Processing condition search device, computer-readable recording medium, and processing condition search method

Abstract

The processing condition search device (120) has: a parameter sorting section (125) that sorts a plurality of parameters into a plurality of variable parameters and one or more fixed parameters; a first dimension reduction section (126) that generates a first characteristic quantity of a first dimension or less from the plurality of variable parameters; a second dimension reduction section (127) that generates a second characteristic quantity of a second dimension or less from the one or more fixed parameters; a machine learning section (128) that performs learning of a relationship of the first characteristic quantity, the second characteristic quantity, and a plurality of evaluation values, thereby generating a learning model; a third dimension processing section (131) that generates a third characteristic quantity of a second dimension or less from the one or more fixed parameters, that is, one or more target fixed parameters; an optimum processing condition search section (132) that searches for optimum values of characteristic quantities of a plurality of target variable parameters using the third characteristic quantity and the learning model; and a dimension restoration section (133) that determines a search processing condition from the optimum values and the one or more target fixed parameters.

Description

Technical Field

[0001] This invention relates to a processing condition search device, a computer-readable recording medium, and a processing condition search method. Background Technology

[0002] Industrial processing machines perform prescribed processing on raw materials, thereby changing their shape or state. Examples of such processing machines include machine tools that cut or grind raw materials, and factory equipment that performs processes such as mixing, reacting, heating, cooling, drying, or calcining of raw materials.

[0003] Generally, these machining centers can be configured with multiple parameters to reflect the user's intent. The machining result depends on the combination of these parameters, i.e., the machining conditions. Therefore, to obtain the desired machining result, appropriate machining conditions need to be set for the machining center.

[0004] However, when multiple parameters exist, and each parameter can be set to continuous or discrete values ​​in stages, the number of possible combinations becomes enormous. Therefore, discovering the processing conditions that yield the desired processing results requires a significant amount of labor and time for trial and error.

[0005] In the past, regarding this parameter, the predicted value was obtained by calculating the predicted value corresponding to the evaluation value of the processing result and the processing conditions corresponding to the evaluation value. The optimal processing conditions were then calculated based on the predicted value. However, the larger the dimension of the parameter, the more difficult it is to search for the global optimal value.

[0006] Therefore, Patent Document 1 proposes the following method: for high-dimensional data, use principal component analysis and other methods to extract its feature quantities, reduce the dimensionality of the extracted feature quantities, and thus make it easier to handle the problem.

[0007] Existing technical documents

[0008] Patent documents

[0009] Patent Document 1: Japanese Patent Publication No. 2012-509190 (paragraph 0014) Summary of the Invention

[0010] The problem that the invention aims to solve

[0011] However, in existing dimensionality reduction, the result of extracting features from all parameters is that all parameters become the search object. Therefore, there is a problem: it cannot be applied to situations where no changes have been made to some parameters, depending on the user's intention or the conditions of the processing site.

[0012] For example, when the processing result depends not only on control parameters but also on characteristic parameters related to the properties of the raw materials, such as size or specific gravity, or environmental parameters related to the processing environment, such as temperature and humidity, it is necessary to consider these parameters to search for optimal processing conditions. However, in the case of dimensional reduction, these unchangeable parameters also become objects of change.

[0013] Therefore, one or more aspects of the present invention aim to enable a global search for optimal processing conditions, even when parameters that cannot be changed exist, in the processing conditions used for processing.

[0014] Methods for solving problems

[0015] One aspect of the processing condition search apparatus of the present invention is characterized by comprising: a processing result evaluation storage unit that stores processing result evaluation information representing multiple processing conditions having multiple parameters and multiple evaluation values ​​for multiple processing results under the multiple processing conditions; a parameter sorting unit that sorts the multiple parameters into multiple variable parameters that can be changed and one or more fixed parameters that cannot be changed; a first dimension processing unit that generates a first feature quantity below a predetermined first dimension based on the multiple variable parameters, thereby generating one or more first feature quantities corresponding to the multiple processing conditions; and a second dimension processing unit that generates a second feature quantity below a predetermined second dimension based on the one or more fixed parameters, thereby generating one or more first feature quantities corresponding to the multiple processing conditions. The system comprises: one or more second feature quantities corresponding to the processing conditions; a machine learning unit that learns the relationship between the one or more first feature quantities, the one or more second feature quantities, and the plurality of evaluation values, thereby generating a learning model; a third dimension processing unit that generates a third feature quantity below the second dimension based on one or more fixed parameters used in the processing conditions that serve as the search object, i.e., one or more object fixed parameters; a search unit that uses the third feature quantity and the learning model to search for the optimal value of the feature quantity of the plurality of variable parameters used in the object processing conditions, i.e., the plurality of object variable parameters; and a determination unit that determines the processing condition searched as the object processing condition, i.e., the search processing condition, based on the optimal value and the one or more object fixed parameters.

[0016] One aspect of the program of the present invention is characterized in that the program enables a computer to function as: a processing result evaluation storage unit that stores processing result evaluation information representing multiple processing conditions having multiple parameters and multiple evaluation values ​​for multiple processing results under the multiple processing conditions; a parameter sorting unit that sorts the multiple parameters into multiple variable parameters that can be changed and one or more fixed parameters that cannot be changed; a first dimension processing unit that generates a first feature quantity below a predetermined first dimension based on the multiple variable parameters, thereby generating one or more first feature quantities corresponding to the multiple processing conditions; and a second dimension processing unit that generates a second feature quantity below a predetermined second dimension based on the one or more fixed parameters, thereby generating one or more first feature quantities corresponding to the multiple processing conditions. The system comprises: one or more second feature quantities corresponding to the processing conditions; a machine learning unit that learns the relationship between the one or more first feature quantities, the one or more second feature quantities, and the plurality of evaluation values, thereby generating a learning model; a third dimension processing unit that generates a third feature quantity below the second dimension based on one or more fixed parameters used in the processing conditions that serve as the search object, i.e., one or more object fixed parameters; a search unit that uses the third feature quantity and the learning model to search for the optimal value of the feature quantity of the plurality of variable parameters used in the object processing conditions, i.e., the plurality of object variable parameters; and a determination unit that determines the processing condition searched as the object processing condition, i.e., the search processing condition, based on the optimal value and the one or more object fixed parameters.

[0017] One aspect of the processing condition search method of the present invention is characterized by: classifying multiple parameters included in processing result evaluation information into multiple variable parameters that can be changed and one or more fixed parameters that cannot be changed; the processing result evaluation information represents multiple processing conditions having the multiple parameters and multiple evaluation values ​​for multiple processing results under the multiple processing conditions; generating a first feature quantity below a predetermined first dimension based on the multiple variable parameters, thereby generating one or more first feature quantities corresponding to the multiple processing conditions; generating a second feature quantity below a predetermined second dimension based on the one or more fixed parameters, thereby generating one or more second feature quantities corresponding to the multiple processing conditions; learning the relationship between the one or more first feature quantities, the one or more second feature quantities, and the multiple evaluation values, thereby generating a learning model; generating a third feature quantity below the second dimension based on one or more fixed parameters used in the processing condition (i.e., object processing condition) that is the search object; using the third feature quantity and the learning model, searching for the optimal value of the feature quantity of the multiple variable parameters (i.e., multiple object variable parameters) used in the object processing condition; and determining the processing condition (i.e., the search processing condition) as the object processing condition based on the optimal value and the one or more object fixed parameters.

[0018] Invention Effects

[0019] According to one or more embodiments of the present invention, in the processing conditions used for processing, even when there are parameters that cannot be changed, it is also possible to search for the optimal global processing conditions regarding parameters that can be changed. Attached Figure Description

[0020] Figure 1 This is a block diagram that roughly shows the structure of the processing system according to embodiments 1 to 6.

[0021] Figure 2 This is a block diagram that schematically shows the structure of the processing condition search device in embodiments 1 to 4.

[0022] Figure 3 This is a schematic diagram illustrating an example of processing result evaluation information.

[0023] Figure 4 (A) and (B) are schematic diagrams showing examples of parameter data representing parameters selected by the parameter sorting section.

[0024] Figure 5 This is a block diagram illustrating an example of the hardware structure of a processing condition search device.

[0025] Figure 6This is a flowchart illustrating the operation of the processing system according to Embodiment 1.

[0026] Figure 7 This is a schematic diagram used to illustrate the search method in Implementation Method 1.

[0027] Figure 8 This is a block diagram that roughly shows the structure of the parameter sorting unit in Embodiment 2.

[0028] Figure 9 (A) and (B) are schematic diagrams illustrating examples of low and high correlations between Qx and Rx.

[0029] Figure 10 This is a flowchart illustrating an example of the parameter sorting operation in the parameter sorting unit of Embodiment 2.

[0030] Figure 11 This is a block diagram that roughly shows the structure of the parameter sorting unit in Embodiment 3.

[0031] Figure 12 This is a flowchart illustrating an example of the parameter allocation operation in the parameter allocation unit of Embodiment 3.

[0032] Figure 13 This is a flowchart illustrating the operation of the optimal processing condition search unit during the initial search in Embodiment 4.

[0033] Figure 14 This is a block diagram that schematically illustrates the structure of the processing condition search device in Embodiment 5.

[0034] Figure 15 This is a flowchart illustrating the operations of the first dimension reduction section, the second dimension reduction section, the fourth dimension reduction section, the first comparison section, and the second comparison section in Embodiment 5.

[0035] Figure 16 This is a block diagram that schematically illustrates the structure of the processing condition search device in Embodiment 6.

[0036] Figure 17 This is a flowchart illustrating the operations of the first dimension reduction section, the second dimension reduction section, the fourth dimension reduction section, the synthesis section, and the comparison section in Embodiment 6. Detailed Implementation

[0037] Implementation Method 1

[0038] Figure 1 This is a block diagram that schematically illustrates the structure of the processing system 100 of Embodiment 1.

[0039] The machining system 100 includes a machining machine 110 and a machining condition search device 120.

[0040] The processing machine 110 performs processing using processing conditions from the processing condition search device 120, and provides processing result information, i.e., processing result information, to the processing condition search device 120.

[0041] The processing condition search device 120 receives processing result information under the processing conditions set for the processing machine 110 and searches for processing conditions suitable for the processing machine 110.

[0042] The processing conditions consist of multiple parameters.

[0043] Figure 2 This is a block diagram that roughly shows the structure of the processing condition search device 120.

[0044] The processing condition search device 120 includes a processing result acquisition unit 121, a processing result evaluation unit 122, a processing result evaluation storage unit 123, a sorting flag storage unit 124, a parameter sorting unit 125, a first dimension reduction unit 126, a second dimension reduction unit 127, a machine learning unit 128, a model storage unit 129, a fixed parameter storage unit 130, a third dimension reduction unit 131, an optimal processing condition search unit 132, a dimension restoration unit 133, and a processing condition instruction unit 134.

[0045] The processing result acquisition unit 121 acquires processing result information, i.e., processing result information, from the processing machine 110. The acquired processing result information is provided to the processing result evaluation unit 122.

[0046] The types of processing result information vary depending on the type of processing machine 110 or the purpose of processing. For example, consider processing result information as inspection data of the processed workpiece, where the inspection result value is the error or defect rate from the target value determined by the processing specifications.

[0047] Furthermore, in Embodiment 1, the processing result acquisition unit 121 acquires processing result information from the processing machine 110; however, Embodiment 1 is not limited to this example. For example, processing result information may also be acquired from an inspection machine or the like, which is different from the processing machine 110. Additionally, the user may input processing result information via an input unit (not shown).

[0048] The processing result evaluation unit 122 evaluates the processing result performed by the processing machine 110, and determines the evaluation value. The determined evaluation value is added to the processing result evaluation information described later, corresponding to the processing conditions used when the processing was performed, i.e., the search processing conditions.

[0049] For example, the processing result evaluation unit 122 evaluates the processing result shown in the processing result information from the processing result acquisition unit 121 and determines its evaluation value. The evaluation value can be a numerical value such as a continuous value or a discrete value, a category value representing an attribute, or a logical value representing the truth or falsity of a proposition.

[0050] Here, the evaluation value indicates whether the processing result is good or bad. For example, the evaluation value can be a continuous or discrete value representing the degree of processing quality. As a specific example, there is a defect rate that uses a continuous value of 0 to 1 to represent the proportion of defective products. In this case, the smaller the value, the better the processing result.

[0051] In addition, the evaluation value can also be a category that indicates whether the processing result is good or bad, or a logical value that indicates whether a predetermined proposition is true or false.

[0052] Then, the processing result evaluation unit 122 stores the evaluation value and the corresponding processing conditions in the processing result evaluation storage unit 123.

[0053] The processing result evaluation storage unit 123 stores processing result evaluation information representing multiple processing conditions and multiple evaluation values ​​for multiple processing results under these multiple processing conditions. As described above, each of the multiple processing conditions has multiple parameters.

[0054] For example, in the processing result evaluation information, by default, processing conditions that differ from the processing conditions searched by the processing condition search device 120 and the evaluation values ​​of the processing results under those processing conditions are stored together. This information can be input via an input unit not shown.

[0055] Furthermore, in the processing result evaluation information, the processing conditions indicated by the processing condition instruction unit 134 to the processing machine 110 and the evaluation value determined by the processing result evaluation unit 122 for the processing conditions are stored together.

[0056] In addition, the processing result evaluation information includes all the processing performed by the processing condition search device 120, and this information is stored in the processing result evaluation storage unit 123.

[0057] Figure 3 This is a schematic diagram illustrating an example of processing result evaluation information.

[0058] like Figure 3 As shown, the processing result evaluation information 101 becomes a matrix that gathers together the various parameters constituting the processing conditions and their corresponding evaluation values.

[0059] exist Figure 3In the example shown, for each of the past N processing operations (N being an integer greater than or equal to 1), a processing identification information, namely a processing number, is assigned to identify each processing operation. Furthermore, according to each processing number, the M types of parameters used in the processing (M being an integer greater than or equal to 2) and their corresponding evaluation values ​​are arranged in a vertical column of a matrix.

[0060] When implementing a new process, a new column is added to the right end of the matrix to record the processing conditions and their evaluation values.

[0061] Specific examples of parameters include control parameters of the processing machine 110, material parameters representing properties such as the type or characteristic values ​​of the material, or environmental parameters such as temperature and humidity at the processing site. These parameters can be numerical values ​​such as continuous or discrete values, categorical values ​​representing attributes, or logical values ​​representing the truth or falsity of propositions.

[0062] The sorting flag storage unit 124 stores sorting flags that indicate whether a parameter is variable or fixed, according to each category of the multiple parameters, in order to sort multiple parameters.

[0063] For example, the sorting mark storage unit 124 stores sorting marks indicating whether the various parameters constituting the processing conditions are variable parameters that can be changed or fixed parameters that cannot be changed. In other words, variable parameters are parameters that can be changed, while fixed parameters are parameters that cannot be changed.

[0064] The sorting flag can also be set by receiving user instructions via an input unit (not shown). Furthermore, the sorting flag can be set automatically based on conditions such as the type or model of the processing machine 110. Additionally, the sorting flag can also be received from other devices via a communication unit (not shown).

[0065] The fixed parameters include control parameters of the processing machine 110 that cannot be changed or that the user does not want to change, material properties, dimensions or quantities and other material-related parameters, or environmental parameters such as air pressure, temperature and humidity of the processing environment.

[0066] The parameter sorting unit 125 sorts the multiple parameters contained in the processing result evaluation information into multiple variable parameters that can be changed and one or more fixed parameters that cannot be changed. Here, the parameter sorting unit 125 sorts the multiple parameters into multiple variable parameters and one or more fixed parameters by referring to sorting marks.

[0067] For example, the parameter sorting unit 125 reads multiple parameters contained in the processing result evaluation information stored in the processing result evaluation storage unit 123 and sorting marks stored in the sorting mark storage unit 124. The parameter sorting unit 125 sorts the parameters into variable parameters or fixed parameters according to the sorting marks. Then, the parameter sorting unit 125 generates variable parameter data representing the sorted variable parameters and fixed parameter data representing the sorted fixed parameters.

[0068] Figure 4 (A) and (B) are schematic diagrams showing examples of parameter data representing parameters sorted by parameter sorting unit 125.

[0069] Figure 4 (A) is an example of variable parameter data 102 that stores variable parameters. Figure 4 (B) is an example of fixed parameter data 103 that stores fixed parameters.

[0070] Variable parameter data 102 stores Mv types of parameters, and fixed parameter data 103 stores Mf types of parameters.

[0071] and Figure 3 Similarly, the processing result evaluation information 101 shown also relates to the past N processing operations. The variable parameter data 102 and fixed parameter data 103 are composed of matrices that maintain parameters according to each processing number. Mv and Mf correspond to the dimensions of the variable parameters and fixed parameters, respectively.

[0072] In variable parameter data 102, the y-th variable parameter in processing number x is set to q. xy In fixed parameter data 103, the z-th fixed parameter in processing number x is set to r. xz They are in Figure 3 The processing conditions stored in the processing result evaluation information 101 shown are extracted from the column of processing number x.

[0073] return Figure 2 The first dimension reduction unit 126 is a first dimension processing unit that generates a first feature quantity below a predetermined first dimension based on multiple variable parameters contained in the variable parameter data, thereby generating one or more first feature quantities corresponding to multiple processing conditions. Here, when the dimensions of the multiple variable parameters are larger than the first dimension, the first dimension reduction unit 126 reduces the dimensions of the multiple variable parameters, thereby generating the first feature quantity.

[0074] For example, the first dimension reduction unit 126 analyzes the variable parameter data generated by the parameter sorting unit 125 and determines whether the dimension Mv of the variable parameter data is greater than a predetermined threshold THv. Then, if the dimension Mv is greater than the threshold THv, the first dimension reduction unit 126 performs a dimension reduction process, namely, a first dimension reduction process, to convert the variable parameter data into first feature quantity data represented by a dimension Lv below the threshold THv. Here, the threshold THv corresponds to the first dimension. Furthermore, if the dimension Lv is 2 or more, the first feature quantity of multiple dimensions contained in the first feature quantity data is also referred to as the first feature quantity set.

[0075] Specifically, the element of the x-th dimension of the first feature data corresponding to the processing number n is set to av. nx At that time, av nx Become a variable parameter q n1 q n2 , ..., q nMv The function is therefore expressed using the following equation (1).

[0076] av nx =fx(q n1 q n2 , ..., q nMv (1)

[0077] Here, fx represents a function that transforms the variable parameters into elements of the x-th dimension of the first feature data. These elements are called the first feature.

[0078] As a method for reducing this dimensionality, consider using principal component analysis (PCA). In this case, the principal components obtained through PCA become eigenvalues. If we extract the first to the kth principal components in descending order of the intrinsic values ​​of the covariance matrix and remove the remaining principal components, we can reduce the dimensionality. Let k be the eigenvalue. <Mv。

[0079] Autoencoders using neural networks are also a preferred example of dimensionality reduction processing. In this case, the output of the encoder network of the autoencoder becomes the feature quantity. Here, the encoder network is part of the neural network that constitutes the autoencoder, meaning a sub-network related to the encoder processing.

[0080] The predicted value is obtained by calculating the predicted value based on the evaluation value of the processing result and the processing conditions corresponding to the evaluation value, and the evaluation value corresponding to the processing conditions without processing. The technique of calculating the optimal processing conditions based on the predicted value is called black box optimization. As a type of black box optimization, namely Bayesian optimization, there are known methods such as Random Embedding Bayesian Optimization (REMBO), which uses a random matrix to embed a high-dimensional matrix into a low-dimensional space, or Line Bayesian Optimization (LINEBO), which restricts the search space to a one-dimensional space. However, these methods can also be used as the dimension reduction method in Implementation Method 1.

[0081] Here, REMBO is described in detail in reference 1 below, and LINEBO is described in detail in reference 2 below.

[0082] Document 1: Wang, Ziyu, et al. "Bayesian optimization in high dimensions viarandom embeddings." Twenty-Third International Joint Conference on ArtificialIntellgence.2013

[0083] Document 2: Kirschner, Johannes, et al. "Adaptive and Safe BayesianOptimization in High Dimensions via One-Dimensional Subspaces." arXiv preprintarXiv: 1902.03229 (2019)

[0084] In addition, other dimensionality reduction methods can be used, such as multidimensional scaling, independent component analysis, nonnegative matrix factor analysis (NMF), locally linear embedding (LLE), locality-preserving projection (LPP), Laplace intrinsic mapping (LEP), core principal component analysis, Karhunen-Loeve expansion, and t-SNE (t-distributed stochastic neighbor embedding).

[0085] Furthermore, when the dimension Mv of the variable parameter data is below the threshold THv, the first dimension reduction unit 126 does not perform dimension reduction and sets the variable parameter data itself as the first feature data.

[0086] Then, the first dimension reduction unit 126 provides the first feature data to the machine learning unit 128.

[0087] The second dimension reduction unit 127 is a second dimension processing unit that generates a second feature quantity below a predetermined second dimension based on one or more fixed parameters shown in the fixed parameter data, thereby generating one or more second feature quantities corresponding to multiple processing conditions. Here, when the dimensions of the multiple fixed parameters are larger than the second dimension, the second dimension reduction unit 127 reduces the dimensions of the multiple fixed parameters, thereby generating the second feature quantity.

[0088] For example, the second dimension reduction unit 127 analyzes the fixed parameter data generated by the parameter sorting unit 125 and determines whether the dimension Mf of the fixed parameter data is greater than a predetermined threshold THf. If the dimension Mf is greater than the threshold THf, the second dimension reduction unit 127 performs a dimension reduction process, namely, a second dimension reduction process, to convert the fixed parameter data into second feature quantity data represented by a dimension Lf below the threshold THf. The specific dimension reduction process is the same as that of the first dimension reduction unit 126. Here, the dimension Mf corresponds to the second dimension. Furthermore, when the dimension Lf is 2 or more, the second feature quantity containing multiple dimensions in the second feature quantity data is also referred to as the second feature quantity set.

[0089] Let the element of the x-th dimension of the second feature data with fixed parameters for processing number n be af. nx At that time, af nx Become a fixed parameter value r n1 r n2 , ..., r nMf The function is therefore expressed using the following equation (2).

[0090] af nx =hx(r n1 r n2 , ..., r nMf (2)

[0091] Here, hx represents a function that transforms fixed parameters into elements of the x-th dimension of the second feature data. These elements are called the second feature.

[0092] Furthermore, when the dimension Mf of the fixed parameter data is below the threshold THf, the second dimension reduction unit 127 does not perform dimension reduction and sets the fixed parameter data itself as the second feature data.

[0093] Then, the second dimension reduction unit 127 provides the second feature data to the machine learning unit 128.

[0094] Furthermore, since the fixed parameters are not the search objects in the optimal processing condition search unit 132 described later, when the dimension Mf of the fixed parameter data is large, the second dimension reduction process can be omitted, and the fixed parameter data itself can be set as the second feature data.

[0095] Machine Learning Unit 128 learns the relationship between one or more first features, one or more second features, and multiple evaluation values, thereby generating a learning model.

[0096] For example, the machine learning unit 128 learns the relationship between the first feature data provided by the first dimension reduction unit 126, the second feature data provided by the second dimension reduction unit 127, and the evaluation values ​​contained in the processing result evaluation information stored in the processing result evaluation storage unit 123, treating each feature as an input value and the evaluation value as a response value, and generates a learning model that uses a mathematical model to represent them.

[0097] For example, if the evaluation value is a continuous or discrete numerical value, the learning model can apply a regression model; if the evaluation value is a categorical or logistic value, the learning model can apply a classification model. Given new input values—the values ​​of each feature—if these are input into the learning model, it can calculate predicted evaluation values ​​for the processing results of those input values. Specific examples of learning algorithms used to generate such learning models include linear regression, nonlinear regression, regression trees, model trees, support vector regression, genetic programming, Gaussian process regression, linear discriminant analysis, logistic regression, k-nearest neighbors, support vector machines, decision trees, random forests, or neural networks.

[0098] The model storage unit 129 stores the learning models generated by the machine learning unit 128.

[0099] The fixed parameter storage unit 130 stores one or more fixed parameters used in the processing conditions, i.e., object processing conditions, which are the search objects. The one or more fixed parameters stored in the fixed parameter storage unit 130 are also referred to as one or more object fixed parameters. Furthermore, the data representing the one or more fixed parameters stored in the fixed parameter storage unit 130 is also referred to as fixed parameter data or object fixed parameter data.

[0100] For example, the fixed parameters stored in the fixed parameter storage unit 130 can be set by the user's instruction, or automatically set under specific conditions, or input from other devices via a communication unit not shown.

[0101] Furthermore, the sorting marks stored in the sorting mark storage unit 124 or the fixed parameters stored in the fixed parameter storage unit 130 can be changed after all the search steps described later have been performed.

[0102] The third-dimensional reduction unit 131 is a third-dimensional processing unit that generates a third feature quantity of less than the second dimension based on one or more fixed parameters stored in the fixed parameter storage unit 130. Here, when the dimension of the multiple fixed parameters is larger than the second dimension, the third-dimensional reduction unit 131 reduces the dimension of the multiple fixed parameters, thereby generating the third feature quantity.

[0103] For example, the third-dimensional reduction unit 131 analyzes the fixed parameter data stored in the fixed parameter storage unit 130 and determines whether the dimension Mf of the fixed parameter data is greater than the threshold THf. If the dimension Mf is greater than the threshold THf, the third-dimensional reduction unit 131 performs a dimension reduction process, namely, a third-dimensional reduction process, to convert the fixed parameter data into third feature data represented by a dimension Lf less than or equal to Mf. The third-dimensional reduction process is the same as the second-dimensional reduction process performed by the second-dimensional reduction unit 127. The third feature data is provided to the optimal processing condition search unit 132. Furthermore, if the dimension Lf is 2 or greater, the multiple third features contained in the third feature data are also referred to as a third feature set.

[0104] If the second dimension reduction process is principal component analysis, then the third dimension reduction unit 131 can extract the same number of principal components as the second feature data using the intrinsic values ​​and intrinsic vectors used at this time.

[0105] Furthermore, when an autoencoder is used in the second dimension reduction process, the third dimension reduction unit 131 can input fixed parameters to the same encoder network as the second dimension reduction unit 127 and set its output as the third feature quantity.

[0106] Furthermore, when the dimension Mf of the fixed parameter data is below the specified threshold THf, the third dimension reduction unit 131 does not perform third dimension reduction processing, but provides the fixed parameter data itself read from the fixed parameter storage unit 130 as the third feature data to the optimal processing condition search unit 132.

[0107] The optimal processing condition search unit 132 is a search unit that searches for the optimal values ​​of multiple variable parameters, i.e., multiple object variable parameters, used in the object processing conditions using a third feature quantity and a learning model.

[0108] For example, the optimal processing condition search unit 132 uses the learning model stored in the model storage unit 129 to search for optimal processing conditions. At this time, the optimal processing condition search unit 132 provides the third feature data provided by the third dimension reduction unit 131 and candidate features of multiple variable parameters generated using a predetermined method as input to the learning model, obtaining a predicted value as the evaluation value obtained by the learning model in response to this input. Then, the optimal processing condition search unit 132 provides the candidate that gives the best predicted value as the optimal processing condition to the dimension restoration unit 133. Furthermore, the candidates included in the optimal processing condition are equivalent to the optimal value.

[0109] The dimension restoration unit 133 is a determination unit that determines the processing conditions, i.e., the search processing conditions, as object processing conditions based on the optimal value and one or more fixed parameters of the object. Here, when the dimensions of the multiple variable parameters are larger than the first dimension, the dimension restoration unit 133 restores the parameters from the optimal value in a manner that makes them the same as the dimensions of the multiple variable parameters.

[0110] For example, if the dimension Mv of the variable parameter data is larger than the threshold THv, the dimension restoration unit 133 will convert the optimal processing conditions provided by the optimal processing condition search unit 132 into variable parameters. For example, the variable parameter value of the x-th dimension after conversion will be set to q. x * Let the element of the y-th dimension of the variable parameter output as the optimal processing condition be av. y * At that time, q x * Become av1 * av2 * , ...,av Lv * The function is therefore expressed using the following equation (3).

[0111] q x * =g(av1) * av2 * , ...,av Lv * (3)

[0112] Here, g represents a function that converts the feature quantity into a variable parameter. For example, if the dimensionality compression process is principal component analysis, the dimensionality restoration unit 133 can convert the optimal processing conditions into variable parameters using the intrinsic values ​​and intrinsic vectors used in the dimensionality compression process.

[0113] Furthermore, if the dimensionality compression process is an autoencoder, then the dimensionality restoration unit 133 can obtain variable parameters as output simply by inputting feature quantities into the decoder network. Here, the decoder network is part of the neural network that constitutes the autoencoder, meaning a sub-network related to the decoder processing.

[0114] Furthermore, if the dimension Mv of the variable parameter data is smaller than the threshold THv, the dimension restoration unit 133 will not perform dimension restoration, but will directly set the optimal processing conditions provided by the optimal processing condition search unit 132 as the variable parameters.

[0115] The processing condition instruction unit 134 provides the processing condition search to the processing machine 110, so that the processing machine 110 performs processing under the processing condition search, and adds the processing condition search to the processing result evaluation information.

[0116] For example, the processing condition instruction unit 134 combines the variable parameters provided by the dimension restoration unit 133 and the fixed parameters read from the fixed parameter storage unit 130 to set the processing conditions, and instructs the processing machine 110 to perform processing under these processing conditions. In addition, the processing condition instruction unit 134 stores the processing conditions in the processing result evaluation information stored in the processing result evaluation storage unit 123.

[0117] At this time, the user can also arbitrarily modify the processing conditions via an input unit (not shown). In this case, the user implements the modified processing conditions, which are then output from the processing condition instruction unit 134 to the processing machine 110 and the processing result evaluation storage unit 123.

[0118] As described above, after processing conditions are provided from the processing condition search device 120, the processing machine 110 performs processing according to those processing conditions. Then, the processing machine 110 provides processing result information, representing the result of the processing, to the processing condition search device 120.

[0119] Next, the hardware structure of the processing condition search device 120 will be described.

[0120] Figure 2 The processing result acquisition unit 121, processing result evaluation unit 122, parameter sorting unit 125, first dimension reduction unit 126, second dimension reduction unit 127, machine learning unit 128, third dimension reduction unit 131, optimal processing condition search unit 132, dimension restoration unit 133 and processing condition instruction unit 134 shown can be implemented by processing circuits.

[0121] The processing circuitry can be a circuit with a processor or dedicated hardware. Furthermore, it can be implemented in a distributed computing environment where they are connected on computer networks such as the cloud. In other words, the processing condition search device 120 can also be implemented using a computer.

[0122] The processing result evaluation storage unit 123, the sorting mark storage unit 124, the model storage unit 129, and the fixed parameter storage unit 130 can be implemented by a storage device.

[0123] Storage devices are semiconductor memories such as DRAM (Dynamic Random Access Memory), SRAM (Static Random Access Memory), or flash memory, or recording media such as disks, optical discs, or magnetic tapes, or data storage devices on computer networks.

[0124] Figure 5 This is a block diagram illustrating an example of the hardware structure of the processing condition search device 120.

[0125] The aforementioned processing circuit 140 includes, for example, a processor 141 and a memory 142.

[0126] When each part of the processing condition search device 120 is implemented by the processing circuit 140, the processor 141 reads and executes the program stored in the memory 142, thereby realizing the processing result acquisition unit 121, the processing result evaluation unit 122, the parameter sorting unit 125, the first dimension reduction unit 126, the second dimension reduction unit 127, the machine learning unit 128, the third dimension reduction unit 131, the optimal processing condition search unit 132, the dimension restoration unit 133, and the processing condition instruction unit 134.

[0127] In other words, when each part of the processing condition search device 120 is implemented by the processing circuit 140, the processing result acquisition unit 121, the processing result evaluation unit 122, the parameter sorting unit 125, the first dimension reduction unit 126, the second dimension reduction unit 127, the machine learning unit 128, the third dimension reduction unit 131, the optimal processing condition search unit 132, the dimension restoration unit 133, and the processing condition instruction unit 134 are implemented using software, i.e., a program.

[0128] Such programs can be provided via a network, or they can be provided recorded on a recording medium. That is, such programs can also be provided as program products, for example.

[0129] Furthermore, the processing result evaluation storage unit 123, the sorting mark storage unit 124, the model storage unit 129, and the fixed parameter storage unit 130 are implemented by the memory 142.

[0130] The memory 142 is also used as the working area of ​​the processor 141.

[0131] The processor 141 is a CPU (Central Processing Unit), etc. The memory 142 is, for example, a non-volatile or volatile semiconductor memory such as RAM (Random Access Memory), ROM (Read Only Memory), flash memory, or a hard disk.

[0132] Furthermore, when the processing circuits that realize the processing result acquisition unit 121, processing result evaluation unit 122, parameter sorting unit 125, first dimension reduction unit 126, second dimension reduction unit 127, machine learning unit 128, third dimension reduction unit 131, optimal processing condition search unit 132, dimension restoration unit 133 and processing condition instruction unit 134 are dedicated hardware, such as FPGA (Field Programmable Gate Array) or ASIC (Application Specific Integrated Circuit).

[0133] Each part of the processing condition search device 120 can also be implemented by combining processing circuits with processors and dedicated hardware. Alternatively, the processing condition search device 120 can also be implemented by connecting multiple of the above-mentioned processing circuits with processors or dedicated hardware via a computer network such as a cloud.

[0134] In other words, each part of the processing condition search device 120 can be realized through the processing circuit network.

[0135] Next, the operation of the processing system 100 of Embodiment 1 will be described.

[0136] Figure 6 This is a flowchart illustrating the operation of the processing system 100 according to Embodiment 1.

[0137] First, the sorting flag storage unit 124 stores the sorting flags (S10). Here, for example, the user of the machining system 100 can set the category of each of the multiple parameters used in the machining machine 110 as either a variable parameter or a fixed parameter via an input unit (not shown), and then store the sorting flags according to that setting.

[0138] Next, the fixed parameter storage unit 130 stores the values ​​of the parameters that are designated as fixed parameters that cannot be changed by the sorting flag stored in step S10 as fixed parameter data (S11). Furthermore, for example, this fixed parameter data can also be set via an input unit (not shown) using a user's instruction.

[0139] Next, the parameter sorting unit 125 reads multiple parameters from the processing result evaluation storage unit 123 and sorts the multiple parameters into variable parameters or fixed parameters by referring to the sorting marks stored in the sorting mark storage unit 124 (S12). Then, the parameter sorting unit 125 generates variable parameter data representing the sorted variable parameters and fixed parameter data representing the sorted fixed parameters, and provides the variable parameter data to the first dimension reduction unit 126 and the fixed parameter data to the second dimension reduction unit 127.

[0140] The first dimension reduction unit 126 analyzes the variable parameter data generated by the parameter sorting unit 125 and determines whether the dimension Mv of the variable parameter data is greater than a predetermined threshold THv (S13). If the dimension Mv is greater than the threshold THv (S13: Yes), the process proceeds to step S14; if the dimension Mv is less than or equal to the threshold THv (S13: No), the process proceeds to step S15. Alternatively, if the dimension Mv is less than or equal to the threshold THv (S13: No), the first dimension reduction unit 126 does not perform dimension reduction and provides the variable parameter data itself as the first feature data to the machine learning unit 128.

[0141] In step S14, the first dimension reduction unit 126 converts the variable parameter data into first feature data represented by a dimension Lv smaller than the threshold THv.

[0142] In step S15, the second dimension reduction unit 127 analyzes the fixed parameter data sorted by the parameter sorting unit 125 and determines whether the dimension Mf of the fixed parameter data is greater than the threshold THf. If the dimension Mf is greater than the threshold THf (S15: Yes), the process proceeds to step S16; if the dimension Mf is less than or equal to the threshold THf (S15: No), the process proceeds to step S17. Alternatively, if the dimension Mf is less than or equal to the threshold THf (S15: No), the second dimension reduction unit 127 does not perform dimension reduction and provides the fixed parameter data itself as second feature data to the machine learning unit 128.

[0143] In step S16, the second dimension reduction unit 127 converts the fixed parameter data into second feature data represented by a dimension Lf that is less than the threshold THf.

[0144] In step S17, the third dimension reduction unit 131 determines whether the dimension Mf of the fixed parameter data stored in the fixed parameter storage unit 130 is greater than the threshold THf. If the dimension Mf is greater than the threshold THf (S17: Yes), the process proceeds to step S18; if the dimension Mf is less than or equal to the threshold THf (S17: No), the process proceeds to step S19. Furthermore, if the dimension Mf is less than or equal to the threshold THf (S17: No), the third dimension reduction unit 131 does not perform dimension reduction and provides the fixed parameter data itself as third feature data to the optimal processing condition search unit 132.

[0145] In step S18, the third dimension reduction unit 131 reads the fixed parameter data from the fixed parameter storage unit 130, performs the same dimension reduction process as the second dimension reduction unit 127 on it, thereby converting it into third feature data, and provides the third feature data to the optimal processing condition search unit 132.

[0146] In step S19, the machine learning unit 128 reads the first feature data provided by the first dimension reduction unit 126, the second feature data provided by the second dimension reduction unit 127, and multiple evaluation values ​​contained in the processing result evaluation information stored in the processing result evaluation storage unit 123. It treats each feature as an input value and its evaluation value as a response value, performs learning on the relationship between the two, and generates a learning model that uses a mathematical model to represent them. The generated learning model is stored in the model storage unit 129.

[0147] Next, the optimal processing condition search unit 132 uses the learning model stored in the model storage unit 129 to search for the optimal processing condition (S20). At this time, the optimal processing condition search unit 132 provides the third feature quantity data provided by the third dimension reduction unit 131 and the candidate feature quantities of multiple variable parameters generated using a predetermined method as input to the learning model, obtains the predicted value as the evaluation value obtained by the learning model in response to the input, and sets the candidate that gives the best predicted value as the optimal processing condition.

[0148] Figure 7 This is a schematic diagram used to illustrate the search method in Implementation Method 1.

[0149] exist Figure 7 In the example shown, the evaluation value is determined by the characteristic quantity av of the variable parameter and the characteristic quantity af of the fixed parameter, and a graph is shown with the characteristic quantity av of the variable parameter set as the horizontal axis and the characteristic quantity af of the fixed parameter set as the vertical axis.

[0150] exist Figure 7 In the quadrilateral, points P01 to P06 represent the searched processing conditions stored in the processing result evaluation storage unit 123.

[0151] Furthermore, regions R11, R12, and R13 represent regions where the learning model generated by the machine learning unit 128, based on the data of the searched processing conditions, predicts the processing result to be poor, good, and best, respectively. Here, as an example, let the evaluation value be the defect rate. Moreover, let the evaluation value be a continuous value from 0% to 100%, defining a defect rate less than 1% as best, a defect rate less than 5% as good, and a defect rate of 5% or more as poor. Therefore, the predicted evaluation value in region R11 is 5% or more, the predicted evaluation value in region R12 is 1% or more but less than 5%, and the predicted evaluation value in region R13 is less than 1%.

[0152] Additionally, it should be noted that these regions are not clearly defined. The values ​​of the variable parameters av corresponding to each coordinate and the values ​​of the fixed parameters af are input into the learning model to obtain their predicted values ​​before observation can be performed.

[0153] Currently, the value of af is shown by a certain third feature quantity as a feature quantity with fixed parameters. * At that time, the search space is Figure 7 The search space is shown on the dashed line L. The dimension of the search space is equal to the dimension of the feature quantity of the variable parameter (in other words, the first feature quantity). In this example, for the sake of illustration, the feature quantity of the variable parameter is set to one dimension; however, in the case of two or more dimensions, the search space also becomes two or more dimensions.

[0154] Candidate points are selected from points in a search space constrained by a third feature; however, the selection method can be arbitrary. For example, the search space can be divided into grids at specified intervals, as in grid search, with each grid point designated as a candidate. Alternatively, a specified number of points can be randomly selected from the search space, as in random search. Another approach is successive optimization, such as using methods like mountain climbing, annealing, particle swarm optimization, or Bayesian optimization to select candidate points one by one, calculate their predicted values, and determine the next candidate point based on the results.

[0155] On the dashed line L, point P21 (triangle), point P22 (circle), and point P23 (double circle) represent search candidates selected using a predetermined method, and the horizontal coordinates of each point become candidates for the characteristic quantities of the variable parameters.

[0156] Currently, the predicted defect rate in the first search candidate corresponding to point P21 (shown by the triangle) is 5% or more, for example, 12%. The predicted defect rate in the second search candidate corresponding to point P22 (shown by the circle) is 1% or more but less than 5%, for example, 3%. The predicted defect rate in the third search candidate corresponding to point P23 (shown by the double circle) is less than 1%, for example, 0.2%. At this time, the optimal processing condition search unit 132 determines the third search candidate shown by point P23 as the best, and selects the characteristic quantity av of the variable parameter in that candidate. * Dimensional restoration section 133 is provided with optimal processing conditions.

[0157] Furthermore, if the learning model generated by the machine learning unit 128 is a Gaussian process regression model, the optimal processing condition search unit 132 can use this model to calculate not only the predicted value of the evaluation value but also its confidence interval. Moreover, the optimal processing condition search unit 132 can calculate a score indicating whether a search should be performed for any unsearched point using an acquisition function calculated based on the calculated confidence interval. In this case, the optimal processing condition search unit 132 can also set the feature quantity of the variable parameter of the search point with the largest score calculated by the acquisition function as the optimal processing condition.

[0158] As described above, the optimal processing condition search unit 132 selects the best candidate as the optimal processing condition only with respect to the feature values ​​of the variable parameters while maintaining the feature values ​​of the fixed parameters. Therefore, it can perform the search only with respect to the variable parameters without changing the fixed parameters. Furthermore, the optimal processing condition search unit 132 selects the candidate that is predicted by the learning model to obtain the best result as the optimal processing condition, thus further reducing the actual trial runs of the machining machine 110 and enabling efficient searching of processing conditions.

[0159] return Figure 6 The dimension restoration unit 133 determines whether the dimension Mv of the variable parameter data is greater than the threshold THv (S21). If the dimension Mv of the variable parameter data is greater than the threshold THv (S21: Yes), the process proceeds to step S22; if the dimension Mv of the variable parameter data is less than the threshold THv (S21: No), the process proceeds to step S23.

[0160] In step S22, the dimension restoration unit 133 converts the optimal processing conditions provided by the optimal processing condition search unit 132 into variable parameters. Then, the dimension restoration unit 133 provides the processing condition instruction unit 134 with the processing conditions formed by combining the variable parameters and the fixed parameters read from the fixed parameter storage unit 130 as search processing conditions.

[0161] On the other hand, if the result in step S21 is "No", the dimension restoration unit 133 does not perform dimension restoration, but directly sets the optimal processing condition provided by the optimal processing condition search unit 132 as a variable parameter, and provides the processing condition formed by combining the variable parameter and the fixed parameter read from the fixed parameter storage unit 130 as the search processing condition to the processing condition instruction unit 134.

[0162] Next, the machining condition instruction unit 134 instructs the machining machine 110 to perform machining under the machining conditions provided by the dimension restoration unit 133 (S23). In addition, the machining condition instruction unit 134 adds the machining conditions to the machining result evaluation information stored in the machining result evaluation storage unit 123.

[0163] Additionally, at this time, the user can arbitrarily modify the processing conditions via an input unit (not shown). In this case, the user implements the modified processing conditions and provides them from the processing condition instruction unit 134 to the processing machine 110 and the processing result evaluation storage unit 123.

[0164] Next, the machining machine 110 performs machining according to the machining conditions provided by the machining condition instruction unit 134 (S24).

[0165] Then, the processing result acquisition unit 121 acquires processing result information from the processing machine 110 (S25).

[0166] The processing result evaluation unit 122 determines the evaluation value of the processing result based on the processing result information obtained by the processing result acquisition unit 121 (S26).

[0167] Then, the processing result evaluation unit 122 stores the evaluation value in the processing result evaluation information stored in the processing result evaluation storage unit 123, and matches it with the processing conditions added by the processing condition instruction unit 134.

[0168] Then, the parameter sorting unit 125 determines whether to end the process (S28). If the process does not end (S28: No), the process returns to step S12, and the above process is repeated. The determination of whether to end the process can be made using any method. For example, an upper limit for any number of repetitions can be determined, or the user can view the processing results and indicate the end. Furthermore, the processing condition search device 120 can automatically end the search according to a certain criterion.

[0169] As described above, according to Implementation 1, the processing conditions are divided into fixed parameters and variable parameters, and their dimensions are reduced separately to convert them into their respective feature quantities. While maintaining the feature quantities of the fixed parameters, the optimal processing conditions are searched for with respect to the feature quantities of the variable parameters. Therefore, even if the parameters constituting the processing conditions are high-dimensional and some of them cannot be changed, the optimal value can be searched efficiently only with respect to the parameters that can be changed.

[0170] Furthermore, according to Implementation 1, even if the variable parameter is high-dimensional, the optimal processing conditions are searched for the feature quantity of the variable parameter that is lower in dimension than the variable parameter obtained by reducing the dimension of the variable parameter. As a result, the search space becomes low-dimensional, and the search for the optimal processing conditions becomes easier.

[0171] Furthermore, according to Implementation 1, even if the parameters constituting the processing conditions are high-dimensional, machine learning and the search for optimal processing conditions are performed based on the lower-dimensional feature quantities obtained through dimension reduction, thereby reducing the computational power or memory capacity required for these processes.

[0172] Furthermore, according to Implementation 1, the candidate that is predicted to yield the best result by the learning model is selected as the optimal processing condition. Therefore, the actual trial of the processing machine 110 is further reduced, and efficient processing condition search can be performed.

[0173] Furthermore, according to Implementation 1, in the machine learning that learns the relationship between processing conditions and evaluation values, not only are the variable parameter features used as the search object for the optimal processing conditions used, but also the fixed parameter features are used to perform machine learning and generate a learning model. This model can take fixed parameters into account to predict the evaluation value of the processing result, thereby improving the prediction accuracy.

[0174] Implementation Method 2

[0175] In Implementation 1, variable parameters and fixed parameters are sorted according to sorting criteria. However, even among parameters sorted as variable parameters by the sorting criteria, there are parameters that have a high correlation with fixed parameters and can be assigned as fixed parameters. In Implementation 2, such parameters are automatically identified and assigned as fixed parameters.

[0176] like Figure 1 As shown, the machining system 200 of Embodiment 2 includes a machining machine 110 and a machining condition search device 220.

[0177] The machining machine 110 in the machining system 200 of Embodiment 2 is the same as the machining machine 110 in the machining system 100 of Embodiment 1.

[0178] like Figure 2As shown, the processing condition search device 220 in Embodiment 2 includes a processing result acquisition unit 121, a processing result evaluation unit 122, a processing result evaluation storage unit 123, a sorting flag storage unit 124, a parameter sorting unit 225, a first dimension reduction unit 126, a second dimension reduction unit 127, a machine learning unit 128, a model storage unit 129, a fixed parameter storage unit 130, a third dimension reduction unit 131, an optimal processing condition search unit 132, a dimension restoration unit 133, and a processing condition instruction unit 134.

[0179] The processing result acquisition unit 121, processing result evaluation unit 122, processing result evaluation storage unit 123, sorting flag storage unit 124, first dimension reduction unit 126, second dimension reduction unit 127, machine learning unit 128, model storage unit 129, fixed parameter storage unit 130, third dimension reduction unit 131, optimal processing condition search unit 132, dimension restoration unit 133, and processing condition instruction unit 134 of the processing condition search device 220 in Embodiment 2 are the same as those in Embodiment 1.

[0180] The parameter sorting unit 225 sorts the multiple parameters contained in the processing result evaluation information stored in the processing result evaluation storage unit 123 into variable parameters or fixed parameters, and generates variable parameter data representing the sorted variable parameters and fixed parameter data representing the sorted fixed parameters.

[0181] Figure 8 This is a block diagram that schematically shows the structure of the parameter sorting unit 225 in Embodiment 2.

[0182] The parameter sorting unit 225 has an initial allocation unit 250, a parameter data storage unit 251, a parameter allocation unit 254, and an output unit 258.

[0183] The initial allocation unit 250 allocates multiple parameters into multiple variable parameters and one or more fixed parameters by referring to the sorting flags. Here, the parameters allocated as variable parameters by the initial allocation unit 250 are also referred to as initial variable parameters, and the parameters allocated as fixed parameters by the initial allocation unit 250 are also referred to as initial fixed parameters.

[0184] For example, the initial allocation unit 250 allocates, as initial states, the multiple parameters included in the processing result evaluation information stored in the processing result evaluation storage unit 123 as fixed parameters or variable parameters respectively according to the sorting flag read out from the sorting flag storage unit 124, thereby generating variable parameter data representing the allocated variable parameters and fixed parameter data representing the allocated fixed parameters.

[0185] Then, the initial allocation unit 250 causes the parameter data storage unit 251 to store the variable parameter data and the fixed parameter data.

[0186] The parameter data storage unit 251 includes a variable parameter data storage unit 252 that stores the variable parameter data generated by the initial allocation unit 250, and a fixed parameter data storage unit 253 that stores the fixed parameter data generated by the initial allocation unit 250.

[0187] The parameter allocation unit 254 finally allocates the variable parameters and fixed parameters that are allocated as initial states by the initial allocation unit 250.

[0188] The parameter allocation unit 254 includes a correlation analysis unit 255 and a reallocation unit 256.

[0189] The correlation analysis unit 255 determines multiple combinations of each of the multiple initial variable parameters and each of the one or more initial fixed parameters, and analyzes the correlation of each of the multiple combinations.

[0190] For example, the correlation analysis unit 255 combines the variable parameter data stored in the variable parameter data storage unit 252 and the fixed parameter data stored in the fixed parameter data storage unit 253 according to each category of the parameters, and analyzes the correlation between the parameters.

[0191] Specifically, as shown in (A) of Figure 4 , in the variable parameter data storage unit 252, Mv types of variable parameter data 102 for the past N times of processing are stored. As shown in (B) of Figure 4 , Mf types of fixed parameter data 103 are stored in the fixed parameter data storage unit 253. In this case, for all combinations of x and y, the correlation score represented by the following formula (4) is calculated

[0192]

[0193] where 1 < x < Mv, 1 < y < Mf, Qx is the past N variable parameter values q of the parameter number x in the variable parameter data 102 1xLet q2x, ..., qNx be a vector of elements, and Ry be the fixed parameter value r of parameter number y for the past N times in fixed parameter data 103. 1y r 2y , ..., r Ny Let it be a vector of elements.

[0194] Furthermore, the function Φ is a function that outputs a numerical value representing the correlation between vectors. (Correlation score) Specific examples include the absolute value of the correlation coefficient, cross-entropy, KL (Kullback-Leibler) information, or other mutual information.

[0195] Figure 9 (A) and (B) show examples of cases where the correlation between Qx and Rx is low and cases where the correlation between Qx and Rx is high.

[0196] Figure 9 (A) shows the case where the correlation between Qx and Rx is low. Figure 9 (B) shows the case where Qx and Rx have a high correlation.

[0197] In addition, Figure 9 In the graphs shown in (A) and (B), the vertical axis represents the variable parameter x, and the horizontal axis represents the fixed parameter y.

[0198] like Figure 9 As shown in (A), under conditions of low correlation, the variable and fixed parameters are distributed approximately uncorrelatedly, whereas, as Figure 9 As shown in (B), a certain relationship is established between the variable parameter and the fixed parameter when the correlation is high. In other words, in the latter case, the variable parameter and the fixed parameter can be considered to be linked. Furthermore, in the latter case, the variable parameter can also be considered as a parameter whose value is automatically determined when the fixed parameter is determined. Therefore, this variable parameter can be included in the fixed parameter. This is also true in the case of inverse correlation.

[0199] Therefore, the redistribution unit 256 redistributes the variable parameter data based on the correlation score calculated by the correlation analysis unit 255. Specifically, the redistribution unit 256 redistributes the data according to a predetermined threshold. Will satisfy The variable parameters are assigned as fixed parameters, and these variable parameters are stored in the fixed parameter data.

[0200] In other words, the redistribution unit 256 redistributes the initial variable parameters contained in the combination of each initial variable parameter in the plurality of initial variable parameters and each initial fixed parameter in one or more initial fixed parameters that has a correlation higher than a predetermined threshold as initial fixed parameters, thereby determining the plurality of initial variable parameters after redistribution as a plurality of variable parameters, and determining the one or more initial fixed parameters after redistribution as one or more fixed parameters.

[0201] After the redistribution is completed by the redistribution unit 256, the output unit 258 provides the variable parameter data stored in the parameter data storage unit 251 to the first dimension reduction unit 126 and provides the fixed parameter data to the second dimension reduction unit 127.

[0202] Figure 10 This is a flowchart illustrating an example of the parameter sorting operation in the parameter sorting unit 225 of Embodiment 2.

[0203] First, as an initial allocation, the initial allocation unit 250 allocates multiple parameters contained in the processing result evaluation information stored in the sorting mark storage unit 124 as the initial state, thereby generating variable parameter data and fixed parameter data (S30). The generated variable parameter data is stored in the variable parameter data storage unit 252, and the generated fixed parameter data is stored in the fixed parameter data storage unit 253.

[0204] Next, the correlation analysis unit 255 analyzes the correlation between the variable parameter data stored in the variable parameter data storage unit 252 and the fixed parameter data stored in the fixed parameter data storage unit 253, according to each category of the parameter (S31). Here, a correlation score is calculated.

[0205] Next, the redistribution unit 256 initializes the parameter number x used to identify the variable parameter to "1" (S32).

[0206] Then, the redistribution unit 256 repeatedly performs the following process until the parameter number x exceeds the maximum value Mv (S33).

[0207] The redistribution unit 256 initializes the parameter number y used to identify the fixed parameter to "1" (S34).

[0208] Then, the redistribution unit 256 repeatedly performs the following process until the parameter number y exceeds the maximum value Mf (S35).

[0209] The redistribution unit 256 determines the correlation score calculated by the correlation analysis unit 255. Has the predetermined threshold been exceeded? (S36). Regarding the correlation score... Exceeding the threshold In the case of (S36: Yes), the process proceeds to step S37, based on the correlation score. Threshold If (S36: No), proceed to step S38.

[0210] In step S37, the redistribution unit 256 redistributes the relevant scores. Exceeding the threshold The variable parameter Qx with parameter number x is used as a fixed parameter. Specifically, the redistribution unit 256 extracts the variable parameter Qx from the variable parameter data storage unit 252 and appends it to the fixed parameter data stored in the fixed parameter data storage unit 253.

[0211] In step S38, the redistribution unit 256 adds "1" to the parameter number y.

[0212] Then, the redistribution unit 256 determines whether the parameter number y is below the maximum value Mf (S39). If the parameter number y is below the maximum value Mf (S39: Yes), the process returns to step S35; if the parameter number y exceeds the maximum value Mf (S39: No), the process proceeds to step S40.

[0213] In step S40, the redistribution unit 256 adds "1" to the parameter number x.

[0214] Then, the redistribution unit 256 determines whether the parameter number x is below the maximum value Mv (S41). If the parameter number x is below the maximum value Mv (S41: Yes), the process returns to step S33; if the parameter number x exceeds the maximum value Mv (S41: No), the process proceeds to step S42.

[0215] In step S42, the output unit 258 outputs the variable parameter data and fixed parameter data stored in the parameter data storage unit 251.

[0216] As described above, according to Embodiment 2, even parameters temporarily identified as variable parameters based on sorting criteria are analyzed for their correlation with fixed parameters. If the correlation is high, they are reassigned as fixed parameters. This further reduces the dimensionality of variable parameters and facilitates the search for optimal processing conditions.

[0217] Implementation Method 3

[0218] In Implementation 2, even if a parameter is a variable parameter in the sorting criteria, it is assigned as a fixed parameter if it has a high correlation with the fixed parameter. In Implementation 3, an analysis is performed to determine whether the parameters that are variable parameters in the sorting criteria contribute to the processing results, and variable parameters that do not contribute are assigned as fixed parameters.

[0219] like Figure 1 As shown, the machining system 300 of Embodiment 3 includes a machining machine 110 and a machining condition search device 320.

[0220] The machining machine 110 in the machining system 300 of Embodiment 3 is the same as the machining machine 110 in the machining system 100 of Embodiment 1.

[0221] like Figure 2 As shown, the processing condition search device 320 in Embodiment 3 includes a processing result acquisition unit 121, a processing result evaluation unit 122, a processing result evaluation storage unit 123, a sorting flag storage unit 124, a parameter sorting unit 325, a first dimension reduction unit 126, a second dimension reduction unit 127, a machine learning unit 128, a model storage unit 129, a fixed parameter storage unit 130, a third dimension reduction unit 131, an optimal processing condition search unit 132, a dimension restoration unit 133, and a processing condition instruction unit 134.

[0222] The processing result acquisition unit 121, processing result evaluation unit 122, processing result evaluation storage unit 123, sorting flag storage unit 124, first dimension reduction unit 126, second dimension reduction unit 127, machine learning unit 128, model storage unit 129, fixed parameter storage unit 130, third dimension reduction unit 131, optimal processing condition search unit 132, dimension restoration unit 133, and processing condition instruction unit 134 of the processing condition search device 320 in Embodiment 3 are the same as those in Embodiment 1.

[0223] The parameter sorting unit 325 sorts the multiple parameters contained in the processing result evaluation information stored in the processing result evaluation storage unit 123 into variable parameters or fixed parameters, and generates variable parameter data representing the sorted variable parameters and fixed parameter data representing the sorted fixed parameters.

[0224] Figure 11It is a block diagram schematically showing the structure of the parameter sorting unit 325 in Embodiment 3.

[0225] The parameter sorting unit 325 has an initial allocation unit 250, a parameter data storage unit 251, a parameter allocation unit 354, and an output unit 258.

[0226] The initial allocation unit 250, the parameter data storage unit 251, and the output unit 258 of the parameter sorting unit 325 in Embodiment 3 are the same as those of the parameter sorting unit 225 in Embodiment 2.

[0227] The parameter allocation unit 354 finally allocates the variable parameters and fixed parameters allocated as the initial state by the initial allocation unit 250.

[0228] The parameter allocation unit 354 has a reallocation unit 356 and a contribution degree analysis unit 357.

[0229] The contribution degree analysis unit 357 analyzes the contribution degree of each of the multiple initial variable parameters to the corresponding evaluation value.

[0230] For example, the contribution degree analysis unit 357 reads the variable parameter data from the variable parameter data storage unit 252 and reads the evaluation values corresponding to the multiple variable parameters included in the variable parameter data from the processing result evaluation storage unit 123. Then, the contribution degree analysis unit 357 analyzes the contribution degree of each of the multiple variable parameters to the evaluation value.

[0231] For example, as Figure 4 shown in (A) of, in the variable parameter data storage unit 252, for the past N times of processing, Mv types of variable parameter data 102 are stored. In addition, the evaluation values of the past N times of processing are stored in the processing result evaluation storage unit 123.

[0232] In this case, for all x satisfying 1 < x < Mv, the contribution degree score ψ expressed by the following formula (5) is calculated x .

[0233] ψ x = Ψ(Qx, J) (5)

[0234] Here, Qx is a vector including the past N variable parameter values q 1x , q 2x , …, q Nx of parameter number x included in the variable parameter data 102.

[0235] In addition, J is a vector including the past N evaluation values j1, j2, …, j N .

[0236] Furthermore, the function Ψ(Qx, J) is a function that calculates the contribution score of Qx to J using numerical representation.

[0237] As a contribution score ψ x For specific examples, give the absolute values ​​of the correlation coefficients between Qx and J.

[0238] Or, contribution score ψ x It can also be the reciprocal of the regression error when performing a single regression analysis of J using Qx.

[0239] Furthermore, the contribution score ψ x It can also be the magnitude of the regression error when performing multiple regression analysis on J using all variable parameters except Qx (in other words, all Qi that satisfy i≠x).

[0240] In addition to linear regression, nonlinear regression or core regression can also be cited as examples of these regression analyses.

[0241] The contribution score calculated in this way indicates whether the variable parameter x contributes to the evaluation value. If the contribution score is low, it is assumed to have little impact on the processing conditions and can be excluded from the search range of optimal processing conditions.

[0242] Regarding the variable parameters stored in the variable parameter data storage unit 252, the contribution score ψ x For the predetermined threshold TH Ψ In the following cases, the redistribution unit 356 assigns the parameter as a fixed parameter.

[0243] In other words, the redistribution unit 356 redistributes the initial variable parameters whose contribution is below a predetermined threshold from among the multiple initial variable parameters as initial fixed parameters. Thus, the multiple initial variable parameters after redistribution are determined as multiple variable parameters, and one or more initial fixed parameters after redistribution are determined as one or more fixed parameters.

[0244] Figure 12 This is a flowchart illustrating an example of the parameter allocation operation in the parameter allocation unit 354 of Embodiment 3.

[0245] First, the contribution analysis unit 357 reads the variable parameter data from the variable parameter data storage unit 252 and reads the evaluation values ​​corresponding to the multiple variable parameters contained in the variable parameter data from the processing result evaluation storage unit 123. Then, the contribution analysis unit 357 analyzes the contribution of the multiple variable parameters to the evaluation values ​​of the parameter categories (S50). Specifically, the contribution analysis unit 357 calculates the contribution score ψ for all parameter numbers x of the variable parameters using the above equation (5). x .

[0246] Next, the redistribution unit 356 initializes the parameter number x used to identify the variable parameter to "1" (S51).

[0247] Then, the redistribution unit 356 repeatedly performs the following process until the parameter number x exceeds its maximum value Mv (S52).

[0248] The redistribution unit 356 determines the contribution score ψ of the variable parameter corresponding to parameter number x. x Is it more than the threshold TH? Ψ Large (S53). In the contribution score ψ x Threshold TH Ψ In the following case (S53: No), proceed to step S54, where the contribution score ψ is... x Compared to threshold TH Ψ In the case of a large number of cases (S53: Yes), proceed to step S55.

[0249] In step S54, the redistribution unit 356 redistributes the contribution score ψ. x Compared to threshold TH Ψ The variable parameter Qx with parameter number x is used as a fixed parameter. Specifically, the redistribution unit 356 extracts the variable parameter Qx from the variable parameter data storage unit 252 and appends it to the fixed parameter data stored in the fixed parameter data storage unit 253. Then, the process proceeds to step S55.

[0250] In step S55, the redistribution unit 356 adds "1" to the parameter number x.

[0251] Then, the redistribution unit 356 determines whether the parameter number x is below the maximum value Mv (S56). If the parameter number x is below the maximum value Mv (S56: Yes), the process returns to step S52; if the parameter number x exceeds the maximum value Mv (S56: No), the process ends.

[0252] As described above, according to Implementation 3, even parameters temporarily identified as variable parameters according to sorting marks are analyzed for their contribution to the evaluation value. If the contribution is low, they are reassigned as fixed parameters. Therefore, the dimensionality of variable parameters can be further reduced, making it easier to search for optimal processing conditions.

[0253] Implementation Method 4

[0254] In Implementation 1, the optimal processing condition search unit 132 provides candidate features of multiple variable parameters generated using a prescribed method as input to the learning model, obtains predicted values ​​as evaluation values ​​obtained by the learning model in response to the input, and outputs the candidate with the best predicted value as the optimal processing condition.

[0255] However, in the initial search, the processing result evaluation storage unit 123 only stores data related to past processing. In this case, compared with the processing conditions predicted by the learning model, sometimes good results are obtained when selecting the processing conditions with similar performance and good evaluation values ​​from past data as the optimal processing conditions.

[0256] Therefore, Implementation 4 shows an example where, in the initial search, the optimal processing conditions are determined based on the first feature data, the second feature data, the third feature data, and the evaluation value.

[0257] like Figure 1 As shown, the machining system 400 of Embodiment 4 includes a machining machine 110 and a machining condition search device 420.

[0258] The machining machine 110 in the machining system 400 of Embodiment 4 is the same as the machining machine 110 in the machining system 100 of Embodiment 1.

[0259] like Figure 2 As shown, the processing condition search device 420 in Embodiment 4 includes a processing result acquisition unit 121, a processing result evaluation unit 122, a processing result evaluation storage unit 123, a sorting flag storage unit 124, a parameter sorting unit 125, a first dimension reduction unit 126, a second dimension reduction unit 127, a machine learning unit 128, a model storage unit 129, a fixed parameter storage unit 130, a third dimension reduction unit 131, an optimal processing condition search unit 432, a dimension restoration unit 133, and a processing condition instruction unit 134.

[0260] The processing result acquisition unit 121, processing result evaluation unit 122, processing result evaluation storage unit 123, sorting flag storage unit 124, parameter sorting unit 125, first dimension reduction unit 126, second dimension reduction unit 127, machine learning unit 128, model storage unit 129, fixed parameter storage unit 130, third dimension reduction unit 131, dimension restoration unit 133, and processing condition instruction unit 134 of the processing condition search device 420 in Embodiment 4 are the same as those in Embodiment 1.

[0261] In the initial search, the optimal processing condition search unit 432 selects all processing numbers that meet predetermined criteria from among the multiple evaluation values ​​included in the processing result evaluation information stored in the processing result evaluation storage unit 123. Then, the optimal processing condition search unit 432 determines the second feature quantity corresponding to the selected processing number from the second feature quantity data, and identifies the processing number whose determined second feature quantity is closest to the third feature quantity shown in the third feature quantity data. For example, it can determine the processing number with the smallest distance between the feature quantities calculated using the distance scale of the second and third feature quantities.

[0262] Then, the optimal processing condition search unit 432 determines the feature quantity corresponding to the determined processing number from the first feature quantity data as the optimal processing condition, and provides the determined optimal processing condition to the dimension restoration unit 133.

[0263] In addition, the optimal processing condition search unit 432 determines the optimal processing conditions in the same way as in Embodiment 1 in searches other than the initial one.

[0264] Figure 13 This is a flowchart illustrating the operation of the optimal processing conditions search unit 432 during the initial search.

[0265] Figure 13 The flowchart shown is only performed during the initial search.

[0266] First, the optimal processing condition search unit 432 selects all processing numbers that meet predetermined criteria from among the multiple evaluation values ​​included in the processing result evaluation information stored in the processing result evaluation storage unit 123. For example, using a predetermined threshold, all processing numbers judged to have a processing result better than the evaluation value corresponding to the threshold can be selected.

[0267] Next, the optimal processing condition search unit 432 determines the processing number n that is closest to the third feature value among the second feature values ​​corresponding to the sorted processing numbers. * (S61).

[0268] Next, the optimal processing condition search unit 432 determines the processing number n from the first feature data. * The corresponding feature quantities are used as optimal processing conditions (S62). Then, the determined optimal processing conditions are provided to the dimension restoration unit 133.

[0269] The third feature is obtained by converting the values ​​of parameters that cannot be changed during the search into feature values. In past data, data with good evaluation values ​​and close to them are searched, and the variable parameters paired with them are selected as the initial optimal processing conditions. Thus, it is expected that good processing conditions can be found with fewer search attempts.

[0270] As described above, in Embodiment 4, the optimal processing condition search unit 432 functions as a search unit that, when searching for the optimal value for the first time, determines one or more evaluation values ​​from a plurality of evaluation values ​​that are higher than a predetermined evaluation value, determines one or more second feature values ​​from a plurality of second feature values ​​that correspond to the determined evaluation value, determines one of the determined second feature values ​​that is closest to a third feature value, determines a first feature value that corresponds to the second feature value, and sets the first feature value as the optimal value.

[0271] As described above, according to Embodiment 4, in the initial search, the conditions that are close to the actual values ​​of fixed parameters that result in good processing and cannot be changed in the current search are searched from the past processing conditions stored in the processing result evaluation storage unit 123. The variable parameters paired with these conditions are set as the initial optimal processing conditions. As a result, good processing conditions can be found with fewer search cycles.

[0272] Furthermore, the parameter sorting unit 125 in Embodiment 4 is the same as the parameter sorting unit 125 in Embodiment 1; however, Embodiment 4 is not limited to this example. For instance, the parameter sorting unit 125 in Embodiment 4 could also be the parameter sorting unit 225 in Embodiment 2 or the parameter sorting unit 325 in Embodiment 3.

[0273] Implementation Method 5

[0274] In Implementation 1, dimensionality reduction is performed separately for the sorted variable parameter data and fixed parameter data. In Implementation 5, to improve the reduction effect, the result of uniform dimensionality reduction without sorting parameter data is used as a reference, and each reduction process is adjusted in a manner similar to that result.

[0275] like Figure 1 As shown, the machining system 500 of Embodiment 5 includes a machining machine 110 and a machining condition search device 520.

[0276] The machining machine 110 in the machining system 500 of Embodiment 5 is the same as the machining machine 110 in the machining system 100 of Embodiment 1.

[0277] Figure 14 This is a block diagram that roughly shows the structure of the processing condition search device 520.

[0278] The processing condition search device 520 includes a processing result acquisition unit 121, a processing result evaluation unit 122, a processing result evaluation storage unit 123, a sorting flag storage unit 124, a parameter sorting unit 125, a first dimension reduction unit 526, a second dimension reduction unit 527, a machine learning unit 128, a model storage unit 129, a fixed parameter storage unit 130, a third dimension reduction unit 131, an optimal processing condition search unit 132, a dimension restoration unit 133, a processing condition instruction unit 134, a fourth dimension reduction unit 560, a first comparison unit 561, and a second comparison unit 562.

[0279] The processing result acquisition unit 121, processing result evaluation unit 122, processing result evaluation storage unit 123, sorting flag storage unit 124, parameter sorting unit 125, machine learning unit 128, model storage unit 129, fixed parameter storage unit 130, third dimension reduction unit 131, optimal processing condition search unit 132, dimension restoration unit 133, and processing condition instruction unit 134 of the processing condition search device 520 in Embodiment 5 are the same as those in Embodiment 1.

[0280] The fourth dimension reduction part 560 is a dimension reduction part that reduces the dimensions of multiple parameters, thereby generating a fourth feature quantity.

[0281] For example, the fourth dimension reduction unit 560 reads the processing conditions from the processing result evaluation information stored in the processing result evaluation storage unit 123, performs dimension reduction on multiple parameters included in the read processing conditions, thereby generating fourth feature data. The generated fourth feature data is provided to the first comparison unit 561 and the second comparison unit 562.

[0282] Similar to Embodiment 1, the first dimension reduction unit 526 generates first feature data and provides the generated first feature data to the first comparison unit 561. However, the first dimension reduction unit 526 reduces the dimensions of the multiple variable parameters regardless of the dimension of the multiple variable parameters, thereby generating the first feature.

[0283] Then, the first dimension reduction unit 526 obtains the similarity score calculated by comparing the first feature data and the fourth feature data from the first comparison unit 561, namely the first similarity score, and determines whether the first similarity score has converged.

[0284] If the first similarity score fails to converge, the first dimension reduction unit 526 changes the dimension reduction process in a way that increases the similarity between the first feature data and the fourth feature data shown by the first similarity score, thereby adjusting the first similarity score. Then, the first dimension reduction unit 526 regenerates the first feature data through the adjusted dimension reduction process and provides the generated first feature data to the first comparison unit 561.

[0285] Repeat the above process until the first similarity score converges.

[0286] Then, if the first similarity score converges, the first dimension reduction unit 526 provides the first feature data that has been determined to be converged to the machine learning unit 128. In addition, the first feature data before the first similarity score converges is also called the first temporary feature data.

[0287] As described above, the first dimension reduction unit 526 changes the processing of reducing multiple variable parameters of dimensions and repeatedly generates the first temporary feature quantity until the first similarity score converges, and determines the first temporary feature quantity when the first similarity score converges as the first feature quantity.

[0288] Similar to Embodiment 1, the second dimension reduction unit 527 generates second feature data and provides the generated second feature data to the second comparison unit 562. However, the second dimension reduction unit 527 reduces the dimensions of the multiple fixed parameters regardless of the dimension of the multiple fixed parameters, thereby generating the second feature.

[0289] Then, the second dimension reduction unit 527 obtains the similarity score, i.e. the second similarity score, calculated by comparing the second feature data and the fourth feature data from the second comparison unit 562, and determines whether the second similarity score has converged.

[0290] If the second similarity score fails to converge, the second dimension reduction unit 527 changes the dimension reduction process in a way that increases the similarity between the second feature data and the fourth feature data shown by the second similarity score, thereby adjusting the dimension reduction process. Then, the second dimension reduction unit 527 regenerates the second feature data through the adjusted dimension reduction process and provides the generated second feature data to the second comparison unit 561.

[0291] Repeat the above process until the second similarity score converges.

[0292] Then, if the second similarity score converges, the second dimension reduction unit 527 provides the converged second feature data to the machine learning unit 128. Furthermore, the second feature data before the second similarity score converges is also referred to as the second temporary feature data.

[0293] As described above, the second dimension reduction unit 527 modifies the processing of reducing multiple dimensions with fixed parameters and repeatedly generates a second temporary feature quantity until the second similarity score converges. The second temporary feature quantity at the time of convergence of the second similarity score is determined as the second feature quantity.

[0294] The first comparison unit 561 calculates a similarity score, i.e., the first similarity score, which represents the degree of similarity between the first temporary feature and the fourth feature.

[0295] The second comparison unit 562 calculates a similarity score, i.e., the second similarity score, which represents the degree of similarity between the second temporary feature and the fourth feature.

[0296] As a similarity score, specifically, the absolute value of the correlation coefficient, cross-entropy, KL information, and other mutual information are used. Furthermore, the more similar the compared data, the lower the value of the cross-entropy or mutual information. Therefore, when used for similarity scores, the signs of positive and negative values ​​are reversed, or a reciprocal form is used.

[0297] The autoencoder is a preferred example of the dimension reduction processing of the first dimension reduction unit 526, the second dimension reduction unit 527 and the fourth dimension reduction unit 560 in Embodiment 5. When the autoencoder is learning, the first dimension reduction unit 526 and the second dimension reduction unit 527 add the aforementioned cross-entropy or KL information to the loss function, thereby obtaining a dimension reduction effect close to that of the fourth dimension reduction unit 560.

[0298] In the fourth feature data and the first or second feature data, when the dimensions are different, for example, considering the combination of all features, the similarity score is calculated, and the maximum or average value is used as the first similarity score or the second similarity score.

[0299] For example, let the dimension of the first feature data be Mv, and the dimension of the fourth feature data be Mo. Then, the similarity score α between the x-th dimension feature of the first feature data and the z-th dimension feature of the fourth feature data is... 1(x，z) It can be expressed using the following equation (6).

[0300] α 1(x，z) =Γ(Av x Ao z (6)

[0301] Here, Γ is the function for calculating the similarity score.

[0302] In addition, Av x The feature value of the x-th dimension of the first feature value data of the processing number n is set as av. nx When will av1x av 2x , ...,av Nx Let it be a vector of elements.

[0303] Furthermore, Ao z The feature quantity of the fourth feature quantity data of the processing number n in the z-th dimension is set as ao. nz When will ao 1z ao 2z , ..., ao Nx Let it be a vector of elements.

[0304] Here, N is the maximum value of the processing number.

[0305] For all combinations of x = 1, 2, ..., Mv and z = 1, 2, ..., Mo, calculate this first similarity score α. 1(x，z) Set its maximum, minimum, or average value as the first similarity score α1.

[0306] The second similarity score α2 is also based on setting the feature quantity of the y-th dimension of the second feature quantity data in the processing number n as af. ny When will af 1y ,af 2y , ..., af Ny Let Af be a vector of elements. y And the z-th dimension feature of the aforementioned fourth feature data is set as the element vector Ao. z For all combinations of y = 1, 2, ..., Mf and z = 1, 2, ..., Mo, calculate the similarity score α expressed by the following equation (7). 2(y，z) Set its maximum, minimum, or average value as the second similarity score α2.

[0307] α 2(y，z) =Γ(Af y Ao z (7)

[0308] Furthermore, similarity scores are considered to have converged if the change in similarity score since the last score is below a predetermined threshold.

[0309] The hardware structure of the processing condition search device 520 in the fifth embodiment described above is the same as that of the processing condition search device 120 in embodiment 1. For example, the fourth dimension reduction unit 560, the first comparison unit 561, and the second comparison unit 562 can also be implemented by the processing circuit 140.

[0310] Figure 15This is a flowchart illustrating the operations of the first dimension reduction unit 526, the second dimension reduction unit 527, the fourth dimension reduction unit 560, the first comparison unit 561, and the second comparison unit 562 in Embodiment 5.

[0311] First, the fourth dimension reduction unit 560 reads the processing conditions (in other words, parameters) contained in the processing result evaluation information stored in the processing result evaluation storage unit 123, performs dimension reduction on them, thereby generating fourth feature data (S70). The generated fourth feature data is provided to the first comparison unit 561 and the second comparison unit 562.

[0312] Similar to Embodiment 1, the first dimension reduction unit 526 reduces the dimensions of the variable parameter data from the parameter sorting unit 125 to generate first feature data (S71). The first feature data is provided to the first comparison unit 561.

[0313] The first comparison unit 561 compares the fourth feature data and the first feature data, and calculates a first similarity score α1 (S73). The first similarity score α1 is provided to the first dimension reduction unit 526.

[0314] The first dimension reduction unit 526 determines whether the first similarity score α1 has converged (S73). If the first similarity score α1 has not converged (S73: No), the process proceeds to step S74; if the first similarity score α1 has converged (S73: Yes), the process proceeds to step S75. Here, the first dimension reduction unit 526 provides the first feature data generated in step S71 to the machine learning unit 128.

[0315] In step S74, the first dimension reduction unit 526 changes the dimension reduction process in a way that increases the first similarity score α1. Then, the process returns to step S71.

[0316] In step S75, similar to Embodiment 1, the second dimension reduction unit 527 reduces the dimensions of the fixed parameter data from the parameter sorting unit 125 to generate second feature data (S72). The second feature data is then provided to the second comparison unit 562.

[0317] The second comparison unit 562 compares the fourth feature data and the second feature data, and calculates the second similarity score α2 (S76). The second similarity score α2 is provided to the second dimension reduction unit 527.

[0318] The second dimension reduction unit 527 determines whether the second similarity score α2 has converged (S77). If the second similarity score α2 has not converged (S77: No), the process proceeds to step S78. If the second similarity score α2 has converged (S77: Yes), the second dimension reduction unit 527 provides the second feature data generated in step S75 to the machine learning unit 128 and ends the process.

[0319] In step S78, the second dimension reduction unit 527 changes the dimension reduction process in a way that increases the second similarity score α2. Then, the process returns to step S75.

[0320] As described above, according to Embodiment 5, when performing dimension reduction on the sorted variable parameter data and fixed parameter data respectively, the result of uniformly performing dimension reduction without sorting parameter data is used as a reference, and each reduction process is adjusted in a manner similar to that of the reference. Therefore, the dimension reduction effect can be improved.

[0321] Furthermore, the parameter sorting unit 125 in Embodiment 5 is the same as the parameter sorting unit 125 in Embodiment 1; however, Embodiment 5 is not limited to this example. For instance, the parameter sorting unit 125 in Embodiment 5 could also be the parameter sorting unit 225 in Embodiment 2 or the parameter sorting unit 325 in Embodiment 3.

[0322] Furthermore, the optimal processing condition search unit 132 in embodiment 5 can also be the optimal processing condition search unit 432 in embodiment 4.

[0323] Implementation Method 6

[0324] In embodiment 5, the first feature and the second feature are compared separately with the fourth feature. In embodiment 6, the first feature and the second feature are synthesized in such a way that they have the same dimension as the fourth feature, and then compared with the fourth feature.

[0325] like Figure 1 As shown, the machining system 600 of Embodiment 6 includes a machining machine 110 and a machining condition search device 620.

[0326] The machining machine 110 in the machining system 600 of Embodiment 6 is the same as the machining machine 110 in the machining system 100 of Embodiment 1.

[0327] Figure 16 This is a block diagram that roughly shows the structure of the processing condition search device 620.

[0328] The processing condition search device 620 includes a processing result acquisition unit 121, a processing result evaluation unit 122, a processing result evaluation storage unit 123, a sorting flag storage unit 124, a parameter sorting unit 125, a first dimension reduction unit 626, a second dimension reduction unit 627, a machine learning unit 128, a model storage unit 129, a fixed parameter storage unit 130, a third dimension reduction unit 131, an optimal processing condition search unit 132, a dimension restoration unit 133, a processing condition instruction unit 134, a fourth dimension reduction unit 560, a synthesis unit 663, and a comparison unit 664.

[0329] The processing result acquisition unit 121, processing result evaluation unit 122, processing result evaluation storage unit 123, sorting flag storage unit 124, parameter sorting unit 125, machine learning unit 128, model storage unit 129, fixed parameter storage unit 130, third dimension reduction unit 131, optimal processing condition search unit 132, dimension restoration unit 133, and processing condition instruction unit 134 of the processing condition search device 620 in Embodiment 1 are the same as those in Embodiment 1.

[0330] The fourth dimension reduction unit 560 of the processing condition search device 620 in Embodiment 6 is the same as the fourth dimension reduction unit 560 of the processing condition search device 520 in Embodiment 5. However, in Embodiment 6, the fourth dimension reduction unit 560 provides the generated fourth feature data to the comparison unit 664.

[0331] Similar to Embodiment 1, the first dimension reduction unit 626 generates first feature data and provides the generated first feature data to the synthesis unit 663. However, the first dimension reduction unit 626 reduces the dimensions of the multiple variable parameters regardless of the dimension of the multiple variable parameters, thereby generating the first feature.

[0332] Then, the first dimension reduction unit 626 obtains from the comparison unit 664 the similarity score calculated by comparing the synthetic feature data representing the synthetic feature data of the first feature data and the second feature data with the fourth feature data, and determines whether the similarity score has converged.

[0333] If the similarity score does not converge, the first dimension reduction unit 626 changes the dimension reduction process in a way that increases the similarity between the synthesized feature data and the fourth feature data shown by the similarity score, thereby adjusting the dimension reduction process. Then, the first dimension reduction unit 626 regenerates the first feature data through the adjusted dimension reduction process and provides the generated first feature data to the synthesis unit 663.

[0334] Repeat the above process until the similarity score converges.

[0335] Then, when the similarity score converges, the first dimension reduction unit 626 provides the first feature data that has been determined to be converged to the machine learning unit 128. In addition, the first feature before the first similarity score converges is also called the first temporary feature.

[0336] As described above, the first dimension reduction unit 626 modifies the process of reducing multiple variable parameters of dimensions and repeatedly generates a first temporary feature quantity until the similarity score converges. The first temporary feature quantity when the similarity score converges is then determined as the first feature quantity.

[0337] Similar to Embodiment 1, the second dimension reduction unit 627 generates second feature data and provides the generated second feature data to the synthesis unit 663. However, the second dimension reduction unit 627 reduces the dimensions of the multiple fixed parameters regardless of the dimension of the multiple fixed parameters, thereby generating the second feature.

[0338] Then, the second dimension reduction unit 627 obtains from the comparison unit 664 the similarity score calculated by comparing the synthetic feature data representing the first feature data and the second feature data with the fourth feature data, and determines whether the similarity score has converged.

[0339] If the similarity score does not converge, the second dimension reduction unit 627 changes the dimension reduction process in a way that increases the similarity between the synthesized feature data and the fourth feature data shown by the similarity score, thereby adjusting the dimension reduction process. Then, the second dimension reduction unit 627 regenerates the second feature data through the adjusted dimension reduction process and provides the generated second feature data to the synthesis unit 663.

[0340] Repeat the above process until the similarity score converges.

[0341] Then, if the second similarity score converges, the second dimension reduction unit 627 provides the converged second feature data to the machine learning unit 128. Furthermore, the second feature data before the second similarity score converges is also referred to as the second temporary feature data.

[0342] As described above, the second dimension reduction unit 627 modifies the processing of reducing multiple dimensions with fixed parameters and repeatedly generates a second temporary feature quantity until the similarity score converges. The second temporary feature quantity at the time of similarity score convergence is then determined as the second feature quantity.

[0343] The synthesis unit 663 synthesizes the first temporary feature quantity and the second temporary feature quantity to generate a synthesized feature quantity, such that the dimension of the synthesized feature quantity is the same as the dimension of the fourth feature quantity.

[0344] For example, the synthesis unit 663 synthesizes the first feature quantity shown by the first feature quantity data provided by the first dimension reduction unit 626 and the second feature quantity shown by the second feature quantity data output by the second dimension reduction unit 627, so that the synthesized feature quantity, i.e., the synthesized feature quantity, has the same dimension as the fourth feature quantity, and generates synthesized feature quantity data representing the synthesized feature quantity. The generated synthesized feature quantity data is provided to the comparison unit 664.

[0345] Specifically, if the method for synthesizing the feature quantity is, for example, linear coupling, then the element of the z-th dimension of the synthesized feature quantity in the processing number n is set as. nz At that time, the synthesis unit 663 can be synthesized by the following formula (8).

[0346]

[0347] Here, wv k and wf k It is the weighting coefficient.

[0348] As another example of a synthesis method, neural networks can be cited. In this case, as nz The input values ​​are av when x = 1, 2, ..., Mv respectively. nx And let af be y = 1, 2, ..., Mf ny The output of the neural network at that time.

[0349] Furthermore, the synthesis unit 663 obtains from the comparison unit 664 a similarity score calculated by comparing the synthesized feature data representing the first feature data and the second feature data with the fourth feature data, and determines whether the similarity score has converged.

[0350] If the similarity score does not converge, the synthesis unit 663 changes the synthesis process in a way that increases the similarity between the synthesized feature data and the fourth feature data shown by the similarity score, thereby adjusting the synthesis process.

[0351] In other words, the synthesis unit 663 modifies the processing of the synthesis of the first temporary feature and the second temporary feature until the similarity score converges.

[0352] The comparison unit 664 calculates a similarity score that represents the degree of similarity between the synthetic feature and the fourth feature.

[0353] For example, the comparison unit 664 compares the fourth feature data with the synthesized feature data provided from the synthesis unit 663 and calculates the similarity score α.

[0354] As a similarity score, specifically, the absolute value of the correlation coefficient, cross-entropy, KL information, and other mutual information are used. Furthermore, the more similar the compared data, the lower the value of the cross-entropy or mutual information. Therefore, when used for similarity scores, the signs of positive and negative values ​​are reversed, or a reciprocal form is used.

[0355] The autoencoder is a preferred example of the dimension reduction processing of the first dimension reduction unit 626, the second dimension reduction unit 627 and the fourth dimension reduction unit 660 in Embodiment 6. When the autoencoder is learning, the first dimension reduction unit 626 and the second dimension reduction unit 627 add the aforementioned cross-entropy or KL information to the loss function, thereby obtaining a dimension reduction effect close to that of the fourth dimension reduction unit 660.

[0356] Furthermore, if the change in similarity score since the last score is below a predetermined threshold, it is considered that the similarity score has converged.

[0357] The hardware structure of the processing condition search device 620 in Embodiment 6 described above is the same as that of the processing condition search device 120 in Embodiment 1. For example, the fourth dimension reduction unit 660, the synthesis unit 663, and the comparison unit 664 can also be implemented by the processing circuit 140.

[0358] Figure 17 This is a flowchart illustrating the operations of the first dimension reduction unit 626, the second dimension reduction unit 627, the fourth dimension reduction unit 660, the synthesis unit 663, and the comparison unit 664 in Embodiment 6.

[0359] First, the fourth dimension reduction unit 560 reads the processing conditions (in other words, parameters) contained in the processing result evaluation information stored in the processing result evaluation storage unit 123, performs dimension reduction on them, and thereby generates fourth feature data (S80). The generated fourth feature data is provided to the comparison unit 664.

[0360] Similar to Embodiment 1, the first dimension reduction unit 626 reduces the dimensions of the variable parameter data from the parameter sorting unit 125 to generate first feature data (S81). The first feature data is provided to the synthesis unit 663.

[0361] Similar to Embodiment 1, the second dimension reduction unit 627 reduces the dimensions of the fixed parameter data from the parameter sorting unit 125 to generate second feature data (S82). The second feature data is then provided to the synthesis unit 663.

[0362] The synthesis unit 663 synthesizes the first feature quantity shown by the first feature quantity data provided by the first dimension reduction unit 626 and the second feature quantity shown by the second feature quantity data output by the second dimension reduction unit 627, so that the synthesized feature quantity, i.e., the synthesized feature quantity, has the same dimension as the fourth feature quantity, and generates synthesized feature quantity data representing the synthesized feature quantity (S83). The generated synthesized feature quantity data is provided to the comparison unit 664.

[0363] The comparison unit 664 compares the fourth feature data and the synthesized feature data, and calculates the similarity score α (S84). The similarity score α is provided to the first dimension reduction unit 626, the second dimension reduction unit 627, and the synthesis unit 663.

[0364] The first dimension reduction unit 626, the second dimension reduction unit 627, and the synthesis unit 663 determine whether the similarity score α has converged (S85). If the similarity score α has not converged (S85: No), the process proceeds to step S86. If the similarity score α has converged (S85: Yes), the first dimension reduction unit 626 provides the first feature data generated in step S81 to the machine learning unit 128, and the second dimension reduction unit 627 provides the second feature data generated in step S82 to the machine learning unit 128, ending the process.

[0365] In step S86, the first dimension reduction unit 626 and the second dimension reduction unit 627 change the dimension reduction process in a way that increases the similarity score α.

[0366] Next, the synthesis unit 663 changes the synthesis process in a way that increases the similarity score α (S87). Then, the process returns to step S81.

[0367] As described above, according to Embodiment 6, when adjusting the dimensionality reduction of the sorted variable parameter data and fixed parameter data by using the result of uniform dimensionality reduction without sorting parameter data as a reference, the feature values ​​of the variable parameter data and fixed parameter data are synthesized, and a similarity score is obtained by comparing them in a way that makes them the same dimension as the feature values ​​obtained by uniform dimensionality reduction without sorting parameter data. Therefore, a reduction effect that is closer to the case of uniform dimensionality reduction without sorting parameter data can be obtained.

[0368] Furthermore, the parameter sorting unit 125 in Embodiment 6 is the same as the parameter sorting unit 125 in Embodiment 1; however, Embodiment 6 is not limited to this example. For instance, the parameter sorting unit 125 in Embodiment 6 could also be the parameter sorting unit 225 in Embodiment 2 or the parameter sorting unit 325 in Embodiment 3.

[0369] Furthermore, the optimal processing condition search unit 132 in Embodiment 6 can also be the optimal processing condition search unit 432 in Embodiment 4.

[0370] Label Explanation

[0371] 100, 200, 300, 400, 500, 600: Processing system; 110: Processing machine; 120, 220, 320, 420, 520, 620: Processing condition search device; 121: Processing result acquisition unit; 122: Processing result evaluation unit; 123: Processing result evaluation storage unit; 124: Sorting mark storage unit; 125, 225, 325: Parameter sorting unit; 126, 526, 626: First dimension reduction unit; 127, 527, 627: Second dimension reduction unit; 128: Machine learning unit; 129: Model storage unit; 130: Fixed parameter storage unit. Section; 131: Third Dimension Reduction Section; 132, 432: Optimal Processing Condition Search Section; 133: Dimension Restoration Section; 134: Processing Condition Instruction Section; 250: Initial Allocation Section; 251: Parameter Data Storage Section; 252: Variable Parameter Data Storage Section; 253: Fixed Parameter Data Storage Section; 254, 354: Parameter Allocation Section; 255: Correlation Analysis Section; 256, 356: Reallocation Section; 357: Contribution Analysis Section; 258: Output Section; 560: Fourth Dimension Reduction Section; 561: First Comparison Section; 562: Second Comparison Section; 663: Synthesis Section; 664: Comparison Section.

Claims

1. A processing condition search device, characterized in that, The processing condition search device has the following features: The processing result evaluation storage unit stores processing result evaluation information that represents multiple processing conditions with multiple parameters and multiple evaluation values ​​for multiple processing results under the multiple processing conditions. The parameter sorting unit sorts the plurality of parameters into a plurality of variable parameters that can be changed and one or more fixed parameters that cannot be changed; The first dimension processing unit generates a first feature quantity below the first dimension, which is a predetermined dimension, based on the plurality of variable parameters, thereby generating one or more first feature quantities corresponding to the plurality of processing conditions. The second dimension processing unit generates a second feature quantity below the second dimension, which is a predetermined dimension, based on the one or more fixed parameters, thereby generating one or more second feature quantities corresponding to the plurality of processing conditions; The machine learning department learns the relationship between the one or more first features, the one or more second features, and the plurality of evaluation values, thereby generating a learning model; The third dimension processing unit generates a third feature quantity below the second dimension based on one or more fixed parameters used in the processing conditions that serve as the search object, i.e., object processing conditions. The search unit uses the third feature quantity and the learning model to search for the optimal value of the feature quantity of multiple variable parameters used in the object processing conditions, i.e., multiple object variable parameters. as well as The determination unit determines the processing conditions, i.e., the search processing conditions, as the processing conditions for the object, based on the optimal value and the one or more fixed parameters of the object.

2. The processing condition search device according to claim 1, characterized in that, When the dimensions of the plurality of variable parameters are larger than the first dimension, the first dimension processing unit reduces the dimensions of the plurality of variable parameters, thereby generating the first feature quantity.

3. The processing condition search device according to claim 1 or 2, characterized in that, When the dimension of the plurality of fixed parameters is larger than the second dimension, the second dimension processing unit reduces the dimension of the plurality of fixed parameters, thereby generating the second feature quantity.

4. The processing condition search device according to any one of claims 1 to 3, characterized in that, When the dimension of the multiple fixed parameters of the objects is larger than the second dimension, the third dimension processing unit reduces the dimension of the multiple fixed parameters of the objects, thereby generating the third feature quantity.

5. The processing condition search device according to any one of claims 1 to 4, characterized in that, When the dimensions of the plurality of variable parameters are larger than the first dimension, the determining unit reconstructs the plurality of parameters from the optimal value in such a way that the dimensions of the plurality of variable parameters are the same as the dimensions of the plurality of variable parameters.

6. The processing condition search device according to claim 1, characterized in that, The first dimension processing unit reduces the dimension of the plurality of variable parameters, thereby generating a first temporary feature quantity. The second dimension processing unit reduces the dimensions of the plurality of fixed parameters, thereby generating a second temporary feature quantity. The processing condition search device also has: A dimension reduction unit reduces the dimensions of the plurality of parameters, thereby generating a fourth feature quantity; A first comparison unit calculates a first similarity score that represents the degree of similarity between the first temporary feature and the fourth feature; as well as The second comparison unit calculates a second similarity score, which represents the degree of similarity between the second temporary feature and the fourth feature. The first dimension processing unit modifies the process of reducing the dimensions of the multiple variable parameters and repeatedly generates the first temporary feature value until the first similarity score converges. The first temporary feature value at the point where the first similarity score converges is then set as the first feature value. The second dimension processing unit modifies the process of reducing the dimensions of the multiple fixed parameters and repeatedly generates the second temporary feature value until the second similarity score converges. The second temporary feature value when the second similarity score converges is set as the second feature value.

7. The processing condition search device according to claim 1, characterized in that, The first dimension processing unit reduces the dimension of the plurality of variable parameters, thereby generating a first temporary feature quantity. The second dimension processing unit reduces the dimensions of the plurality of fixed parameters, thereby generating a second temporary feature quantity. The processing condition search device also has: A dimension reduction unit reduces the dimensions of the plurality of parameters, thereby generating a fourth feature quantity; A synthesis unit that synthesizes the first temporary feature and the second temporary feature to generate a synthesized feature, such that the dimension of the synthesized feature is the same as the dimension of the fourth feature; and The comparison unit calculates a similarity score, which represents the degree of similarity between the synthesized feature and the fourth feature. The first dimension processing unit modifies the process of reducing the dimensions of the multiple variable parameters and repeatedly generates the first temporary feature value until the similarity score converges. The first temporary feature value at which the similarity score converges is then set as the first feature value. The second dimension processing unit modifies the process of reducing the dimensions of the multiple fixed parameters and repeatedly generates the second temporary feature quantity until the similarity score converges. The second temporary feature quantity at the point of convergence is then set as the second feature quantity. The synthesis unit modifies the processing of synthesizing the first temporary feature and the second temporary feature until the similarity score converges.

8. The processing condition search device according to claim 6 or 7, characterized in that, The determining part reconstructs the parameters from the optimal value in a manner that makes it the same dimension as the plurality of variable parameters.

9. The processing condition search device according to any one of claims 1 to 8, characterized in that, The processing condition search device also includes a sorting flag storage unit, which stores sorting flags indicating whether a parameter is variable or fixed according to each category of the plurality of parameters for sorting the plurality of parameters. The parameter sorting unit sorts the plurality of parameters into the plurality of variable parameters and the one or more fixed parameters by referring to the sorting mark.

10. The processing condition search device according to any one of claims 1 to 8, characterized in that, The processing condition search device also includes a sorting flag storage unit, which stores sorting flags indicating whether a parameter is variable or fixed according to each category of the plurality of parameters for sorting the plurality of parameters. The parameter sorting unit has: An initial allocation unit, by referring to the sorting flag, allocates the plurality of parameters into a plurality of initial variable parameters and one or more initial fixed parameters; The correlation analysis unit determines multiple combinations of each initial variable parameter among the plurality of initial variable parameters and each initial fixed parameter among the one or more initial fixed parameters, and analyzes the correlation of each of the plurality of combinations. as well as The redistribution unit redistributes the initial variable parameters contained in the combinations among the plurality of combinations whose correlation is higher than a predetermined threshold as initial fixed parameters, thereby setting the plurality of initial variable parameters after redistribution as the plurality of variable parameters, and setting one or more initial fixed parameters after redistribution as the one or more fixed parameters.

11. The processing condition search device according to any one of claims 1 to 8, characterized in that, The processing condition search device also includes a sorting flag storage unit, which stores sorting flags indicating whether a parameter is variable or fixed according to each category of the plurality of parameters for sorting the plurality of parameters. The parameter sorting unit has: An initial allocation unit, by referring to the sorting flag, allocates the plurality of parameters into a plurality of initial variable parameters and one or more initial fixed parameters; The contribution analysis unit analyzes the contribution of each of the plurality of initial variable parameters to the evaluation value; as well as The redistribution unit redistributes the initial variable parameters whose contribution is below a predetermined threshold from the plurality of initial variable parameters as initial fixed parameters, thereby setting the redistributed plurality of initial variable parameters as the plurality of variable parameters, and setting the redistributed one or more initial fixed parameters as the one or more fixed parameters.

12. The processing condition search device according to any one of claims 1 to 11, characterized in that, The processing condition search device also has: A processing condition instruction unit provides the searched processing conditions to a processing machine, causing the processing machine to perform processing under the searched processing conditions, and appends the searched processing conditions to the processing result evaluation information; and The processing result evaluation unit evaluates the processing result performed by the processing machine, thereby determining an evaluation value. The determined evaluation value is then added to the processing result evaluation information, corresponding to the search processing conditions. When performing the search for the optimal value for the first time, the search unit determines one or more evaluation values ​​that are higher than a predetermined evaluation value from the plurality of evaluation values, determines one or more second feature values ​​from the plurality of second feature values ​​that correspond to the one or more evaluation values, determines one of the more than one second feature values ​​that is closest to the third feature value, determines a first feature value that corresponds to the first second feature value, and sets the first feature value as the optimal value.

13. A computer-readable recording medium storing a computer program, characterized in that, When the computer program is executed by the processor The processing result evaluation information includes multiple parameters that can be changed and one or more fixed parameters that cannot be changed. The processing result evaluation information represents multiple processing conditions with the multiple parameters and multiple evaluation values ​​for multiple processing results under the multiple processing conditions. Based on the plurality of variable parameters, a first feature quantity below the first dimension, which is a predetermined dimension, is generated, thereby generating one or more first feature quantities corresponding to the plurality of processing conditions. Based on the one or more fixed parameters, a second feature quantity below the second dimension, which is a predetermined dimension, is generated, thereby generating one or more second feature quantities corresponding to the plurality of processing conditions. The relationship between the one or more first features, the one or more second features, and the plurality of evaluation values ​​is learned to generate a learning model. Based on one or more fixed parameters used in the processing conditions (i.e., object processing conditions) that serve as the search object, a third feature quantity below the second dimension is generated. Using the third feature and the learning model, the optimal values ​​of the feature quantities of the multiple variable parameters used in the object processing conditions, i.e., the multiple object variable parameters, are searched. Based on the optimal value and the fixed parameters of the one or more objects, the processing conditions searched for as the processing conditions of the objects are determined, namely the search processing conditions.

14. A method for searching processing conditions, characterized in that, The processing result evaluation information includes multiple parameters that can be changed and one or more fixed parameters that cannot be changed. The processing result evaluation information represents multiple processing conditions with the multiple parameters and multiple evaluation values ​​for multiple processing results under the multiple processing conditions. Based on the plurality of variable parameters, a first feature quantity below the first dimension, which is a predetermined dimension, is generated, thereby generating one or more first feature quantities corresponding to the plurality of processing conditions. Based on the one or more fixed parameters, a second feature quantity below the second dimension, which is a predetermined dimension, is generated, thereby generating one or more second feature quantities corresponding to the plurality of processing conditions. The relationship between the one or more first features, the one or more second features, and the plurality of evaluation values ​​is learned to generate a learning model. Based on one or more fixed parameters used in the processing conditions (i.e., object processing conditions) that serve as the search object, a third feature quantity below the second dimension is generated. Using the third feature and the learning model, the optimal values ​​of the feature quantities of the multiple variable parameters used in the object processing conditions, i.e., the multiple object variable parameters, are searched. Based on the optimal value and the fixed parameters of the one or more objects, the processing conditions searched for as the processing conditions of the objects are determined, namely the search processing conditions.

Images