Steel evaluation method, program, and model generation method

Abstract

The invention provides a steel evaluation method and the like, which can evaluate steel satisfying specifications related to composition in more detail. The steel evaluation method analyzes the composition of the steel; a calculation unit that, when the composition data obtained by the analysis satisfies a predetermined specification, inputs the composition data into a model that is machine-learned to output the same data as the input data, and calculates an error loss between output data output from the model and the composition data; and determining whether the error loss exceeds a predetermined reference value.

Description

Technical Field

[0001] This invention relates to methods, procedures, and models for evaluating steel. Background Technology

[0002] Steel is an alloy made of iron with small amounts of additives such as carbon, silicon, manganese, and phosphorus. The properties of steel vary greatly depending on the proportions of its components, i.e., its composition. Therefore, various types of steel are manufactured and used, such as carbon steel, tool steel, and spring steel. For each type of steel, specifications related to its composition are specified (Patent Document 1).

[0003] Existing technical documents

[0004] Patent documents

[0005] Patent Document 1: Japanese Patent Application Publication No. 2006-152356 Summary of the Invention

[0006] The problem the invention aims to solve

[0007] In the above specifications, the proportions of each component are determined by a width. During steel manufacturing, the content of each component is adjusted to fall within the defined width, thus meeting the composition-related specifications. Various outgoing inspections, including composition analysis, are performed on the manufactured steel to confirm that the composition-related specifications are actually met. Steel that passes the outgoing inspections is then shipped.

[0008] However, experience shows that when comparing multiple batches of steel manufactured to meet specifications related to the same composition, some steels sometimes exhibit superior properties compared to other batches. By detecting the occurrence of these superior steel properties and incorporating them into the management of the manufacturing process, it is possible to improve the quality of the steel.

[0009] In one aspect, the aim is to provide an evaluation method for steel that can evaluate steel that meets composition-related specifications in more detail.

[0010] Solution for solving the problem

[0011] A method for evaluating steel involves analyzing its composition; when the composition data obtained through analysis meets predetermined specifications, the composition data is input into a model that is machine-learned to output data identical to the input data, and the error loss between the output data from the model and the composition data is calculated; and it is determined whether the error loss exceeds a predetermined benchmark value.

[0012] Invention Effects

[0013] In one aspect, it is possible to provide an evaluation method for steel that can evaluate steels that meet composition-related specifications in more detail. Attached Figure Description

[0014] Figure 1 This is an illustrative diagram illustrating the process of determining and applying methods for evaluating steel.

[0015] Figure 2 This is an illustrative diagram illustrating the application stages of steel evaluation methods.

[0016] Figure 3 This is a table showing the composition of the samples used in the preliminary research phase.

[0017] Figure 4 It is a chart showing the rate of dimensional change of steels with different molybdenum contents after heat treatment.

[0018] Figure 5 It is a chart that shows the rate of dimensional change of steels with different chromium contents after heat treatment.

[0019] Figure 6 It is a chart that shows the rate of dimensional change of steels with different vanadium contents after heat treatment.

[0020] Figure 7 It is a chart that shows the rate of dimensional change of steels with different silicon contents after heat treatment.

[0021] Figure 8 It is a chart that shows the rate of dimensional change of steels with different manganese contents after heat treatment.

[0022] Figure 9 This is an explanatory diagram illustrating the structure of the information processing device used to generate the model.

[0023] Figure 10 This is an explanatory diagram illustrating the model.

[0024] Figure 11 It is a flowchart illustrating the processing flow of the program that generates the model.

[0025] Figure 12 It is a chart that shows the variation in the mean absolute error loss generated during the manufacturing period.

[0026] Figure 13 This is an explanatory diagram illustrating the structure of the steel evaluation system.

[0027] Figure 14 It is a flowchart illustrating the processing flow of the program during the application phase. Detailed Implementation

[0028] [Implementation Method 1]

[0029] Figure 1This is an explanatory diagram illustrating the process of determining and applying a steel evaluation method. In this embodiment, an example is given where an evaluation method is determined that, for steel already in mass production, the composition is fine-tuned to stably reproduce superior properties, and the fine-tuned steel is manufactured stably. Figure 1 A summary of the process for determining the evaluation method for steel is provided.

[0030] First, a preliminary research phase is conducted. In this phase, samples of several steels with slightly different compositions are prepared. These samples include steels with standard properties from mass production, as well as steels with intentionally increased or decreased content of various components.

[0031] Compositional analysis was performed on each sample individually. Furthermore, the properties of each sample were evaluated. Based on the results of the compositional analysis and property evaluation, a target composition was determined to stably manufacture steel with the desired properties.

[0032] The target composition is determined by increasing or decreasing the proportion of specific components within a specified range of their content determined by the specifications of the mass-produced steel. In other words, the target composition will not deviate from the specifications of the mass-produced steel. Details of the preliminary study phase will be provided later.

[0033] Next, the data collection phase is implemented. During this phase, multiple batches of steel are manufactured based on manufacturing conditions modified from the target composition established in the preliminary research phase. Furthermore, batch-to-batch variations exist in the composition of the actual manufactured steel.

[0034] Compositional analysis is performed on each batch of steel. A training database (DB) 51 is created, containing multiple sets of compositional analysis data. A model 56 is generated using the training database 51. Model 56 is, for example, an autoencoder trained through machine learning to recover and output data identical to the input compositional analysis data. Details of model 56 will be explained later.

[0035] Next, the validation phase is implemented. In the validation phase, compositional analysis data from multiple batches of steel manufactured prior to the preliminary study phase, as well as compositional analysis data from multiple batches of steel manufactured during the data collection phase, are used. Each compositional analysis data point is input into model 56, and output data is output from model 56. The error loss between the input compositional analysis data and the output data is calculated.

[0036] In this implementation, the error loss is defined by the Loss Mean Absolute Error (LAE). The error loss is smaller when the composition of the input data is similar to the training data used to generate model 56. Conversely, the error loss is larger when the composition of the input data differs significantly from the training data.

[0037] Therefore, the error loss calculated for steel manufactured before the data collection phase tends to be larger than the error loss calculated for steel manufactured during the data collection phase. A baseline value for the error loss is established so that its level can distinguish the deviation range of steel batches manufactured after the data collection phase from the deviation range of steel batches manufactured before the data collection phase. Details of the verification phase will be explained later.

[0038] Finally, an overview of the application phase is provided. Similar to the verification phase, error losses are calculated for each steel manufacturing batch. Evaluations of the manufactured steel are conducted based on established benchmark values. Details of the application phase will be provided later.

[0039] This section provides a summary of the significance of establishing benchmark values ​​for error loss. Steel with a composition similar to the target composition established in the preliminary research phase is expected to achieve good properties. However, because the content of each component varies due to manufacturing deviations, it is difficult for manufacturing engineers to determine whether the characteristics of the actual measured composition are similar to the target composition when they see the composition data.

[0040] By using the benchmark value for error loss established during the verification phase, it is easy to determine whether the composition is similar to the target composition. That is, when the calculated error loss is below the benchmark value, the composition of the manufactured steel is similar to that of the steel manufactured during the data collection phase, and it can be determined that it has the desired characteristics. When the calculated error loss exceeds the benchmark value, the composition of the manufactured steel, even if it meets the predetermined specifications, is different from that of the steel manufactured during the data collection phase, and it can be determined that it does not possess the particularly desired characteristics.

[0041] Figure 2 This diagram illustrates the application stages of steel evaluation methods. Newly manufactured steel undergoes various quality checks, including compositional analysis. Based on the quality check results, a pass / fail determination is made to determine whether the steel meets the specified specifications. Steel deemed non-compliant is not shipped as it is not a qualified product but is used as raw material for manufacturing new steel. The pass / fail determination process for steel quality checks has been implemented continuously, therefore detailed explanations are omitted.

[0042] For steel deemed acceptable (acceptable product), compositional analysis data is input into model 56. The error loss between the output data of model 56 and the compositional analysis data is calculated. When the error loss is below a benchmark value, the steel is considered a high-quality product with the desired characteristics. High-quality products are typically delivered to customers requiring exceptionally high reliability.

[0043] When the error loss exceeds the benchmark value, the steel, although meeting the predetermined specifications, is deemed not to be a superior product. In the following description, products that are acceptable but not superior are recorded as near-superior products. For near-superior products, additional inspections are conducted, for example, based on their intended use at the delivery location. Based on the results of these additional inspections, the near-superior steel is categorized into general acceptable products and specific-purpose acceptable products, and shipped to the appropriate customer.

[0044] As mentioned above, steel that has passed routine quality inspection and is deemed qualified is classified into excellent and near-excellent products based on benchmark values. Near-excellent products are further inspected and sorted according to customer applications, so that each type of steel can be shipped to the appropriate customer with corresponding characteristics.

[0045] [Preliminary Research Phase]

[0046] Figure 3 This table shows the composition of the samples used in the preliminary research phase. Eight samples were used. Each sample was... Figure 3 The sample symbols shown on the left are used for differentiation. For each sample, the components are expressed as a percentage by mass.

[0047] "Std" indicates steel with standard properties produced in batches. "MoL" indicates steel with a molybdenum (Mo) content less than "Std". "MoH" indicates steel with a molybdenum content more than "Std". "CrL" indicates steel with a chromium (Cr) content less than "Std". "CrH" indicates steel with a chromium content more than "Std". "VH" indicates steel with a vanadium (V) content more than "Std". "SiL" indicates steel with a silicon (Si) content less than "Std". "MnH" indicates steel with a manganese (Mn) content more than "Std".

[0048] against Figure 3 For each sample shown, six cuboid specimens were prepared. Each specimen was cut from a piece of steel, with the direction in which the steel was stretched by heat treatment defined as the length direction, the direction in which the steel was compressed defined as the thickness direction, and the direction orthogonal to both the length and thickness directions defined as the width direction. In the diagrams used in the following description, the length direction is represented by the symbol L, the width direction by the symbol W, and the thickness direction by the symbol T.

[0049] For each sample, the dimensional change rate before and after heat treatment was measured. The dimensional change rate is expressed as the percentage of change in size obtained by dividing the dimensional change caused by heat treatment by the size before heat treatment. The dimensional change rate was measured using... Figure 1 An example of a characteristic evaluation method used in the preliminary research phase of the study.

[0050] The heat treatment consisted of quenching and tempering. The quenching conditions were all identical. A quenching process was performed from 1020°C with a 10-minute semi-cooling period. The semi-cooling figure refers to the time required to cool to half the temperature of 1020°C (510°C). Tempering was carried out at five tempering temperatures: 485°C, 500°C, 515°C, 530°C, and 545°C. Two tempering processes were performed at each temperature.

[0051] Figure 4 This is a chart showing the rate of dimensional change after heat treatment of steels with different molybdenum contents side-by-side. Figure 4 Three charts sharing a common vertical axis are arranged horizontally in the middle. The leftmost (a) represents the dimensional change rate of "MoL" after heat treatment, the middle (b) represents the dimensional change rate of "Std" after heat treatment, and the rightmost (c) represents the dimensional change rate of "MoH" after heat treatment.

[0052] The horizontal axis of the graph represents the tempering temperature, and the vertical axis represents the dimensional change rate after heat treatment. The dimensional change rate on the vertical axis is calculated by measuring the sample dimensions before and after heat treatment, dividing the change in dimensions by the original size. Positive values ​​on the vertical axis indicate expansion due to heat treatment, while negative values ​​indicate contraction. "As(Q)" on the horizontal axis indicates a sample that has only undergone quenching.

[0053] In the chart, solid circles represent the dimensional change rate of heat treatment along the length direction (L). Solid diamonds represent the dimensional change rate of heat treatment along the width direction (W). Solid triangles represent the dimensional change rate of heat treatment along the thickness direction (T). For each temperature, the average ave of the three dimensional change rates in the length, width, and thickness directions, as well as the sample standard deviation s, are calculated, with the range ave ± 3s represented in gray.

[0054] Figure 5 This is a chart showing the rate of dimensional change after heat treatment of steels with different chromium contents side-by-side. Figure 5 Three charts sharing a common vertical axis are arranged horizontally. The leftmost (a) represents the dimensional change rate of "CrL" after heat treatment, the middle (b) represents the dimensional change rate of "Std" after heat treatment, and the rightmost (c) represents the dimensional change rate of "CrH" after heat treatment. The vertical axis, horizontal axis, and plot symbols are consistent with... Figure 4 The same applies, so I will not elaborate further. Figure 4 (b) and Figure 5 (b) is the same chart.

[0055] Figure 6 This is a chart showing the rate of dimensional change after heat treatment of steels with different vanadium contents side-by-side. Figure 6 Two charts sharing a common vertical axis are arranged horizontally. The left chart (a) represents the dimensional change rate of heat treatment for "Std", and the right chart (b) represents the dimensional change rate of heat treatment for "VH". The vertical axis, horizontal axis, and plot symbols are consistent with... Figure 4 The same applies, so I will not elaborate further. Figure 4 (b) and Figure 6 (a) is the same chart.

[0056] Figure 7 This is a chart showing the rate of dimensional change after heat treatment of steels with different silicon contents side-by-side. Figure 7 Two charts sharing a common vertical axis are arranged horizontally. The left chart (a) represents the dimensional change rate of "SiL" after heat treatment, and the right chart (b) represents the dimensional change rate of "Std" after heat treatment. The vertical axis, horizontal axis, and plot symbols are consistent with... Figure 4 The same applies, so I will not elaborate further. Figure 4 (b) and Figure 7 (b) is the same chart.

[0057] Figure 8 This is a chart showing the rate of dimensional change after heat treatment of steels with different manganese contents side-by-side. Figure 8 Two charts sharing a common vertical axis are arranged horizontally. The left chart (a) represents the dimensional change rate of "Std" after heat treatment, and the right chart (b) represents the dimensional change rate of "MnH" after heat treatment. The vertical axis, horizontal axis, and plot symbols are consistent with... Figure 4 The same applies, so I will not elaborate further. Figure 4 (b) and Figure 8 (a) is the same chart.

[0058] When machining steel with a large anisotropy in the rate of dimensional change after heat treatment, shape differences arise due to dimensional changes caused by the heat treatment following machining. This leads to increased machining allowances or machining times during the shaping process. Therefore, as a characteristic of steel, it is desirable for each steel to have a low anisotropy in the rate of dimensional change after heat treatment, i.e., it is desirable... Figures 4 to 8 The range of ave±3s, represented in gray in the chart, is narrower.

[0059] Furthermore, since the tempering temperature is appropriately selected based on the equipment used and the mechanical properties of the manufactured mechanical parts, it is difficult to account for dimensional changes during heat treatment. Therefore, as a characteristic of steel, it is desirable that the difference in the rate of dimensional change caused by tempering temperature during heat treatment be small, i.e., it is desirable that... Figures 4 to 8The vertical axis of the chart changes less.

[0060] Based on the above judgment criteria, it can be concluded that Figure 5 (c) shows steels with a higher chromium content and Figure 6 (b) shows that steels with higher vanadium content exhibit smaller deviations in the rate of dimensional change due to orientation and temperature during heat treatment, and demonstrate favorable properties. Therefore, a target composition aimed at increasing the chromium and vanadium content was established in the preliminary study phase.

[0061] [Data Collection Phase]

[0062] Based on the target composition, 18 batches of steel (after the target composition was established) were manufactured, with chromium and vanadium added within the specified range. Composition analysis was performed on each batch of steel, and a training DB51 dataset (described later) was created to record the composition analysis data. Figure 9 Training DB51 record layout and usage. Figure 3 The table illustrating the sample composition is the same, recording the content of each component for each batch. A diagram of the record layout used for training DB51 is omitted.

[0063] Figure 9 This is an explanatory diagram illustrating the structure of the information processing device 10 used to generate model 56. The information processing device 10 includes a control unit 11, a main storage device 12, an auxiliary storage device 13, a communication unit 14, an output unit 15, an input unit 16, a reading unit 19, and a bus.

[0064] The control unit 11 is an arithmetic control device that executes the program of this embodiment. The control unit 11 uses one or more CPUs (Central Processing Units), GPUs (Graphics Processing Units), TPUs (Tensor Processing Units), or multi-core CPUs. The control unit 11 is connected to the various hardware components constituting the information processing device 10 via a bus.

[0065] The main storage device 12 is a storage device such as SRAM (Static Random Access Memory), DRAM (Dynamic Random Access Memory), or flash memory. The main storage device 12 temporarily stores information required for the processing performed by the control unit 11 and the program being executed in the control unit 11.

[0066] The auxiliary storage device 13 is a storage device such as SRAM, flash memory, hard disk, or magnetic tape. The auxiliary storage device 13 stores the training DB 51, the model 56 being generated, the program executed by the control unit 11, and various data required for program execution. The training DB 51 can also be stored on an external high-capacity storage device connected to the information processing device 10. The communication unit 14 is the interface for communication between the information processing device 10 and the network.

[0067] Output unit 15 is, for example, a liquid crystal display device or an organic EL (Electro Luminescence) display device. Input unit 16 is, for example, an input device such as a keyboard, mouse, trackball, or microphone. Output unit 15 and input unit 16 can be stacked together to form a touch panel.

[0068] Portable storage medium 96 is, for example, a USB (Universal Serial Bus) memory, a CD-ROM (CompactDisc Read-only memory), an optical disk, other optical disc media, or an SD memory card. The portable storage medium 96 stores the program 97, which will be described later.

[0069] The reading unit 19 is an interface capable of reading portable storage media 96, such as a USB connector, CD-ROM drive, or SD card reader. The semiconductor memory 98 stores program 97 and is a memory that can be installed inside the information processing device 10.

[0070] The information processing device 10 is a general-purpose personal computer, tablet computer, mainframe computer, virtual machine running on a mainframe computer, or quantum computer. The information processing device 10 can be composed of hardware consisting of multiple computers performing distributed processing, or mainframe computers, etc. The information processing device 10 can be composed of a cloud computing system. The information processing device 10 can be composed of hardware consisting of multiple computers or mainframe computers operating collaboratively.

[0071] Program 97 is stored in portable storage medium 96. Control unit 11 reads program 97 via reading unit 19 and saves it in auxiliary storage device 13. Furthermore, control unit 11 can also read program 97 stored in semiconductor memory 98. Moreover, control unit 11 can also download program 97 from another server computer (not shown) connected via communication unit 14 and a network (not shown) and save it in auxiliary storage device 13.

[0072] Program 97 is installed as a control program for the information processing device 10 and executed after being loaded into the main storage device 12. Program 97 in this embodiment is an example of a program product.

[0073] Program 97 can be built and used on an open software platform. By using an open software platform, Program 97 can be easily ported to various hardware.

[0074] Figure 10 This is an explanatory diagram illustrating Model 56. As mentioned above, Model 56 is a machine learning model trained to output data identical to the input compositional analysis data. Furthermore, if using... Figure 3 As explained, the compositional analysis data is a one-dimensional matrix arranged with numerical values ​​representing the content of each component. That is, Model 56 takes a one-dimensional matrix as input and outputs a one-dimensional matrix of the same size.

[0075] Model 56, for example, has the structure of an autoencoder that combines an encoder and a decoder. Model 56 can be generated using algorithms such as transformers or LSTM (Long Short-Term Memory).

[0076] Figure 11 This is a flowchart illustrating the processing flow of the program that generates model 56. During execution... Figure 11 Before the procedure, for example, prepare an unlearned autoencoder model. Utilize Figure 11 The program adjusts the parameters of the unlearned model to generate a model that can be used... Figure 10 Model 56 is used for explanation.

[0077] The control unit 11 acquires training records from the training DB 51 (step S601). The control unit 11 inputs the compositional analysis data contained in the acquired training records into the model 56 under training, and acquires the output one-dimensional matrix (step S602). In the following description, the matrix output from the model 56 under training will be referred to as the training output data.

[0078] The control unit 11 uses methods such as backpropagation of error to adjust the parameters of the model 56 so that the difference between the composition analysis data input into the model 56 and the training output data output from the model 56 is small (step S603).

[0079] The control unit 11 determines whether to end the parameter adjustment (step S604). For example, when learning using all the training data stored in the training DB51 has ended, the control unit 11 determines to end the parameter adjustment. When the amount of training data stored in the training DB51 is sufficient, and the learning has been repeated a predetermined number of times as specified by the hyperparameters, the control unit 11 may determine to end the process.

[0080] If the process is determined not to end (No in step S604), the control unit 11 returns to step S601. If the process is determined to end (Yes in step S604), the control unit 11 stores the model 56, including the adjusted parameters, in the auxiliary storage device 13 (step S605). Then, the control unit 11 ends the process. Thus, the generation of model 56 is complete.

[0081] [Verification Phase]

[0082] Figure 12 This is a graph showing the variation in the mean absolute error loss during the manufacturing period. The horizontal axis represents the manufacturing period of the steel. The vertical axis represents the mean absolute error loss calculated based on the steel composition analysis data and the output data from model 56 after inputting the composition analysis data into model 56. The mean absolute error loss is defined by formula (1). Based on the target composition described in the [Data Collection Phase], 18 batches of steel (18 solid circles, after the target composition determination period) were manufactured, with chromium and vanadium added within the specification range.

[0083] [Formula 1]

[0084]

[0085] N is the number of components included in the composition analysis data.

[0086] X i It is the content rate of the i-th component in the composition analysis data.

[0087] Y i It is the i-th data point in the matrix output from model 56.

[0088] Figure 12 This chart is obtained by inputting the compositional analysis data of each batch of steel into the completed model 56 and plotting the calculated mean absolute error loss. The horizontal axis represents the manufacturing period of the steel. The scale of the horizontal axis represents the number of years since the start of manufacturing the steel related to this embodiment. The scale line represents January 1st of each year. The vertical axis represents the mean absolute error loss. Solid circles represent data related to a single batch of steel.

[0089] like Figure 12 As shown in the chart, the steel used in this embodiment was manufactured starting at the end of the first year, with production frequency increased from the first half of the sixth year. Manufacturing was carried out according to the same specifications from the end of the first year to the middle of the eighth year. A preliminary research phase was conducted in the latter half of the eighth year, and the target composition was determined. Samples from the preliminary research phase are not shown in the chart. Figure 12 In the chart.

[0090] Starting at the beginning of the ninth year, the data collection phase began, and manufacturing was carried out according to the modified manufacturing conditions based on the target composition. Samples from the data collection phase are depicted in... Figure 12 The chart shows that the compositional analysis data of steel manufactured to target specifications from the beginning of the ninth year to around the autumn of the tenth year is stored in training DB51 and used to generate model 56.

[0091] Compared with the batches of steel manufactured after the target composition was established, the batches of steel manufactured after the target composition was established show that the average absolute error loss is significantly reduced.

[0092] The dashed lines represent manufacturing technicians and other experts based on Figure 12 The baseline value for the mean absolute error loss is determined by the data plotted in the image. Figure 12 In this context, the benchmark value is set to be slightly larger than the maximum value of the mean absolute error loss after setting the target. That is, the benchmark value is set to be smaller than the maximum value of the mean absolute error loss before setting the target loss.

[0093] [Application Phase]

[0094] Figure 13 This is an explanatory diagram illustrating the structure of the steel evaluation system 40. The steel evaluation system 40 includes an information processing unit 20 and an analysis unit 30. The analysis unit 30 is, for example, an ICP-OES (Inductively Coupled Plasma Optical Emission Spectrometer) used for the compositional analysis of steel.

[0095] The information processing device 20 includes a control unit 21, a main storage device 22, an auxiliary storage device 23, a communication unit 24, an output unit 25, an input unit 26, a reading unit 29, and a bus.

[0096] The control unit 21 is an arithmetic control device that executes the program of this embodiment. The control unit 21 uses one or more CPUs, GPUs, or multi-core CPUs, etc. The control unit 21 is connected to the various hardware components constituting the information processing device 20 via a bus.

[0097] The main storage device 22 is a storage device such as SRAM, DRAM, or flash memory. The main storage device 12 temporarily stores information required for the processing performed by the control unit 21 and the program being executed by the control unit 21.

[0098] The auxiliary storage device 23 is a storage device such as SRAM, flash memory, hard disk, or magnetic tape. The auxiliary storage device 23 stores the model 56, the program executed by the control unit 21, and various data required for program execution. The communication unit 24 is the interface for communication between the information processing device 20 and the network.

[0099] Output unit 25 is, for example, a liquid crystal display device or an organic EL display device. Input unit 26 is, for example, an input device such as a keyboard, mouse, trackball, or microphone. Output unit 25 and input unit 26 can be stacked together to form a touch panel.

[0100] Portable storage medium 91 is, for example, a USB memory, CD-ROM, optical disk media, other optical disc media, or SD memory card. Program 92, described later, is stored in portable storage medium 91.

[0101] The reading unit 29 is an interface capable of reading portable storage media 91, such as a USB connector, CD-ROM drive, or SD card reader. The semiconductor memory 93 stores program 92 and is a memory that can be installed inside the information processing device 20.

[0102] The information processing device 20 can be a general-purpose personal computer, tablet computer, mainframe computer, virtual machine running on a mainframe computer, or quantum computer. The information processing device 20 can be composed of hardware consisting of multiple computers performing distributed processing, or mainframe computers, etc. The information processing device 20 can be composed of a cloud computing system. The information processing device 20 can be composed of hardware consisting of multiple computers or mainframe computers operating collaboratively.

[0103] The information processing device 20 can be integrated with the analysis device 30, and also serve as the control device for the analysis device 30. The analysis device 30 can be configured to automatically send component analysis data to the information processing device 20. Users can read the component analysis data stored in the analysis device 30 into the information processing device 20 via, for example, a network drive or any storage medium.

[0104] Program 92 is stored in portable storage medium 91. Control unit 21 reads program 92 via reading unit 29 and saves it in auxiliary storage device 23. Furthermore, control unit 21 can also read program 92 stored in semiconductor memory 93. Moreover, control unit 21 can also download program 92 from another server computer (not shown) connected via communication unit 24 and a network (not shown) and save it in auxiliary storage device 23.

[0105] Program 92 is installed as a control program for the information processing device 20 and executed after being loaded into the main storage device 22. Program 92 in this embodiment is an example of a program product.

[0106] Figure 14 This is a flowchart illustrating the processing flow of the procedure during the application phase. The control unit 21 acquires composition analysis data from the analysis device 30 or a network driver, etc. (step S501). The control unit 21 determines whether the content of each component is within the specified range (step S502).

[0107] When it is determined that there are components outside the specified range (No in step S502), the control unit 21 notifies that the steel related to the composition analysis data obtained in step S501 is a non-conforming product (step S511). The notification may be sent to automated equipment, which then marks the steel as unshippable. The notification may also be sent to a manufacturing engineer.

[0108] When it is determined that all components are within the specified range (Yes in step S502), the control unit 21 inputs the composition analysis data obtained in step S501 into the model 56 and obtains the output data output from the model 56 (step S503). The control unit 21 calculates the error loss between the composition analysis data and the output data, for example, according to formula (1) (step S504).

[0109] Control unit 21 determines whether the calculated error loss is below a reference value (step S505). When it is determined to be below the reference value (yes in step S505), control unit 21 notifies that the steel related to the composition analysis data obtained in step S501 is of excellent quality (step S512). The notification may be sent to an automated device, which then marks the steel as excellent. The notification may also be sent to a manufacturing engineer.

[0110] When the value is determined to be greater than the benchmark value (No in step S505), the control unit 21 notifies that the steel related to the composition analysis data obtained in step S501 is neither a superior product nor a defective product, but a near-superior product (step S513). This notification is sent, for example, to the manufacturing engineer. The manufacturing engineer then uses the information... Figure 2 Various additional checks are required to provide explanations.

[0111] After step S511, step S512 or step S513 is completed, the control unit 21 ends the processing.

[0112] According to this embodiment, a method for evaluating steel can be provided that allows for a more detailed evaluation of steel that has been deemed acceptable by existing inspection methods. Since it is easy to identify superior steel with particularly good properties and other near-superior steel among the acceptable steel that meets specifications, superior steel can be prioritized for delivery to customers requiring particularly high reliability.

[0113] According to this embodiment, by setting a target composition, the frequency of producing high-quality products can be increased. Therefore, the quality of steel can be improved.

[0114] According to this embodiment, by applying the compositional analysis data of steel manufactured before the target composition is determined to the baseline value, the baseline value can be quickly determined without the need to manufacture and test samples for comparative studies after the data collection phase is completed.

[0115] According to this embodiment, by utilizing reference values, deviations in the composition during the manufacturing process can be quickly detected, resulting in the production of steel with stable quality.

[0116] Error loss can be achieved using the mean squared error (LOSS MSE). The mean squared error is defined by equation (2).

[0117] [Formula 2]

[0118]

[0119] N is the number of components included in the composition analysis data.

[0120] X i It is the content rate of the i-th component in the composition analysis data.

[0121] Y i It is the i-th data point in the matrix output from model 56.

[0122] Even when the error loss is defined using mean squared error loss, the error loss decreases when the composition of the input data is similar to the training data used to generate Model 56. Conversely, the error loss increases when the composition of the input data differs significantly from the training data.

[0123] Other error losses can use any loss function used for generating and evaluating machine learning models, such as the square root of the mean squared error loss, the mean squared logarithmic error loss (LOSS MSLE), the square root of the mean squared logarithmic error loss, Huber loss, Poisson loss, hinge loss, or KL divergence (KLD).

[0124] Furthermore, when using the trained DB51 to adjust the parameters of model 56 via machine learning ( Figure 11 The loss function used in step S603 can be the same as or a different function than the error loss used when setting the baseline during the validation phase. Figure 14 The same function is used in step S504.

[0125] Computer programs can be configured on a single computer or at a single site, or executed on multiple computers distributed across multiple sites and interconnected by a communication network.

[0126] The technical features (structural elements) described in each embodiment can be combined with each other to form new technical features.

[0127] All aspects of the embodiments disclosed herein should be considered as examples rather than limiting descriptions. The scope of the invention is defined by the claims rather than by the foregoing meaning, and is intended to include all modifications within the meaning and scope equivalent to the claims.

[0128] The independent and dependent claims recited in the claims statement can be combined with each other in any combination, regardless of the form of reference. Furthermore, although the claims statement employs the form of a claim referencing two or more other claims (multiple claim form), it is not limited to this. It may also employ the form of a multiple claim that refers to at least one of the multiple claims (multiple-reference multiple-claim form).

[0129] Explanation of reference numerals in the attached figures

[0130] 10. Information processing device

[0131] 11 Control Department

[0132] 12 Main storage devices

[0133] 13. Auxiliary storage device

[0134] 14 Ministry of Communications

[0135] 15 Output Section

[0136] 16 Input Section

[0137] 19 Reading Department

[0138] 20 Information processing devices

[0139] 21 Control Department

[0140] 22 Main storage device

[0141] 23. Auxiliary storage device

[0142] 24 Ministry of Communications

[0143] 25 Output Section

[0144] 26 Input Section

[0145] 29 Reading Department

[0146] 30 Analytical apparatus

[0147] 40 Steel Evaluation System

[0148] 51 Training the DB

[0149] 56 Model

[0150] 91 Portable storage media

[0151] 92 Program

[0152] 93 Semiconductor memory

[0153] 96 Portable storage media

[0154] 97 Program

[0155] 98 Semiconductor memory

Claims

1. A method for evaluating steel, characterized in that, Analyze the composition of the steel; When the composition data obtained through analysis meets the predetermined specifications, the composition data is input into a model that is machine-learned to output the same data as the input data, and the error loss between the output data from the model and the composition data is calculated. as well as Determine whether the error loss exceeds a predetermined reference value.

2. The method for evaluating steel according to claim 1, characterized in that, When the error loss is determined to exceed a predetermined benchmark value, the steel is subjected to additional inspection.

3. The method for evaluating steel according to claim 1, characterized in that, The model is an autoencoder model that includes an encoder and a decoder.

4. The method for evaluating steel according to any one of claims 1 to 3, characterized in that, The model used the composition data of steel manufactured after changing the manufacturing process as training data for machine learning. The baseline value is set to be greater than the maximum error loss between the output data and the component data when the component data contained in the training data are input into the model after machine learning and output from the model, and less than the maximum error loss between the output data and the component data when the component data of the steel manufactured before the change of manufacturing process are input into the model after machine learning and output from the model.

5. A program, characterized in that, The computer will perform the following processes: Obtain the composition data of the steel; When the component data meets the predetermined specifications, the component data is input into a model that is machine-learned to output the same data as the input data, and the error loss between the output data from the model and the component data is calculated. as well as Determine whether the error loss exceeds a predetermined reference value.

6. A method for generating a model, characterized in that, Acquire training data, which records multiple sets of composition data representing the content of each component in the steel; and Based on the acquired training data, a model is generated through machine learning. When the model is given input component data, it outputs data that is identical to the input component data.

Images