WO2025215908A1 - 金属材料の品質予測モデル生成方法、金属材料の品質予測方法、金属材料の品質影響因子推定方法、金属材料の製造方法、品質予測モデル生成方法、金属材料の品質予測モデル生成装置、金属材料の品質予測装置および金属材料の品質影響因子推定装置 - Google Patents

金属材料の品質予測モデル生成方法、金属材料の品質予測方法、金属材料の品質影響因子推定方法、金属材料の製造方法、品質予測モデル生成方法、金属材料の品質予測モデル生成装置、金属材料の品質予測装置および金属材料の品質影響因子推定装置

Info

Publication number
WO2025215908A1
WO2025215908A1 PCT/JP2025/002222 JP2025002222W WO2025215908A1 WO 2025215908 A1 WO2025215908 A1 WO 2025215908A1 JP 2025002222 W JP2025002222 W JP 2025002222W WO 2025215908 A1 WO2025215908 A1 WO 2025215908A1
Authority
WO
WIPO (PCT)
Prior art keywords
quality
quality prediction
prediction model
predetermined range
metal material
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
PCT/JP2025/002222
Other languages
English (en)
French (fr)
Japanese (ja)
Inventor
宏征 高木
洋 中川
丈英 平田
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
JFE Steel Corp
Original Assignee
JFE Steel Corp
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by JFE Steel Corp filed Critical JFE Steel Corp
Priority to JP2025529959A priority Critical patent/JPWO2025215908A1/ja
Publication of WO2025215908A1 publication Critical patent/WO2025215908A1/ja
Pending legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B21—MECHANICAL METAL-WORKING WITHOUT ESSENTIALLY REMOVING MATERIAL; PUNCHING METAL
    • B21B—ROLLING OF METAL
    • B21B37/00—Control devices or methods specially adapted for metal-rolling mills or the work produced thereby
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B21—MECHANICAL METAL-WORKING WITHOUT ESSENTIALLY REMOVING MATERIAL; PUNCHING METAL
    • B21B—ROLLING OF METAL
    • B21B38/00—Methods or devices for measuring, detecting or monitoring specially adapted for metal-rolling mills, e.g. position detection, inspection of the product
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B21—MECHANICAL METAL-WORKING WITHOUT ESSENTIALLY REMOVING MATERIAL; PUNCHING METAL
    • B21B—ROLLING OF METAL
    • B21B99/00—Subject matter not provided for in other groups of this subclass
    • G—PHYSICS
    • G05—CONTROLLING; REGULATING
    • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
    • G05B19/00—Program-control systems
    • G05B19/02—Program-control systems electric
    • G05B19/418—Total factory control, i.e. centrally controlling a plurality of machines, e.g. direct or distributed numerical control [DNC], flexible manufacturing systems [FMS], integrated manufacturing systems [IMS] or computer integrated manufacturing [CIM]

Definitions

  • the present invention relates to a method for generating a quality prediction model for metallic materials, a method for predicting the quality of metallic materials, a method for estimating factors influencing the quality of metallic materials, a method for manufacturing metallic materials, a quality prediction model generation method, a device for generating a quality prediction model for metallic materials, a device for predicting the quality of metallic materials, and a device for estimating factors influencing the quality of metallic materials.
  • Patent Document 1 discloses a method for predicting material quality using a model trained on data that links manufacturing condition data during material production with quality data from the final process, taking into account the quality determination position in the longitudinal direction of the material.
  • Material quality can vary depending on one or more manufacturing conditions, and the manifestation of quality defects also varies depending on the manufacturing conditions that cause them. Therefore, in order to accurately predict quality defects or estimate the causes of quality defects with high accuracy, it is necessary to appropriately utilize information on the manufacturing conditions that correspond to the causes, depending on the type of quality defect.
  • the length of the metal material increases, mainly in the longitudinal direction, as it passes through each process. If the length units of the metal material in the final process are aligned, the length of the metal material in the corresponding intermediate process may become shorter than the measurement period for the actual manufacturing condition data of the intermediate process as you go back through the processes.
  • the manufacturing condition data for the intermediate process metal material is converted into data in units of length for the final process metal material by performing interpolation (for example, linear interpolation) to align it with the length units for the final process metal material whose quality will be evaluated.
  • a dataset is then generated that associates the converted manufacturing condition data for the intermediate process with the quality data and manufacturing condition data for the final process.
  • the present invention has been made in consideration of the above, and aims to provide a quality prediction model generation method for metallic materials, a quality prediction method for metallic materials, a method for estimating factors influencing quality of metallic materials, a manufacturing method for metallic materials, a quality prediction model generation method, a quality prediction model generation device for metallic materials, a quality prediction device for metallic materials, and a quality prediction device for metallic materials, which determine the coarseness (granularity) of data for aligning and associating manufacturing condition data and quality data for each manufacturing process based on specific indices including an index indicating the predictive accuracy of the quality prediction model and an importance index indicating the impact on the quality prediction of the final process, and by appropriately utilizing the granularity of the manufacturing conditions for each process, it is possible to accurately predict the quality of a material for any manufacturing condition and accurately estimate the causes of quality defects.
  • the method for generating a quality prediction model for metallic materials includes a first collection step of collecting the manufacturing conditions for metallic materials in each process for each predetermined range of the metallic material defined for each process, a second collection step of evaluating and collecting the quality of the metallic material in the final process of each of the processes for each predetermined range defined for the final process, and a second collection step of evaluating and collecting the manufacturing conditions for the metallic material in each process collected in the first collection step and the quality of the metallic material in the final process collected in the second collection step, based on the width of the predetermined range for a specific process among the processes, and comparing the manufacturing conditions and quality in other processes.
  • a model generation step that generates one or more quality prediction models for each unified predetermined range from the manufacturing conditions for each unified predetermined range saved in the storage step, and that predicts the quality for each unified predetermined range; and, if multiple quality prediction models are generated in the model generation step, a model selection step that selects the quality prediction model to be used for predicting the quality of the metallic material from the multiple quality prediction models based on an evaluation of a specific index.
  • the saving step targets all of the processes, individually selects the specific process, sets a predetermined range for the selected specific process as the unified predetermined range, and repeatedly associates and saves the manufacturing conditions and the quality for one or more of the unified predetermined ranges.
  • the model selection step selects, from among a plurality of quality prediction models, a quality prediction model to be used for predicting the quality of the metallic material based on an index indicating the predictive accuracy of the quality prediction model as the specific index.
  • the model selection step selects a quality prediction model to be used to predict the quality of the metallic material from among a plurality of quality prediction models based on an importance index indicating the impact that the manufacturing conditions that are input to each quality prediction model have on the prediction of the quality that is output.
  • the model selection step generates a quality prediction model from performance data associated with a predetermined range of the final process as the unified predetermined range, calculates an importance index for each manufacturing condition for the generated quality prediction model, extracts the manufacturing condition with the maximum importance index, and selects the quality prediction model generated from performance data associated with a predetermined range of the process to which the extracted manufacturing condition belongs as the unified predetermined range as the quality prediction model to be used for predicting the quality of the metallic material.
  • the model selection step generates a first quality prediction model from performance data associated with a predetermined range of the final process as the unified predetermined range, calculates a first importance index for each manufacturing condition for the first quality prediction model, extracts the manufacturing condition for which the first importance index is maximized, generates a second quality prediction model from performance data associated with a predetermined range of the process to which the extracted manufacturing conditions belong as the unified predetermined range, calculates a second importance index for each manufacturing condition for the second quality prediction model, extracts the manufacturing condition for which the second importance index is maximized, and selects the second quality prediction model when the manufacturing condition for which the second importance index is maximized matches the manufacturing condition for which the first importance index is maximized as the quality prediction model to be used to predict the quality of the metallic material.
  • the method for generating a quality prediction model for metal materials according to the present invention is the same as the above invention, in which the model generation step generates the quality prediction model using statistical analysis methods and machine learning methods, including linear regression, local regression, principal component regression, PLS regression, logistic regression, support vector machines, decision trees, regression trees, random forests, gradient boosting trees, and neural networks.
  • statistical analysis methods and machine learning methods including linear regression, local regression, principal component regression, PLS regression, logistic regression, support vector machines, decision trees, regression trees, random forests, gradient boosting trees, and neural networks.
  • the metallic material quality prediction method of the present invention includes a quality prediction step that uses a quality prediction model generated by the above-mentioned metallic material quality prediction model generation method to predict the quality of a metallic material manufactured under any manufacturing conditions for each specified range.
  • the method for estimating factors affecting the quality of metallic materials includes an influencing factor estimation step that uses a quality prediction model generated by the above-mentioned method for generating a quality prediction model for metallic materials to estimate manufacturing conditions that are factors that affect the quality of the metallic material.
  • the influencing factor estimation step calculates an importance index for the quality prediction model for each manufacturing condition, and estimates manufacturing conditions with high importance indexes as manufacturing conditions that are factors affecting the quality of the metallic material.
  • the method for manufacturing a metallic material includes a quality prediction step for predicting output quality using a quality prediction model that includes as input manufacturing conditions that are factors that affect the quality of the metallic material estimated by the above-mentioned method for estimating factors that affect the quality of the metallic material; a manufacturing condition determination step for determining manufacturing conditions that are factors that affect the quality of the metallic material so that the predicted quality falls within a predetermined range; and a metallic material manufacturing step for manufacturing a metallic material in accordance with the determined manufacturing conditions.
  • the quality prediction model generation method of the present invention includes a first collection step of collecting the manufacturing conditions of materials in each process for each predetermined range of the material defined for each process, a second collection step of evaluating and collecting the quality of the materials in the final process of each of the processes for each predetermined range defined for the final process, and editing the manufacturing conditions and quality of the materials in other processes based on the width of the predetermined range for a specific process among the processes collected in the first collection step and the quality of the materials in the final process collected in the second collection step, and
  • the method includes a storage step of repeatedly associating and storing the manufacturing conditions and the quality for each unified predetermined range corresponding to the number of processes, the unified predetermined range being a predetermined range that compensates for the variations in the manufacturing conditions and quality in each process; a model generation step of generating one or more quality prediction models for each unified predetermined range from the manufacturing conditions for each unified predetermined range stored in the storage step, the quality prediction model predicting the
  • the metallic material quality prediction model generation device of the present invention comprises a first collection unit that collects the manufacturing conditions of the metallic material in each process for each predetermined range of the metallic material defined for each process, a second collection unit that evaluates and collects the quality of the metallic material in the final process of each of the processes for each predetermined range defined for the final process, and a second collection unit that evaluates and collects the manufacturing conditions of the metallic material in each process collected by the first collection unit and the quality of the metallic material in the final process collected by the second collection unit based on the width of the predetermined range for a specific one of the processes, and compares the manufacturing conditions and quality in other processes.
  • the system includes a storage unit that compiles the quality and repeatedly associates and stores the manufacturing conditions and the quality for each unified predetermined range that corresponds to the number of processes and that is a predetermined range that complements the differences in the manufacturing conditions and quality in each process; a model generation unit that generates one or more quality prediction models for each unified predetermined range from the manufacturing conditions for each unified predetermined range stored in the storage unit, and that predicts the quality for each unified predetermined range; and a model selection unit that, when multiple quality prediction models are generated by the model generation unit, selects the quality prediction model to be used to predict the quality of the metallic material from the multiple quality prediction models based on an evaluation of a specific index.
  • the metallic material quality prediction device of the present invention includes a quality prediction unit that uses the quality prediction model generated by the metallic material quality prediction model generation device described above to predict the quality of metallic materials manufactured under any manufacturing conditions for each specified range.
  • the metallic material quality influence factor estimation device includes an influence factor estimation unit that uses the quality prediction model generated by the metallic material quality prediction model generation device described above to estimate manufacturing conditions that are factors that affect the quality of the metallic material.
  • the metallic material quality prediction model generation method, metallic material quality prediction method, metallic material quality influencing factor estimation method, metallic material manufacturing method, quality prediction model generation method, metallic material quality prediction model generation device, metallic material quality prediction device, and metallic material quality influencing factor estimation device make it possible to reflect the maximum amount of information regarding the target quality in the generated quality prediction model by appropriately utilizing the granularity of the manufacturing condition data for each process in manufacturing metallic materials.As a result, quality can be predicted with high accuracy, and the causes of quality defects can be estimated with high accuracy.
  • FIG. 1 is a block diagram showing an example of the configuration of an information processing device that functions as a quality prediction model generation device, a quality prediction device, and a quality influence factor estimation device for metallic materials according to an embodiment of the present invention.
  • FIG. 2 is a flowchart showing the flow of the method for generating a quality prediction model and the method for predicting quality of a metallic material according to an embodiment of the present invention.
  • Figure 3 is a diagram showing an example of manufacturing condition data for each specified range in the longitudinal direction of a metal material collected by the manufacturing condition data collection unit in the metal material quality prediction model generation device, quality prediction device, and quality influence factor estimation device according to an embodiment of the present invention.
  • Figure 4 is a diagram showing an example of a case in which the manufacturing condition data and quality data for the specified ranges of processes 2 and 3 are edited by the integrated process data editing unit in the quality prediction model generation device, quality prediction device, and quality influence factor estimation device for metal materials according to an embodiment of the present invention, using the width of the specified range of process 1 as the standard (unified specified range).
  • Figure 5 is a diagram showing an example of a case in which the manufacturing condition data and quality data for the specified ranges of processes 1 and 3 are edited by the integrated process data editing unit in the quality prediction model generation device, quality prediction device, and quality influence factor estimation device for metal materials according to an embodiment of the present invention, using the width of the specified range of process 2 as the standard (unified specified range).
  • Figure 6 is a diagram showing an example of a case in which the manufacturing condition data and quality data for the specified ranges of processes 1 and 2 are edited by the integrated process data editing unit in the quality prediction model generation device, quality prediction device, and quality influence factor estimation device for metal materials according to an embodiment of the present invention, using the width of the specified range of process 3 as the standard (unified specified range).
  • Figure 7 is a diagram showing an example of performance data when the integrated process data editing unit combines manufacturing condition data and quality data for the specified ranges of all processes using the width of the specified range of process 1 as the standard (unified specified range) in the quality prediction model generation device, quality prediction device, and quality influence factor estimation device for metal materials according to an embodiment of the present invention.
  • Figure 8 is a diagram showing an example of performance data when the integrated process data editing unit combines manufacturing condition data and quality data for the specified ranges of all processes using the width of the specified range of process 2 as the standard (unified specified range) in the quality prediction model generation device, quality prediction device, and quality influence factor estimation device for metal materials according to an embodiment of the present invention.
  • Figure 9 is a diagram showing an example of performance data when the integrated process data editing unit combines manufacturing condition data and quality data for the specified ranges of all processes using the width of the specified range of process 3 as the standard (unified specified range) in the quality prediction model generation device, quality prediction device, and quality influence factor estimation device for metal materials according to an embodiment of the present invention.
  • FIG. 10 is a flowchart showing the flow of a method for estimating quality influence factors of a metallic material according to an embodiment of the present invention.
  • the quality prediction model generation device is a device for generating a quality prediction model for predicting the quality of a metallic material manufactured through one or more processes.
  • the quality prediction device is a device for predicting the quality of a metallic material manufactured through one or more processes using the quality prediction model.
  • the quality influencing factor estimation device is a device for estimating manufacturing conditions, which are factors that affect the quality of the metallic material, using the quality prediction model.
  • the metallic material in this embodiment may be, for example, a steel product, such as a semi-finished product such as a slab, or a finished product such as a steel plate manufactured by rolling this slab.
  • the quality prediction model generation device, quality prediction device, and quality influence factor estimation device can be realized, for example, by an information processing device 1 as shown in Figure 1.
  • the information processing device 1 is composed of a personal computer, a workstation, or the like.
  • the main components of the information processing device 1 include a processor such as a CPU (Central Processing Unit), and memory (main storage unit) such as RAM (Random Access Memory) and ROM (Read Only Memory).
  • a processor such as a CPU (Central Processing Unit)
  • memory main storage unit
  • RAM Random Access Memory
  • ROM Read Only Memory
  • the information processing device 1 includes a manufacturing condition data collection unit 11, a quality data collection unit 12, an integrated process data editing unit 13, an integrated process database 14, a model generation unit 15, a model selection unit 16, a quality prediction unit 17, and an influencing factor estimation unit 18.
  • the metallic material quality prediction model generation device is composed of the elements of the information processing device 1 excluding the quality prediction unit 17 and the influencing factor estimation unit 18.
  • the metallic material quality prediction device is composed of the elements of the information processing device 1 excluding the influencing factor estimation unit 18.
  • the quality influencing factor estimation device according to the embodiment is composed of all the elements of the information processing device 1.
  • a sensor (not shown) is connected to the manufacturing condition data collection unit 11.
  • the manufacturing condition data collection unit 11 uses this sensor to collect data on the manufacturing conditions of each process (hereinafter referred to as “manufacturing condition data") and outputs it to the integrated process data editing unit 13.
  • “metal material manufacturing conditions” include the components of the metal material in each process, the temperature of the metal material, the pressure applied to the metal material, the tension applied to the metal material, the thickness of the metal material before and after rolling, and the threading speed of the metal material.
  • the manufacturing condition data for each process collected by the manufacturing condition data collection unit 11 includes not only the actual values of the manufacturing conditions measured by sensors, but also the preset values of the manufacturing conditions. In other words, since sensors may not be installed in some processes, in such cases the set values are collected as manufacturing condition data instead of actual values.
  • the manufacturing condition data collection unit 11 collects manufacturing condition data for metal materials in each process for each predetermined range of the metal material defined for each process.
  • the "predetermined range” refers to a certain range (position) in the longitudinal direction of the metal material defined in advance for each process.
  • the predetermined range for each process may be different or the same for each process.
  • the predetermined range in the longitudinal direction of the metal material is determined, for example, based on the movement distance of the metal material according to the conveying direction in each process.
  • a sensor (not shown) or a data input terminal (not shown) is connected to the quality data collection unit 12.
  • the quality data collection unit 12 collects data on the quality of the metal material manufactured through each process (hereinafter referred to as "quality data") using this sensor or data input terminal, and outputs this data to the integrated process data editing unit 13.
  • quality data data on the quality of the metal material manufactured through each process
  • Examples of the "quality of the metal material” mentioned above include the tensile strength of the metal material in the final process, the defect contamination rate (number of defects appearing per unit area) of the metal material in the final process, etc.
  • the quality data for metal materials collected by the quality data collection unit 12 includes not only actual quality measurements taken by sensors, but also quality judgment values entered from a data input terminal, where pass/fail is evaluated based on certain standards or visual inspection by an inspector, etc., based on the actual measurements.
  • the quality data collection unit 12 evaluates and collects the quality of the metal material in the final step of each process within a specified range established for the final step. Note that the "specified range” here is synonymous with the “specified range” in the manufacturing condition data collection unit 11.
  • the integrated process data editing unit 13 edits the performance data input from the manufacturing condition data collection unit 11 and the quality data collection unit 12.
  • the integrated process data editing unit 13 associates the manufacturing condition data for each process collected by the manufacturing condition data collection unit 11 with the quality data for the metal material collected by the quality data collection unit 12 for each specified range within a specific process within each process, and stores the data in the integrated process database 14.
  • the integrated process data editing unit 13 edits the manufacturing condition data and quality data for the other processes, using the width of a specified range for a specific process among the processes as a standard, for the manufacturing condition data for the metal material in each process and the quality data for the metal material in the final process among the processes. In doing so, the integrated process data editing unit 13 repeatedly associates the manufacturing condition data and quality data with a unified specified range corresponding to the number of processes, which is a specified range that compensates for differences in the density of the manufacturing condition data and quality data in each process, and stores the data in the integrated process database 14.
  • the integrated process data editing unit 13 selects a specific process individually from all processes, sets a specified range for the selected specific process as a unified specified range, and repeatedly associates the manufacturing condition data and quality data for one or more unified specified ranges and stores them in the integrated process database 14.
  • the integrated process data editing unit 13 creates only one set of data set (e.g., a learning data set) in which the manufacturing condition data and quality data have been edited, and stores this in the integrated process database 14.
  • the integrated process data editing unit 13 creates multiple sets of data set (e.g., a learning data set) in which the manufacturing condition data and quality data have been edited, and stores these in the integrated process database 14. Details of the processing by the integrated process data editing unit 13 will be described later (see Figures 4 to 9).
  • the model generation unit 15 generates one or more quality prediction models for each unified specified range, using the manufacturing condition data and quality data for each unified specified range stored in the integrated process database 14.
  • the model generation unit 15 generates only one quality prediction model.
  • multiple unified specified ranges are set. In this case, the model generation unit 15 generates multiple quality prediction models.
  • the model generation unit 15 uses, for example, gradient boosting trees as a statistical analysis method and machine learning method when generating a quality prediction model.
  • various other statistical analysis methods and machine learning methods can also be used, such as linear regression, local regression, principal component regression, PLS regression, logistic regression, support vector machines, decision trees, regression trees, random forests, and neural networks.
  • the model selection unit 16 selects a quality prediction model to be used to predict the quality of the metal material (processing in the quality prediction unit 17) from among multiple quality prediction models, for example, based on the evaluation of a specific index. In this case, the model selection unit 16 selects a quality prediction model to be used to predict the quality of the metal material from among multiple quality prediction models, based on, for example, an index indicating the prediction accuracy of the quality prediction model as the specific index.
  • model selection unit 16 may select a quality prediction model to be used to predict the quality of the metal material from among multiple quality prediction models based on an importance index that indicates the impact that the manufacturing conditions that are input to each quality prediction model have on the predicted quality that is output. Details of the processing by the model selection unit 16 will be described later.
  • the quality prediction unit 17 uses the quality prediction model generated by the model generation unit 15 and selected by the model selection unit 16 as necessary to predict the quality of metal materials manufactured under any manufacturing conditions for each specified range.
  • the influencing factor estimation unit 18 uses the quality prediction model generated by the model generation unit 15 and selected by the model selection unit 16 as needed to estimate manufacturing conditions that are factors that affect the quality of the metal material. Details of the processing performed by the influencing factor estimation unit 18 will be described later.
  • a quality prediction method and a quality prediction model generation method according to this embodiment will be described with reference to Figures 2 to 9.
  • the quality prediction method according to this embodiment performs the processes of steps S11 to S17 shown in Figure 2.
  • the quality prediction model generation method according to this embodiment performs the processes of steps S11 to S16 shown in Figure 2, excluding step S17.
  • step S11 corresponds to the first collection step
  • step S12 corresponds to the second collection step
  • step S13 corresponds to the storage step
  • step S14 corresponds to the model generation step
  • steps S15 and S16 correspond to the model selection step
  • step S17 corresponds to the prediction step.
  • the manufacturing condition data collection unit 11 collects manufacturing condition data for each process for a predetermined range of metal materials that is predefined for each process (step S11).
  • the manufacturing condition data collected by the manufacturing condition data collection unit 11 is data in which the actual values or set values of multiple manufacturing conditions are listed for each predetermined range of metal materials that is predefined for each process, as shown in Figure 3, for example.
  • the manufacturing condition data shown in Fig. 3 has items consisting of longitudinal positions l1 , l2 , ... in each process and multiple manufacturing conditions x11 , x12 , ..., x21 , x22 , ... measured by a sensor at those positions.
  • the quality data collection unit 12 evaluates and collects quality data for the metal material manufactured through each process within a predetermined range (step S12).
  • the quality data collected by the quality data collection unit 12 is collected at the final process of multiple processes, and is data containing quality evaluation results according to the longitudinal position of the metal material from the leading end to the trailing end in each process.
  • the integrated process data editing unit 13 associates and saves the manufacturing condition data for each process with the quality data for the metal material manufactured under those manufacturing conditions (step S13). Based on the manufacturing condition data and quality data for the longitudinal position of the metal material collected in steps S11 and S12, the integrated process data editing unit 13 edits the manufacturing condition data and quality data for other processes, using the width of the specified range for a specific one of the processes as a reference. The integrated process data editing unit 13 then aligns, combines, and associates the manufacturing condition data and quality data for each process into a unified specified range that interpolates and compensates for differences in the density of granularity within the specified range.
  • “granularity of the specified range” refers to, for example, the length of the acquisition (collection) cycle of manufacturing condition data and quality data. If the acquisition cycle of manufacturing condition data and quality data is short, the granularity of the specified range will be fine. On the other hand, if the acquisition cycle of manufacturing condition data and quality data is long, the granularity of the specified range will be coarse. Furthermore, “uniform specified range” refers to the width of the specified range for a specific process among the processes.
  • the integrated process data editing unit 13 associates and stores the manufacturing condition data for each process with the quality data for the metal material manufactured under those manufacturing conditions for each uniform, predetermined range in the longitudinal direction of the metal material.
  • An example of processing by the integrated process data editing unit 13 is described below with reference to Figures 4 to 9.
  • manufacturing condition data that takes a predetermined range represented by manufacturing condition xa is collected in step 1
  • manufacturing condition data that takes a predetermined range represented by manufacturing condition xb is collected in step 2
  • manufacturing condition data that takes a predetermined range represented by manufacturing condition xc is collected in step 3, along with quality data (not shown).
  • the integrated process data editing unit 13 edits the manufacturing condition data and quality data for the specified ranges of processes 2 and 3, using the width of the specified range of process 1 as a standard (unified specified range), as shown in Figure 4. This interpolates and compensates for the differences in density of the granularity of the specified ranges of processes 2 and 3 relative to the granularity of the specified range of process 1.
  • the integrated process data editing unit 13 compiles manufacturing condition data for two specified ranges in process 2, which correspond to the width of the specified range in process 1, and associates the data with each specified range in process 1 as representative data for the two specified ranges in process 2. Similarly, the integrated process data editing unit 13 compiles manufacturing condition data and quality data for four specified ranges in process 3, which correspond to the width of the specified range in process 1, and creates a data set associated with each specified range in process 1 as representative data for the four specified ranges in process 3.
  • the integrated process data editing unit 13 edits the manufacturing condition data and quality data for the specified ranges of processes 1 and 3, using the width of the specified range of process 2 as a standard (unified specified range), as shown in Figure 5. This interpolates and compensates for the differences in density of the granularity of the specified ranges of processes 1 and 3 relative to the granularity of the specified range of process 2.
  • the integrated process data editing unit 13 associates the manufacturing condition data for a specified range of process 1, which corresponds to the width of the specified range of process 2, as representative data for each specified range of process 2.
  • the representative data for process 1 is the same data for two adjacent specified ranges of process 2.
  • the integrated process data editing unit 13 compiles the manufacturing condition data and quality data for two specified ranges of process 3, which correspond to the width of the specified range of process 2, and creates a data set associated with each specified range of process 2 as representative data for the two specified ranges of process 3.
  • the integrated process data editing unit 13 edits the manufacturing condition data for the specified ranges of processes 1 and 2, using the width of the specified range of process 3 as a standard (unified specified range), as shown in Figure 6. This interpolates and compensates for the differences in density of the granularity of the specified ranges of processes 1 and 2 relative to the granularity of the specified range of process 3.
  • the integrated process data editing unit 13 creates a data set in which the manufacturing condition data for a specified range of process 1, which corresponds to the width of the specified range of process 3, is used as representative data and associated with each specified range of process 3. At this time, the representative data for process 1 will be the same data in the four adjacent specified ranges of process 3. Furthermore, the integrated process data editing unit 13 associates the actual data of the manufacturing conditions for a specified range of process 2, which corresponds to the width of the specified range of process 3, as representative data for each specified range of process 3. At this time, the representative data for process 2 will be the same data in the two adjacent specified ranges of process 3.
  • each sub-step is actually repeated K times, corresponding to the K number of processes.
  • the integrated process data editing unit 13 creates K data sets corresponding to the K number of processes.
  • commonly known statistics such as averages, maximum values, and minimum values may be used to determine representative data through aggregation.
  • performance data selected based on arbitrary conditions from performance data within multiple specified ranges may be used.
  • the model generation unit 15 generates one or more quality prediction models that predict the quality of each specified range of the metal material from the manufacturing condition data (dataset) for each unified specified range in each process, depending on the granularity of the unified specified range in which the manufacturing condition data and quality data are associated (step S14). If there is one process for manufacturing the metal material and only one unified specified range is set, the model generation unit 15 generates only one quality prediction model. On the other hand, if there are multiple processes for manufacturing the metal material and multiple unified specified ranges are set, the model generation unit 15 generates multiple quality prediction models for the number of data sets created.
  • the model selection unit 16 determines whether multiple quality prediction models were generated in the model generation step (step S15). If multiple quality prediction models were generated in the model generation step (Yes in step S15), the process proceeds to step S16. On the other hand, if multiple quality prediction models were not generated in the model generation step (No in step S15), the process proceeds to step S17.
  • the model selection unit 16 selects a quality prediction model to use from among the multiple quality prediction models based on an evaluation of an index indicating the prediction accuracy of the quality prediction model (step S16).
  • the model selection unit 16 evaluates the index indicating the prediction accuracy of the quality prediction model, for example, by holdout validation or cross validation, and selects a quality prediction model based on the granularity of the unified specified range in which the index indicates the smallest or largest value, depending on the index used.
  • the above-mentioned indicators can be, for example, the mean squared error (MSE), the root mean squared error (RMSE), the mean absolute error (MAE), the coefficient of determination (R2), etc.
  • MSE mean squared error
  • RMSE root mean squared error
  • MAE mean absolute error
  • R2 coefficient of determination
  • the quality prediction model is a classification model
  • the above-mentioned indicators can be, for example, the accuracy rate (Accuracy), precision rate (Precision), recall rate (Recall), F-measure, etc.
  • the quality prediction unit 17 uses the quality prediction model generated by the model generation unit 15 and selected by the model selection unit 16 as necessary to predict the quality of the metal material manufactured under any manufacturing conditions for each specified range (step S17).
  • step S17 quality prediction is performed in step S17 after quality prediction model selection is performed in step S16, but it is not necessary to perform all steps up to step S16 each time step S17 is performed.
  • steps S11 to S16 may be performed in advance to select a quality prediction model, and then only steps S11, S13, and S17 may be performed each time step S17 is performed using the selected quality prediction model.
  • step S21 corresponds to the first collection step
  • step S22 corresponds to the second collection step
  • step S23 corresponds to the storage step
  • step S24 corresponds to the model generation step
  • steps S25 and S26 correspond to the model selection step
  • step S27 corresponds to the quality influence factor estimation step.
  • the model selection unit 16 selects a quality prediction model to use from among multiple quality prediction models based on an evaluation of an index indicating the prediction accuracy of the quality prediction model (step S26).
  • the model selection unit 16 selects a quality prediction model to use to predict the quality of the metal material based on an importance index indicating the impact that the manufacturing condition data, which is the input to the multiple quality prediction models generated by the model generation unit 15, has on the prediction of the output quality data.
  • the model selection unit 16 calculates an importance index (variable importance) for each manufacturing condition for a quality prediction model generated from performance data associated with a predetermined range for the final process in which quality data is collected, as a unified predetermined range, and then extracts the manufacturing condition with the maximum importance index.
  • the importance index calculates an index corresponding to the influence of each factor on the entire predictive model.
  • an algorithm may be used that calculates an importance index specific to the algorithm that created the model, such as importance based on Gini impurity in decision tree algorithms such as random forests.
  • a well-known method such as "Permutation Importance,” that calculates a general-purpose importance index regardless of the model creation algorithm may be used.
  • the model selection unit 16 selects, as the quality prediction model to be used for predicting quality, a quality prediction model generated from performance data associated with a predetermined range of processes to which the manufacturing conditions with the highest importance index belong, as a unified predetermined range.
  • the model selection unit 16 calculates a first importance index (variable importance) for each manufacturing condition for a first quality prediction model generated from performance data associated with a predetermined range for the final process in which quality data is collected among the processes as a unified predetermined range, and then extracts the manufacturing condition with the largest first importance index.
  • the importance index is calculated as an index corresponding to the influence of each factor on the entire predictive model.
  • an algorithm may be used that calculates an importance index specific to the algorithm that created the model, such as importance based on Gini impurity in decision tree algorithms such as random forests.
  • a known method may be used that calculates an importance index generically, regardless of the model creation algorithm, such as "Permutation Importance.”
  • the model selection unit 16 calculates a second importance index for each manufacturing condition for a second quality prediction model generated from associated performance data, with the specified range of the process to which the manufacturing condition with the maximum first importance index belongs being used as the unified specified range.
  • the model selection unit 16 then extracts the manufacturing condition with the maximum second importance index.
  • the model selection unit 16 selects the second quality prediction model as the model to be used.
  • the model selection unit 16 extracts a third importance index for each manufacturing condition for the third quality prediction model generated from the associated performance data, with the specified range of the process to which the manufacturing condition with the largest second importance index belongs being used as the unified specified range. Then, the model selection unit 16 extracts the manufacturing condition with the largest third importance index.
  • the model selection unit 16 extracts the kth importance index for each manufacturing condition, and repeats the above process until the process to which the manufacturing condition with the largest kth importance index belongs matches the process to which the manufacturing condition with the largest k-1th importance index belongs.
  • the Nth quality prediction model is selected as the model to be used. Note that an arbitrary upper limit may be set for the number of times the process is repeated.
  • a metal material material
  • FIGS. 4 to 6 i.e., where three quality prediction models are generated from three pieces of actual data.
  • a first importance index is calculated for each manufacturing condition for a first quality prediction model generated from actual data (see FIG. 6 ) associated with a predetermined range for the final process, process 3, as the unified predetermined range.
  • the importance index for manufacturing condition x a is the largest
  • a second importance index is calculated for each manufacturing condition for a second quality prediction model generated from actual data (see FIG. 4 ) associated with a predetermined range for process 1, to which manufacturing condition x a belongs, as the unified predetermined range.
  • a third importance index is calculated for each manufacturing condition for a third quality prediction model generated from actual data (see FIG. 5 ) associated with a predetermined range for process 2, to which manufacturing condition x b belongs, as the unified predetermined range. For example, if the importance index of the manufacturing condition xb is the largest, the process to which the manufacturing condition with the largest third importance index belongs coincides with the process to which the manufacturing condition with the largest second importance index belongs, and therefore the third quality prediction model is selected as the quality prediction model to be used for predicting quality.
  • the importance index of the manufacturing condition xb is the largest
  • the process to which the manufacturing condition with the largest third importance index belongs coincides with the process to which the manufacturing condition with the largest second importance index belongs, and therefore the third quality prediction model is selected as the quality prediction model to be used for predicting quality.
  • the influencing factor estimation unit 18 uses the quality prediction model selected by the model selection unit 16 to estimate manufacturing conditions that are factors that affect the quality of the metal material (step S27). For example, the influencing factor estimation unit 18 calculates the variable importance of the quality prediction model for each manufacturing condition, and extracts any number of manufacturing conditions in descending order starting with the highest variable importance value. The influencing factor estimation unit 18 then presents the extracted manufacturing conditions as a list of item names of manufacturing conditions estimated to be influencing factors, or presents them graphically, such as in a descending bar graph.
  • the method for manufacturing a metallic material includes a quality prediction step, a manufacturing condition determination step, and a metallic material manufacturing step.
  • the influencing factor estimation unit 18 predicts the output quality using a quality prediction model selected by the model selection unit 16, which includes as input factors that affect the quality of the metal material and estimated manufacturing conditions.
  • the metal material is manufactured based on manufacturing conditions that have been adjusted so that the manufacturing conditions estimated to be factors that affect the quality of the metal material are within the determined control range.
  • Example 2 An example of the method for predicting quality of a metallic material according to the embodiment will be described.
  • the method for predicting quality of a metallic material according to the embodiment was applied to predicting the number of surface defects in a surface-treated steel sheet (hot-dip galvanized steel sheet).
  • the specified grain size ranges for the casting process, hot rolling/rough rolling process, hot rolling/finish rolling process, cold rolling process, and surface treatment process were 0.2m pitch, 0.2m pitch, 0.5m pitch, 1m pitch, and 1m pitch, respectively.
  • one or more quality prediction models are generated that associate the manufacturing conditions of each process with the quality of the metallic material manufactured under those manufacturing conditions for a predetermined range of each process, and the quality prediction model to be used is selected based on a specific index. This makes it possible to predict the quality of metallic materials for any manufacturing conditions with higher accuracy than before, and to estimate manufacturing conditions that are factors that affect the quality of metallic materials with high accuracy.

Landscapes

  • Engineering & Computer Science (AREA)
  • Mechanical Engineering (AREA)
  • General Engineering & Computer Science (AREA)
  • Manufacturing & Machinery (AREA)
  • Quality & Reliability (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Automation & Control Theory (AREA)
  • General Factory Administration (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
PCT/JP2025/002222 2024-04-09 2025-01-24 金属材料の品質予測モデル生成方法、金属材料の品質予測方法、金属材料の品質影響因子推定方法、金属材料の製造方法、品質予測モデル生成方法、金属材料の品質予測モデル生成装置、金属材料の品質予測装置および金属材料の品質影響因子推定装置 Pending WO2025215908A1 (ja)

Priority Applications (1)

Application Number Priority Date Filing Date Title
JP2025529959A JPWO2025215908A1 (https=) 2024-04-09 2025-01-24

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
JP2024-062571 2024-04-09
JP2024062571 2024-04-09

Publications (1)

Publication Number Publication Date
WO2025215908A1 true WO2025215908A1 (ja) 2025-10-16

Family

ID=97349387

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/JP2025/002222 Pending WO2025215908A1 (ja) 2024-04-09 2025-01-24 金属材料の品質予測モデル生成方法、金属材料の品質予測方法、金属材料の品質影響因子推定方法、金属材料の製造方法、品質予測モデル生成方法、金属材料の品質予測モデル生成装置、金属材料の品質予測装置および金属材料の品質影響因子推定装置

Country Status (2)

Country Link
JP (1) JPWO2025215908A1 (https=)
WO (1) WO2025215908A1 (https=)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2009151383A (ja) * 2007-12-18 2009-07-09 Nippon Steel Corp 製造プロセスにおける操業と品質の関連解析装置、解析方法、プログラム、及びコンピュータ読み取り可能な記録媒体
JP2013526751A (ja) * 2010-05-21 2013-06-24 フィッシャー−ローズマウント システムズ，インコーポレイテッド 多段処理のモデル構築法
US20220043435A1 (en) * 2020-08-07 2022-02-10 At&S Austria Technologie & Systemtechnik Aktiengesellschaft AI-Based Determination of Action Plan for Manufacturing Component Carriers
JP2022048038A (ja) * 2020-09-14 2022-03-25 Ｊｆｅスチール株式会社 材料特性値予測システム及び金属板の製造方法
WO2022149344A1 (ja) * 2021-01-06 2022-07-14 Ｊｆｅスチール株式会社 品質異常解析方法、金属材料の製造方法および品質異常解析装置

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP2009151383A (ja) * 2007-12-18 2009-07-09 Nippon Steel Corp 製造プロセスにおける操業と品質の関連解析装置、解析方法、プログラム、及びコンピュータ読み取り可能な記録媒体
JP2013526751A (ja) * 2010-05-21 2013-06-24 フィッシャー−ローズマウント システムズ，インコーポレイテッド 多段処理のモデル構築法
US20220043435A1 (en) * 2020-08-07 2022-02-10 At&S Austria Technologie & Systemtechnik Aktiengesellschaft AI-Based Determination of Action Plan for Manufacturing Component Carriers
JP2022048038A (ja) * 2020-09-14 2022-03-25 Ｊｆｅスチール株式会社 材料特性値予測システム及び金属板の製造方法
WO2022149344A1 (ja) * 2021-01-06 2022-07-14 Ｊｆｅスチール株式会社 品質異常解析方法、金属材料の製造方法および品質異常解析装置

Also Published As

Publication number Publication date
JPWO2025215908A1 (https=) 2025-10-16
TW202540911A (zh) 2025-10-16

Similar Documents

Publication Publication Date Title
KR102939974B1 (ko) 품질 예측 모델 생성 방법, 품질 예측 모델, 품질 예측 방법, 금속 재료의 제조 방법, 품질 예측 모델 생성 장치 및 품질 예측 장치
JP5169096B2 (ja) 品質予測装置、品質予測方法及び製品の製造方法
US20240295874A1 (en) Quality abnormality analysis method, method of manufacturing metal material, and quality abnormality analysis device
CN110569566B (zh) 一种板带力学性能预测方法
CN120055031B (zh) 一种复合板复层厚度的预测方法及相关设备
CN111080054A (zh) 一种热轧带钢宽度质量自动评价方法及系统
CN117862247A (zh) 一种基于机器学习的轧制过程带钢板形预测方法
JP5962290B2 (ja) 鋼材の熱伝達係数予測装置及び冷却制御方法
CN103028614B (zh) 一种带钢热连轧生产过程控制系统的优化方法
WO2025215908A1 (ja) 金属材料の品質予測モデル生成方法、金属材料の品質予測方法、金属材料の品質影響因子推定方法、金属材料の製造方法、品質予測モデル生成方法、金属材料の品質予測モデル生成装置、金属材料の品質予測装置および金属材料の品質影響因子推定装置
CN120494344B (zh) 基于数据挖掘的铝吊顶生产优化方法及系统
CN119446332A (zh) 一种钢材的工艺数据在线预测及诊断的方法及相关设备
JP7800785B1 (ja) 金属材料の品質予測モデル生成方法、金属材料の品質予測方法、金属材料の品質影響要因推定方法、金属材料の製造方法、金属材料の品質予測モデル生成装置、金属材料の品質予測装置および金属材料の品質影響要因推定装置
TWI916223B (zh) 金屬材料的品質預測模型生成方法、金屬材料的品質預測方法、金屬材料的品質影響因子推估方法、金屬材料的製造方法、品質預測模型生成方法、金屬材料的品質預測模型生成裝置、金屬材料的品質預測裝置及金屬材料的品質影響因子推估裝置
JP7715316B1 (ja) 金属材料の品質予測モデル生成方法、金属材料の品質予測方法、金属材料の製造方法、品質予測モデル生成方法、データ加工方法、金属材料の品質予測モデル生成装置、金属材料の品質予測装置およびデータ加工装置
WO2026048122A1 (ja) 金属材料の品質予測モデル生成方法、金属材料の品質予測方法、金属材料の品質影響要因推定方法、金属材料の製造方法、金属材料の品質予測モデル生成装置、金属材料の品質予測装置および金属材料の品質影響要因推定装置
WO2025215895A1 (ja) 金属材料の品質予測モデル生成方法、金属材料の品質予測方法、金属材料の製造方法、品質予測モデル生成方法、データ加工方法、金属材料の品質予測モデル生成装置、金属材料の品質予測装置およびデータ加工装置
TW202609480A (zh) 金屬材料的品質預測模型生成方法、金屬材料的品質預測方法、金屬材料的品質影響因子推定方法、金屬材料的製造方法、金屬材料的品質預測模型生成裝置、金屬材料的品質預測裝置及金屬材料的品質影響因子推定裝置
Ayed et al. Monitoring of Product Conformity Based on Machine Learning Prediction Evaluation
CN121189874A (zh) 一种多预测模型耦合的企业过程安全指标预测及评价方法
JPWO2025215908A5 (https=)
JP2024159169A (ja) 操業条件決定装置、操業条件決定方法及びプログラム
CN119539162A (zh) 一种基于人工智能的生产能耗值预测方法及系统
CN120975143A (zh) 一种基于Rao-1-Informer的时序数据预测方法及装置
CN121616140A (zh) 一种罩退带材的擦伤粘结预测方法及系统

Legal Events

Date Code Title Description
ENP Entry into the national phase

Ref document number: 2025529959

Country of ref document: JP

Kind code of ref document: A

WWE Wipo information: entry into national phase

Ref document number: 2025529959

Country of ref document: JP

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 25786571

Country of ref document: EP

Kind code of ref document: A1