disk drive 2224 stores programs and data used by CPU 2212 within computer 2200. DVD-ROM drive 2226 reads programs or data from DVD-ROM 2201 and provides programs or data to hard disk drive 2224 via RAM 2214. IC card drive reads programs and data from IC card and / or writes programs and data to IC card.

[0134] ROM 2230 stores a boot program and / or programs that depend on the hardware of computer 2200 and are executed by computer 2200 when activated. Input / output chip 2240 can also connect various input / output units to input / output controller 2220 via parallel port, serial port, keyboard port, mouse port, etc.

[0135] The program is provided by a computer-readable medium such as a DVD-ROM 2201 or an IC card. The program is read from the computer-readable medium and installed in a hard disk drive 2224, RAM 2214, or ROM 2230, which are also examples of computer-readable media, and executed by the CPU 2212. The information processing described within these programs is read into the computer 2200, thereby enabling cooperation between the program and the aforementioned various types of hardware resources. An apparatus or method can be constructed to perform the manipulation or processing of information by using the computer 2200.

[0136] For example, when communication is performed between computer 2200 and an external device, CPU 2212 can execute a communication program loaded in RAM 2214, and instruct communication interface 2222 to perform communication processing based on the processing described in the communication program. Under the control of CPU 2212, communication interface 2222 reads transmission data stored in a transmission buffer processing area provided in a recording medium such as RAM 2214, hard disk drive 2224, DVD-ROM 2201, or IC card, sends the read transmission data to the network, or writes received data received from the network to a receive buffer processing area provided on the recording medium, etc.

[0137] Furthermore, the CPU 2212 can read all or necessary portions of files or databases stored on external recording media such as hard disk drive 2224, DVD-ROM drive 2226 (DVD-ROM 2201), IC cards, etc., into RAM 2214, and perform various types of processing on the data in RAM 2214. Then, the CPU 2212 writes the processed data back to the external recording media.

[0138] Information of various types, such as programs, data, tables, and databases, can be stored in recording media and processed. CPU 2212 performs various types of processing described throughout this disclosure on data read from RAM 2214 and writes the results back to RAM 2214. These various types of processing include operations specified by a sequence of program instructions, information processing, conditional judgments, conditional branches, unconditional branches, information retrieval / replacement, etc. Furthermore, CPU 2212 can retrieve information from files, databases, etc., within the recording medium. For example, when multiple entries, each having an attribute value associated with a second attribute, are stored in the recording medium, CPU 2212 can retrieve from these multiple entries an entry that matches a condition specifying the attribute value of the first attribute, and read the attribute value of the second attribute stored in that entry, thereby obtaining the attribute value of the second attribute associated with the first attribute that satisfies a predetermined condition.

[0139] The programs or software modules described above can be stored on or near the computer 2200 on a computer-readable medium. Furthermore, recording media such as hard disks or RAM provided in a server system connected to a dedicated communication network or the Internet can be used as computer-readable media, thereby providing the program to the computer 2200 via the network.

[0140] The present invention has been described above using embodiments, but the technical scope of the present invention is not limited to the scope described in the above embodiments. It will be apparent to those skilled in the art that various modifications or improvements can be made to the above embodiments. As can be seen from the claims, such modifications or improvements may also be included within the technical scope of the present invention.

[0141] The execution order of actions, processes, steps, and stages in the apparatus, systems, programs, and methods shown in the claims, description, and drawings is not specifically stated as "earlier" or "before." Furthermore, it should be noted that any order is permissible as long as the output of the preceding process is not used in the subsequent process. Even if the flow of actions in the claims, description, and drawings is described using terms such as "firstly," "next," etc., for ease of explanation, it does not imply that the actions must be performed in that order.

Claims

1. A learning device, characterized in that... include: The correspondence receiving unit receives the correspondence between the quality evaluations of each product of the target process and each downstream product produced in the downstream process using the products of the target process. The learning processing unit uses at least one production parameter for the production of each product in the target process and the quality evaluation of each downstream product produced using each product in the target process to generate a predictive model that infers the quality evaluation of the downstream products based on the at least one production parameter. The computational unit calculates a model evaluation based on at least one of the accuracy or complexity of the predicted model. as well as The model evaluation sending unit sends the model evaluation calculated by the calculation unit to the evaluation device, which evaluates the at least one upstream process using the model evaluation for each of the at least one upstream process that is upstream of the downstream process.

2. The learning device according to claim 1, characterized in that, It also includes an improvement request receiving unit, which receives improvement request information sent by the evaluation device, which determines the target process to be improved based on model evaluation for each of the at least one upstream process and sends improvement request information accordingly.

3. The learning device according to claim 2, characterized in that, It also includes a parameter selection unit, which selects the production parameter to be adjusted from at least one production parameter in the target process based on the received improvement request information.

4. The learning device according to claim 3, characterized in that, It also includes a quality evaluation prediction unit, which predicts the quality evaluation of downstream products produced in downstream processes using each product of the target process when the production parameters selected by the parameter selection unit are adjusted.

5. An evaluation device, characterized in that... include: The model evaluation receiving unit receives, for each of at least one upstream process, a model evaluation based on at least one of the accuracy or complexity of the inference model, from a learning device that generates an inference model by learning and using at least one production parameter for the production of each upstream product of that upstream process and a quality evaluation of each downstream product produced using each upstream product of that upstream process, and inferring the quality evaluation of the downstream product based on the at least one production parameter. as well as The process evaluation department evaluates the at least one upstream process based on the model evaluation for each of the at least one upstream process.

6. The evaluation device according to claim 5, characterized in that, The process evaluation department determines that an upstream process in at least one upstream process for which the accuracy of the predicted model is greater than the benchmark value or the complexity is less than the benchmark value should be improved.

7. The evaluation device according to claim 6, characterized in that, It also includes an information output section that outputs improvement request information for upstream processes that are determined to require improvement.

8. The evaluation apparatus according to any one of claims 5 to 7, characterized in that, Also includes: The corresponding acquisition unit acquires, for each of the at least one upstream process, the correspondence between each product supplied from the upstream process and each product supplied to the downstream process. The correspondence generation unit uses the correspondences obtained by the correspondence acquisition unit to generate the correspondence between the quality evaluation of each product of the at least one upstream process and each downstream product. as well as The correspondence sending unit sends the correspondence to the learning device.

9. The evaluation device according to any one of claims 5 to 7, characterized in that, It also includes a start determination unit, which determines whether to begin the evaluation of the at least one upstream process by the process evaluation unit based on the quality evaluation of at least one downstream product.

10. An evaluation system, characterized in that... include: At least one learning device as described in any one of claims 1 to 4, with each of at least one upstream process as the target process; as well as The evaluation device as described in any one of claims 5 to 7.

11. A learning method, characterized in that, The learning device receives the correspondence between the quality evaluations of each product of the object process and each downstream product produced in the downstream process using the products of the object process. The learning device uses at least one production parameter for the production of each product in the target process, and quality evaluations of each downstream product produced using the products in the target process, to generate a predictive model that infers the quality evaluations of downstream products based on the at least one production parameter. The learning device calculates a model evaluation based on at least one of the accuracy or complexity of the inferred model. The learning device sends a calculated model evaluation to the evaluation device, which evaluates the at least one upstream process using model evaluations for each of at least one upstream process that is upstream of the downstream process.

12. A recording medium containing a learning program, characterized in that, The computer functions as a correspondence receiving unit, a learning processing unit, a calculation unit, and a model evaluation sending unit by executing the learning program. The correspondence receiving unit receives the correspondence between the quality evaluations of each product of the target process and each downstream product produced in the downstream process using the products of the target process. The learning processing unit uses at least one production parameter for the production of each product in the target process, and the quality evaluation of each downstream product produced using each product in the target process, to generate a predictive model that infers the quality evaluation of downstream products based on the at least one production parameter. The computing unit calculates a model evaluation based on at least one of the accuracy or complexity of the inferred model. The model evaluation sending unit sends the model evaluation calculated by the calculation unit to the evaluation device, and the evaluation device evaluates the at least one upstream process using the model evaluation for each of the at least one upstream process that is upstream of the downstream process.

13. An evaluation method, characterized in that, The evaluation device, for each of at least one upstream process, learns from at least one production parameter relating to the production of each upstream product of that upstream process and the quality evaluation of each downstream product produced using the upstream products of that upstream process, and generates a predictive model that infers the quality evaluation of the downstream products based on the at least one production parameter. It then receives a model evaluation based on at least one of the accuracy or complexity of the predictive model. The evaluation device evaluates the at least one upstream process based on the model evaluation for each of the at least one upstream process.

14. A recording medium for recording an evaluation procedure, characterized in that, The computer functions as both a model evaluation receiving unit and a process evaluation unit by executing the evaluation program. The model evaluation receiving unit receives, for each of at least one upstream process, a model evaluation based on at least one of the following: a learning device that generates a predictive model for predicting the quality evaluation of downstream products based on at least one production parameter relating to the production of each upstream product of that upstream process and the quality evaluation of each downstream product produced using each upstream product of that upstream process; and a model evaluation based on at least one of the accuracy or complexity of the predictive model. The process evaluation department evaluates the at least one upstream process based on the model evaluation for each of the at least one upstream process.

Images