Processing condition exploration device and processing condition exploration method

Abstract

The processing condition exploration device (1) comprises: a processing condition generating unit (11) which generates processing conditions to be set for a processing machine (2); a processing state collecting unit (12) which collects processing states; a processing result collecting unit (13) which collects processing results of processing performed according to the processing conditions; a processing result evaluating unit (14) which calculates an evaluation value of processing based on the processing results; an evaluation value prediction model (16) which predicts an evaluation value corresponding to an untried processing condition based on the processing conditions, the processing state and the evaluation value; and a model A construction unit (15) constructs an evaluation value prediction model (16) based on the degree of change in the relationship between the processing conditions and the evaluation value, and performs weighting corresponding to the processing state on the evaluation value prediction model (16). The processing condition generation unit (11) uses the predicted value of the evaluation value to generate the processing conditions that should be tried next until it is determined that the exploration is completed, and repeats the various processes involved in the processing state collection unit (12), the processing result collection unit (13), the processing result evaluation unit (14), the evaluation value prediction model (16) and the model construction unit (15).

Description

Technical Field

[0001] The present invention relates to a processing condition exploration device and a processing condition exploration method for exploring processing conditions. Background Art

[0002] Generally speaking, different machining results are achieved depending on the machining conditions set on the machining machine. Furthermore, even with the same machining conditions, variations in the machining state, which represents the state of the machining machine and the workpiece, can also lead to different machining results. Therefore, in order to achieve machining results that meet the required machining specifications, it is necessary to search for appropriate machining conditions that correspond to the machining state.

[0003] In contrast, for example, Patent Document 1 describes a control device that predicts the machining results of a workpiece obtained by a machining machine. This control device predicts the impact of changes in state quantities on the machining results by using a learning model that represents the correlation between state variables representing changes in state quantities, which represent machining conditions, and judgment data representing machining results.

[0004] Patent Document 1: Japanese Patent Application Laid-Open No. 2019-32649 Summary of the Invention

[0005] If the machining state changes, even if the machining conditions set on the machining machine remain the same, the machining results will be different, and the relationship between the machining conditions and the evaluation value of the machining performed under those conditions will also change. Therefore, in order to use the control device described in Patent Document 1 to search for machining conditions that meet the required machining specifications, it is necessary to build a learning model that collects a large amount of learning data in advance and can accurately predict the impact of changes in the machining state on the machining results.

[0006] The present invention is proposed to solve the above-mentioned problems, and its purpose is to obtain a processing condition exploration device and a processing condition exploration method that can explore processing conditions without collecting a large amount of data in advance to construct a learning model.

[0007] The processing condition exploration device involved in the present invention comprises: a processing condition generating unit, which generates processing conditions set for a processing machine; a processing state collecting unit, which collects processing states representing various states of the processing machine, the workpiece to be processed, and the setting environment of the processing machine that performs processing according to the processing conditions; a processing result collecting unit, which collects processing results of processing performed according to the processing conditions; a processing result evaluating unit, which calculates an evaluation value of processing performed according to the processing conditions based on the processing results collected by the processing result collecting unit; an evaluation value prediction model, which calculates an evaluation value prediction model based on the processing conditions generated by the processing condition generating unit, the processing states collected by the processing state collecting unit, and The evaluation value calculated by the processing result evaluation unit predicts the evaluation value corresponding to the processing condition that has not been tried; and the model construction unit, which constructs an evaluation value prediction model when it is determined that the degree of change in the processing state is greater than or equal to the threshold, and performs weighting corresponding to the processing state on the evaluation value prediction model. The processing condition generation unit uses the predicted value of the evaluation value calculated by the evaluation value prediction model to generate the processing condition that should be tried next until the processing condition generation unit determines that the exploration of the processing condition is terminated, and repeats the various processes involved in the processing state collection unit, the processing result collection unit, the processing result evaluation unit, the evaluation value prediction model and the model construction unit.

[0008] Effects of the Invention

[0009] According to the present invention, based on the machining conditions, machining state, and machining evaluation values ​​set for the machining machine, an evaluation value corresponding to an untested machining condition is predicted. An evaluation value prediction model is constructed based on the degree of change in the relationship between the machining conditions and the evaluation values, and the evaluation value prediction model is weighted according to the machining state. Even if the relationship between the machining conditions and the evaluation values ​​changes due to a change in the machining state, a new evaluation value prediction model is constructed in accordance with the degree of change, and the evaluation value prediction model is weighted according to the machining state. Thus, the machining condition exploration device of the present invention can explore machining conditions without having to collect a large amount of data in advance to construct a learning model. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 This is a block diagram showing the configuration of the processing condition search device according to the first embodiment.

[0011] Figure 2 This is a flowchart showing the processing condition search method involved in the first embodiment.

[0012] Figure 3 This is a graph showing the relationship between processing conditions, predicted values ​​of evaluation values ​​corresponding thereto, and an index showing the unreliability of the prediction of the evaluation values.

[0013] Figure 4A is a block diagram showing a hardware configuration for realizing the functions of the processing condition search device according to the first embodiment. Figure 4B This is a block diagram showing a hardware configuration for executing software that realizes the functions of the processing condition search device according to the first embodiment. DETAILED DESCRIPTION

[0014] Implementation method 1.

[0015] Figure 1 This is a block diagram showing the structure of a processing condition search device 1 according to Embodiment 1. The processing condition search device 1 searches for optimal processing conditions from among a plurality of processing conditions that can be set for a processing machine 2 and sets the processing conditions resulting from the search on the processing machine 2. For example, the optimal processing conditions are those that produce processing results that meet the required processing specifications. Furthermore, a display unit 3 displays the processing conditions and the like discovered by the processing condition search device 1. For example, the display unit 3 displays the processing conditions set on the processing machine 2 and the evaluation values ​​of the processing performed by the processing machine 2 according to those processing conditions.

[0016] Processing machine 2 is an industrial device that performs processing according to processing conditions. For example, it cuts or grinds a metal object, or removes unwanted portions using electrical or other energy, thereby shaping the object into a desired shape. The object to be processed is not limited to metal; it can also be ceramic, glass, or wood. Examples of processing machine 2 include laser processing machines, electrical discharge machines, cutting machines, grinding machines, electrolytic machines, ultrasonic machines, and electron beam machines. The following description assumes that processing machine 2 is an electrical discharge machine, specifically a contour-carving electrical discharge machine.

[0017] Machining conditions consist of a combination of multiple control parameters used to control the processing machine 2. Generally speaking, different machining conditions result in different machining results. Furthermore, even when the machining conditions set for the processing machine 2 remain the same, variations in the machining state—the state of the processing machine 2, the workpiece being processed, and the environment in which the processing machine 2 is installed—can sometimes lead to different machining results. In other words, in order to achieve the optimal machining result that meets the required machining specifications, it is necessary to search for appropriate machining conditions that correspond to the machining state.

[0018] For example, in the machining of a shape carving electric discharge machine, there are three control parameters that can be adjusted. If the value of each control parameter can be selected through 10 levels, there are 10 machining conditions consisting of the combination of each control parameter. 3= 1000. In contrast, the processing condition search device 1 does not sequentially search for many processing conditions that can be set in the processing machine 2, but rather narrows the number of processing conditions to be searched in accordance with changes in the processing state, thereby reducing the number of trials for searching processing conditions.

[0019] exist Figure 1 In the embodiment, the processing condition exploration device 1 includes a processing condition generation unit 11, a processing state collection unit 12, a processing result collection unit 13, a processing result evaluation unit 14, a model construction unit 15, and an evaluation value prediction model 16. Furthermore, the processing condition exploration device 1 includes a processing state storage unit 17A, a search result storage unit 17B, a model storage unit 17C, a prediction result storage unit 17D, and an unreliability storage unit 17E. Furthermore, all or part of the storage units 17A to 17E may be provided by an external device provided independently of the processing condition exploration device 1.

[0020] The processing condition generating unit 11 generates processing conditions and sets the generated processing conditions in the processing machine 2. The processing condition generating unit 11 includes a processing condition calculating unit 11a, an actual processing instruction unit 11b, and a search completion determining unit 11c.

[0021] The processing condition calculation unit 11a calculates the processing conditions set for the processing machine 2. For example, the processing condition calculation unit 11a selects a combination corresponding to the processing content from a combination of multiple control parameters of the processing machine 2 and the range of values ​​these control parameters can take, and calculates the processing conditions based on the selected combination. Examples of control parameters include laser output, cutting speed, beam magnification, focal position, and gas pressure.

[0022] The actual processing instruction unit 11b causes the processing machine 2 to perform processing based on the processing conditions calculated by the processing condition calculation unit 11a. For example, the actual processing instruction unit 11b generates a command for operating the processing machine 2 based on the processing conditions calculated by the processing condition calculation unit 11a, and outputs the generated command to the processing machine 2.

[0023] The search completion determination unit 11c determines whether to terminate the search for processing conditions based on the data stored in the prediction result storage unit 17D or the unreliability storage unit 17E. If the search completion determination unit 11c determines that additional search for processing conditions is necessary, the processing condition calculation unit 11a generates processing conditions for the next processing to be performed. If it determines that additional search for processing conditions is not necessary, the processing condition calculation unit 11a uses the predicted evaluation values ​​stored in the prediction result storage unit 17D to determine the processing condition with the highest predicted evaluation value as the optimal processing condition.

[0024] The processing status collection unit 12 collects data representing the processing status of processing performed by the processing machine 2. The processing status includes the status of the processing machine 2, the status of the workpiece being processed, and the status of the environment in which the processing machine 2 is installed. For example, the processing status includes the temperature of the processing machine 2, the temperature and thickness of the workpiece, and the temperature and humidity of the room in which the processing machine 2 is installed, which can affect the processing results but are beyond the user's control. The data representing the processing status is a numerical representation of each state represented by the processing status. The data collected by the processing status collection unit 12 is stored in the processing status storage unit 17A.

[0025] The processing result collection unit 13 collects processing results from the processing machine 2 according to the processing conditions. For example, the processing result collection unit 13 collects processing-related information obtained during and after processing. This processing-related information includes, for example, detection data of sound or light observed during processing, the number of discharge pulses, and the surface condition of the workpiece after processing.

[0026] The processing result evaluation unit 14 calculates an evaluation value for the processing performed by the processing machine 2 according to the processing conditions based on the processing results collected by the processing result collection unit 13. The evaluation value indicates whether the processing is acceptable or not, and is, for example, a value between 0 and 1. A larger evaluation value indicates a better processing result. The evaluation value is 1 for the best processing, and 0 for the worst processing. Furthermore, the processing result evaluation unit 14 stores the combination of processing conditions and evaluation values ​​as a search result in the search result storage unit 17B. The search result storage unit 17B stores the search results.

[0027] The model construction unit 15 constructs an evaluation value prediction model 16 based on the degree of change in the relationship between the processing conditions and the evaluation values, and applies weighting to the evaluation value prediction model 16 according to the processing state. For example, the model construction unit 15 calculates the degree of change in the relationship between the processing conditions and the evaluation values ​​based on the processing state information stored in the processing state storage unit 17A and the evaluation values ​​stored in the search result storage unit 17B. If the degree of change in the relationship between the processing conditions and the evaluation values ​​is greater than or equal to a threshold, the model construction unit 15 determines that a change in the processing state has occurred, requiring the construction of a new evaluation value prediction model, and constructs the evaluation value prediction model.

[0028] Furthermore, the model construction unit 15 weights the evaluation value prediction models 16 according to the machining state. For example, the predicted values ​​of each evaluation value obtained by multiple evaluation value prediction models 16 are combined using a weight corresponding to the magnitude of the state quantity obtained by digitizing the machining state corresponding to the evaluation value prediction model 16. In the combined predicted values ​​of the evaluation value, the more weighted the evaluation value prediction model 16, the more weighted its prediction result is given.

[0029] The evaluation value prediction model 16 predicts the evaluation value corresponding to the untried processing conditions based on the processing conditions generated by the processing condition generation unit 11, the processing status collected by the processing status collection unit 12, and the evaluation value calculated by the processing result evaluation unit 14. Figure 1 In FIG, the evaluation value prediction model 16 includes an evaluation value prediction unit 16 a and an unreliability evaluation unit 16 b .

[0030] The evaluation value prediction unit 16a predicts the evaluation value corresponding to the processing conditions that have not been tested (not yet processed) based on the processing conditions and the corresponding evaluation values ​​stored in the search result storage unit 17B. Furthermore, the evaluation value prediction unit 16a stores the processing conditions and the predicted evaluation values ​​corresponding thereto in the prediction result storage unit 17D. The prediction result storage unit 17D stores the untested processing conditions and the predicted evaluation values ​​corresponding thereto in association with each other.

[0031] The unreliability evaluation unit 16b calculates an index representing the unreliability of the evaluation value predicted by the evaluation value prediction unit 16a. For example, the unreliability evaluation unit 16b uses the exploration results stored in the exploration result storage unit 17B to calculate the unreliability of the evaluation value relative to the predicted value, that is, an index representing the ease of deviation from the prediction. The unreliability evaluation unit 16b stores unreliability information including the calculated index value and processing conditions in the unreliability storage unit 17E. In the unreliability storage unit 17E, untested processing conditions are associated with the index value representing the unreliability of the prediction of the corresponding evaluation value and are stored.

[0032] Figure 2This is a flowchart showing the processing condition search method involved in embodiment 1, showing a series of processes until the optimal processing conditions are searched by the processing condition search device 1. When the search process for the optimal processing conditions begins, the processing condition generation unit 11 generates initial processing conditions (step ST1). The processing condition calculation unit 11a selects a certain number of combinations from all combinations of control parameters that can be set for the processing machine 2 as initial processing conditions. As a method for selecting the initial processing conditions, there are experimental planning methods, optimal planning methods, or random sampling. In addition, when the user finds the optimal processing conditions based on the past usage performance of the processing machine 2, the processing conditions specified by the user can be used as the initial processing conditions.

[0033] For example, if there are three control parameters constituting the machining conditions, and if a value set to the machining machine 2 is selected from ten ranges for each control parameter, the total number of machining condition combinations is 1,000. The machining condition calculation unit 11a selects a certain number of machining conditions from among these 1,000 machining conditions. In the following description, the certain number is, for example, 5, and the machining condition calculation unit 11a selects 5 machining conditions from among the 1,000 machining conditions.

[0034] Next, the processing condition generation unit 11 sets the initial processing conditions on the processing machine 2, and causes the processing machine 2 to perform processing according to the initial processing conditions (step ST2). For example, the processing condition calculation unit 11a selects one of five initial processing conditions and outputs the selected initial processing condition to the actual processing instruction unit 11b. The actual processing instruction unit 11b generates a command for operating the processing machine 2 according to the initial processing conditions output from the processing condition calculation unit 11a and outputs the generated command to the processing machine 2. The processing machine 2 performs processing according to the initial processing conditions. In the following description, processing according to the initial processing conditions is also referred to as "initial processing."

[0035] The machining state collecting unit 12 collects data representing the machining state of machining performed by the machining machine 2 according to the initial machining conditions (step ST3). The machining state collecting unit 12 stores the machining state data collected from the machining machine 2 in the machining state storage unit 17A. Thus, the initial machining conditions and the machining state are associated and stored in the machining state storage unit 17A.

[0036] The processing result collecting unit 13 collects data indicating the processing results of the processing performed by the processing machine 2 according to the initial processing conditions (step ST4 ). The data collected by the processing result collecting unit 13 is output to the processing result evaluating unit 14 .

[0037] The processing result evaluation unit 14 calculates an evaluation value for the processing performed by the processing machine 2 according to the initial processing conditions based on the processing results collected by the processing result collection unit 13 (step ST5). The processing result evaluation unit 14 digitizes the processing results collected by the processing result collection unit 13 and calculates an evaluation value indicating whether the processing is acceptable or not. For example, the processing result evaluation unit 14 measures the sound or light detection data or the number of discharge pulses observed during processing according to the initial processing conditions and converts this measured value into an evaluation value indicating whether the processing is acceptable or not, either as a continuous value or as a discrete value with multiple levels (e.g., 10 levels). The search result storage unit 17B stores the combinations of processing conditions and evaluation values.

[0038] The processing condition calculation unit 11a confirms whether initial processing has been completed for all processing conditions selected as initial processing conditions (step ST6). If there are initial processing conditions for which initial processing has not yet been completed (step ST6: NO), the processes from steps ST1 to ST5 are sequentially repeated for the initial processing conditions for which initial processing has not yet been completed. Consequently, all initial processing conditions (e.g., five initial processing conditions) are associated with and stored in the processing status storage unit 17A. Furthermore, all initial processing conditions are associated with and stored in the search result storage unit 17B.

[0039] When the initial processing of all the initial processing conditions is completed (step ST6; YES), the processing condition calculation unit 11a confirms whether the processing according to the found optimal processing conditions is completed (step ST7). When the processing machine 2 completes the processing according to the optimal processing conditions (step ST7; YES), Figure 2 A series of processing is completed.

[0040] If machining under the optimal machining conditions has not yet been completed (step ST7; NO), the model building unit 15 uses the machining state data stored in the machining state storage unit 17A to determine whether the value of an indicator representing the degree of change in the relationship between the machining conditions and the evaluation value is greater than or equal to a threshold value (step ST8). One method for determining the degree of change is change point detection. Change point detection uses an abnormality degree as an indicator representing the degree of change. For example, the abnormality degree is a value obtained by squaring the difference between the predicted value of the machining state and the actual measured value of the machining state stored in the machining state storage unit 17A.

[0041] When calculating the predicted value of the processing state, an autoregressive model is used, for example. In the autoregressive model, the predicted value hat y(t) of the processing state at time t one step ago is calculated using the following formula (1). In the following formula (1), y(t-1) is the measured value of the processing state at the current time t-1, y(t-2) is the measured value of the processing state at the past time t-2, and y(t-3) is the measured value of the processing state at the past time t-3. The coefficient α1 is the coefficient relative to the measured value y(t-1), the coefficient α2 is the coefficient relative to the measured value y(t-2), and the coefficient α3 is the coefficient relative to the measured value y(t-3). In addition, in addition to the autoregressive model, the model construction unit 15 can use a supervised learning method such as a multiple regression model, a decision tree, or a neural network to evaluate the degree of change in the relationship between the processing conditions and the evaluation value.

[0042]

[0043] If the model construction unit 15 determines that the value of the index indicating the degree of change in the relationship between the processing conditions and the evaluation value is greater than or equal to the threshold value (step ST8; YES), it determines that a change in the processing state has occurred, for which a new evaluation value prediction model 16 should be constructed, and constructs the evaluation value prediction model 16 corresponding to the changed processing state (step ST9). When constructing the evaluation value prediction model 16, for example, a supervised learning method such as an autoregressive model, a multiple regression model, a decision tree, or a neural network can be used.

[0044] If the value of the indicator indicating the degree of change in the relationship between the machining conditions and the evaluation values ​​is determined to be less than the threshold value (step ST8; NO) or if step ST9 is completed, the model construction unit 15 weights each of the constructed evaluation value prediction models 16 according to the machining state stored in the machining state storage unit 17A (step ST10). Examples of weighting methods include linear interpolation and boosting. Alternatively, the model construction unit 15 may select and weight a single evaluation value prediction model 16 rather than weighting multiple evaluation value prediction models 16.

[0045] For example, the model construction unit 15 uses linear interpolation to weight the evaluation value prediction model 16. Using the processing state s1 at the past time t-2 and the processing state s2 at the past time t-3 stored in the processing state storage unit 17A, and the processing state s at the current time t-1 collected by the processing state collection unit 12, the model construction unit 15 calculates the weight β1(s) for the function f1 determined by the evaluation value prediction model 16 corresponding to processing state s1 and the weight β2(s) for the function f2 determined by the evaluation value prediction model 16 corresponding to processing state s2 according to the following equation (2). Using these weights, the output f(x, s) of the evaluation value prediction model 16 for processing condition x and processing state s is calculated according to the following equation (3). While the weighted values ​​are determined for two evaluation value prediction models 16, the model construction unit 15 may also calculate the weighted values ​​for three or more evaluation value prediction models 16.

[0046] β1(s)=(s2-s) / (s2-s1)

[0047] β2(s)=(s-s1) / (s2-s1) · · · (2)

[0048] f(x,s)=β1(s)f1(x)+β2(s)f2(x) · · · (3)

[0049] When processing under, for example, five initial processing conditions is completed, the evaluation value prediction unit 16a in the evaluation value prediction model 16 predicts evaluation values ​​for all 1,000 processing conditions using the processing conditions and their corresponding evaluation values ​​stored in the search result storage unit 17B (step ST11). The evaluation values ​​predicted by the evaluation value prediction unit 16a are stored in the prediction result storage unit 17D (step ST12).

[0050] As a prediction method of the evaluation value involved in the evaluation value prediction unit 16a, there is, for example, Gaussian process regression. In Gaussian process regression, the evaluation value prediction unit 16a is a probability model of the processing conditions corresponding to the evaluation values ​​constructed by assuming that the evaluation values ​​corresponding to the processing conditions are probability variables. The evaluation value prediction unit 16a is set to k, and the vector arranging the values ​​of the kernels corresponding to the processing conditions and the kernels stored in the exploration result storage unit 17B is set to t. The evaluation value prediction unit 16a can calculate the predicted value m(x) of the evaluation value for the processing condition x according to the following formula (4). In the following formula (4), C NThe evaluation value prediction unit 16a may predict the evaluation value using a supervised learning method such as a decision tree, linear regression, boosting method, or neural network instead of using Gaussian process regression.

[0051] m(x)=k T ·(C N －1 )·t···(4)

[0052] When machining under, for example, five initial machining conditions is completed, the unreliability evaluation unit 16b in the evaluation value prediction model 16 calculates an index representing the predicted unreliability of the evaluation values ​​for all 1,000 machining conditions using the machining conditions and their corresponding evaluation values ​​stored in the search result storage unit 17B (step ST13). The value of the index calculated by the unreliability evaluation unit 16b is stored in the unreliability storage unit 17E (step ST14).

[0053] As a calculation method of the above-mentioned index involved in the unreliability evaluation unit 16b, there is, for example, Gaussian process regression. In Gaussian process regression, when a vector that arranges the values ​​of the kernels corresponding to the processing conditions stored in the exploration result storage unit 17B is set as k, and a vectorless value obtained by adding the value of the kernel between the processing conditions x to the accuracy parameter of the evaluation value prediction unit 16a is set as c, the unreliability evaluation unit 16b can calculate the index σ representing the unreliability of the prediction of the evaluation value for the untested processing condition x according to the following formula (5): 2 (x) is calculated. In the following formula (5), C N The unreliability evaluation unit 16b may calculate the above-mentioned index by using density estimation, mixed density network, or regression of KL divergence instead of Gaussian process regression.

[0054] σ 2 (x) = c - k T ·(C N －1 )·k···(5)

[0055] Figure 3 The graph shows the relationship between machining conditions, the predicted values ​​of the corresponding evaluation values, and an index indicating the uncertainty of the evaluation value prediction. The evaluation value prediction model 16 uses, for example, Gaussian process regression, and predicts the evaluation value according to a Gaussian distribution. Figure 3 The black plot points shown are the processing conditions and evaluation values ​​stored in the search result storage unit 17B. Figure 3In the example, when the predicted value of the evaluation value is set to the mean m(x) of the Gaussian distribution and the indicator representing the unreliability of the prediction of the evaluation value is set to the standard deviation σ(x) of the Gaussian distribution, even if the prediction of the evaluation value deviates, the black plotted point will statistically show that it enters the range less than or equal to m(x)+2σ(x) and greater than or equal to m(x)-2σ(x) with a probability of approximately 95%.

[0056] The search completion determination unit 11c of the processing condition generating unit 11 determines whether to terminate the search for processing conditions using the predicted values ​​of the evaluation values ​​of the processing conditions stored in the prediction result storage unit 17D and the index indicating the predicted unreliability of the evaluation values ​​stored in the unreliability storage unit 17E (step ST15). For example, the search completion determination unit 11c compares the value of the index indicating the predicted unreliability of the evaluation values ​​of all processing conditions that have been searched so far, stored in the unreliability storage unit 17E, with a threshold value. If the value of the index is less than or equal to the threshold value, the search for processing conditions is determined to have been terminated, and the search for processing conditions is terminated.

[0057] Furthermore, the search end determination unit 11c compares an indicator indicating the predicted unreliability of the evaluation values ​​for all processing conditions with a threshold value. This determines that the search for processing conditions ends when the number of processing conditions that are likely to significantly deviate from the predicted evaluation values ​​is less than or equal to a specified number. This is because if there are processing conditions where the actual evaluation values ​​differ significantly from the predicted evaluation values, the predicted values ​​will vary significantly across a wide range of the search space for processing conditions.

[0058] For example, the search completion determination unit 11c uses a processing condition x, a predicted value m(x) for the evaluation value of that processing condition x, and an indicator (standard deviation) σ(x) indicating the uncertainty of the prediction of that evaluation value. The larger the value of m(x) + κσ(x), the higher the value, the more valuable it is to search for that processing condition. κ is a parameter determined before the search for processing conditions begins. A smaller value for κ selects a processing condition with a higher predicted evaluation value, while a larger value for κ selects a processing condition with a higher probability of a significant deviation from the predicted evaluation value. The value of κ can remain unchanged or be changed midway.

[0059] If the search for processing conditions is determined to be completed (step ST15; YES), the search completion determination unit 11c extracts the processing condition with the highest predicted evaluation value from the predicted evaluation values ​​of all processing conditions stored in the prediction result storage unit 17D, and outputs the extracted processing condition to the actual processing instruction unit 11b. The actual processing instruction unit 11b outputs a command including the processing condition output from the search completion determination unit 11c to the processing machine 2, thereby setting the processing condition in the processing machine 2 (step ST16).

[0060] When it is determined that additional exploration of the processing conditions is necessary (step ST15; NO), the exploration completion determination unit 11c will continue the exploration and output it to the processing condition calculation unit 11a. When the processing condition calculation unit 11a is instructed to continue the exploration by the exploration completion determination unit 11c, it uses the predicted value of the evaluation value of the processing condition stored in the prediction result storage unit 17D to generate the processing condition to be tried next (step ST17). The processing condition to be tried next calculated by the processing condition calculation unit 11a is output to the actual processing instruction unit 11b. The actual processing instruction unit 11b outputs the instruction including the processing condition to be tried next to the processing machine 2, and sets the processing condition in the processing machine 2.

[0061] After the optimal machining conditions are set in step ST16 or the machining conditions to be tested next are set in step ST17, the machining machine 2 performs machining (step ST18). During machining performed by the machining machine 2, the machining status collection unit 12 collects data representing the machining status and stores the machining conditions and the machining status in association in the machining status storage unit 17A. The machining result collection unit 13 collects data representing the machining results and outputs them to the machining result evaluation unit 14. Based on the machining results collected by the machining result collection unit 13, the machining result evaluation unit 14 calculates an evaluation value for the machining performed by the machining machine 2 (step ST19). Next, the process proceeds to step ST7 to execute the aforementioned processing.

[0062] The display unit 3 displays the processing conditions and the evaluation values ​​corresponding to the processing conditions obtained during the processing condition search performed by the processing condition search device 1. Furthermore, the display unit 3 displays the processing conditions and the predicted values ​​of the evaluation values ​​corresponding to the processing conditions, or the optimal processing conditions resulting from the search. Specifically, the display unit 3 displays at least one of the processing conditions and the evaluation values ​​corresponding to the processing conditions read from the search result storage unit 17B, the processing conditions and the predicted values ​​of the evaluation values ​​corresponding to the processing conditions read from the predicted result storage unit 17D, or the optimal processing conditions resulting from the search output from the processing condition calculation unit 11a. Thus, the processing operator can identify the status of the processing condition search and the results of the search by referring to the information displayed on the display unit 3.

[0063] The hardware configuration for realizing the functions of the processing condition search device 1 is as follows.

[0064] The functions of the processing condition generation unit 11, the processing state collection unit 12, the processing result collection unit 13, the processing result evaluation unit 14, the model construction unit 15 and the evaluation value prediction model 16 in the processing condition exploration device 1 are realized by the processing circuit. Figure 2 The processing circuit may be dedicated hardware or a CPU (Central Processing Unit) that executes a program stored in a memory.

[0065] Figure 4A 1 is a block diagram showing a hardware configuration for realizing the functions of the processing condition search device 1. Figure 4B 1 is a block diagram showing the hardware structure of the software that executes the functions of the processing condition search device 1. Figure 4A and Figure 4B In the embodiment, the input interface 100 relays data indicating the machining status and machining results output from the machining machine 2 to the machining condition search device 1, and relays stored data output from the storage units 17A to 17E to the machining condition search device 1. The output interface 101 relays information output from the machining condition search device 1 to the display unit 3, or data output from the machining condition search device 1 to the storage units 17A to 17E.

[0066] In the processing circuit Figure 4AIn the case of dedicated hardware processing circuit 102 shown, processing circuit 102 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field-Programmable Gate Array), or a combination thereof. The functions of processing condition generation unit 11, processing state collection unit 12, processing result collection unit 13, processing result evaluation unit 14, model construction unit 15, and evaluation value prediction model 16 in processing condition exploration device 1 may be implemented by separate processing circuits, or these functions may be integrated and implemented by a single processing circuit.

[0067] In the processing circuit Figure 4B In the case of the processor 103 shown, the functions of the processing condition generation unit 11, processing state collection unit 12, processing result collection unit 13, processing result evaluation unit 14, model construction unit 15, and evaluation value prediction model 16 in the processing condition search device 1 are implemented by software, firmware, or a combination of software and firmware. Furthermore, the software or firmware is described as a program and stored in the memory 104.

[0068] The processor 103 reads and executes the program stored in the memory 104, thereby realizing the functions of the processing condition generation unit 11, the processing state collection unit 12, the processing result collection unit 13, the processing result evaluation unit 14, the model construction unit 15 and the evaluation value prediction model 16 in the processing condition exploration device 1. For example, the processing condition exploration device 1 includes the memory 104, which is used to generate the evaluation value prediction model 16 when the processor 103 executes the program. Figure 2 The memory 104 stores programs that ultimately execute the processing from step ST1 to step ST19 in the flowchart shown. These programs cause a computer to execute the procedures or methods of the processing condition generation unit 11, the processing status collection unit 12, the processing result collection unit 13, the processing result evaluation unit 14, the model construction unit 15, and the evaluation value prediction model 16. The memory 104 may be a computer-readable storage medium storing a program for causing a computer to function as the processing condition generation unit 11, the processing status collection unit 12, the processing result collection unit 13, the processing result evaluation unit 14, the model construction unit 15, and the evaluation value prediction model 16.

[0069] The memory 104 is, for example, a nonvolatile or volatile semiconductor memory such as RAM (Random Access Memory), ROM (Read Only Memory), flash memory, EPROM (Erasable Programmable Read Only Memory), or EEPROM (Electrically-EPROM), or a magnetic disk, a floppy disk, an optical disk, a compact disk, a minidisc, or a DVD.

[0070] The functions of the processing condition generation unit 11, processing status collection unit 12, processing result collection unit 13, processing result evaluation unit 14, model construction unit 15, and evaluation value prediction model 16 in the processing condition exploration device 1 may be partially implemented using dedicated hardware, while others may be implemented using software or firmware. For example, the processing condition generation unit 11, processing status collection unit 12, processing result collection unit 13, processing result evaluation unit 14, and model construction unit 15 may be implemented using dedicated hardware, namely, the processing circuit 102, while the evaluation value prediction model 16 may be implemented by the processor 103 reading and executing a program stored in the memory 104. As described above, the processing circuit may implement the above functions using hardware, software, firmware, or a combination thereof.

[0071] As described above, the processing condition exploration device 1 according to the first embodiment predicts evaluation values ​​corresponding to untested processing conditions based on the processing conditions, processing status, and processing evaluation values ​​set for the processing machine 2. An evaluation value prediction model 16 is constructed based on the degree of change in the relationship between the processing conditions and the evaluation values, and the evaluation value prediction model 16 is weighted according to the processing status. Even if the relationship between the processing conditions and the evaluation values ​​changes due to a change in the processing status, a new evaluation value prediction model 16 is constructed according to the degree of change, and the evaluation value prediction model 16 is weighted according to the processing status. Thus, even if the relationship between the processing conditions and their evaluation values ​​changes in response to a change in the processing status, the processing condition exploration device 1 can explore processing conditions without having to collect a large amount of data and construct a learning model in advance. Furthermore, the processing condition exploration device 1 does not sequentially explore a large number of processing conditions that can be set for the processing machine 2, but rather narrows the number of processing conditions to be explored in response to changes in the processing status. This reduces the number of trials required to explore processing conditions.

[0072] Furthermore, any constituent elements of the embodiments may be modified or omitted.

[0073] The machining condition search device according to the present invention can be used, for example, to search for machining conditions of an electric discharge machine.

[0074] Description of the label

[0075] 1. Processing condition exploration device, 2. Processing machine, 3. Display unit, 11. Processing condition generation unit, 11a. Processing condition calculation unit, 11b. Actual processing instruction unit, 11c. Exploration completion determination unit, 12. Processing state collection unit, 13. Processing result collection unit, 14. Processing result evaluation unit, 15. Model construction unit, 16. Evaluation value prediction model, 16a. Evaluation value prediction unit, 16b. Unreliability evaluation unit, 17A. Processing state storage unit, 17B. Exploration result storage unit, 17C. Model storage unit, 17D. Prediction result storage unit, 17E. Unreliability storage unit, 100. Input interface, 101. Output interface, 102. Processing circuit, 103. Processor, 104. Memory.

Claims

1. A processing condition exploration device, characterized in that: have: a processing condition generating unit that generates processing conditions to be set for the processing machine; a processing state collecting unit for collecting processing states indicating states of the processing machine performing processing according to the processing conditions, a workpiece to be processed, and an installation environment of the processing machine; a processing result collecting unit for collecting processing results of processing performed according to the processing conditions; a processing result evaluation unit that calculates an evaluation value of processing performed according to the processing conditions based on the processing results collected by the processing result collection unit; an evaluation value prediction model for predicting the evaluation value corresponding to the untried processing condition based on the processing condition generated by the processing condition generation unit, the processing status collected by the processing status collection unit, and the evaluation value calculated by the processing result evaluation unit; as well as a model construction unit that, when it is determined that the degree of change in the machining state is greater than or equal to a threshold, constructs the evaluation value prediction model and performs weighting corresponding to the machining state on the evaluation value prediction model; The processing condition generating unit generates the processing condition to be tried next using the predicted value of the evaluation value calculated by the evaluation value prediction model. The processing of the processing state collecting unit, the processing result collecting unit, the processing result evaluating unit, the evaluation value prediction model and the model building unit is repeated until the processing condition generating unit determines that the search for the processing conditions is completed.

2. The processing condition exploration device according to claim 1, characterized in that: The evaluation value prediction model calculates an index indicating the unreliability of the prediction of the evaluation value according to the processing conditions and the processing state. The processing condition generating unit generates the processing condition to be tried next based on the predicted value of the evaluation value and an index indicating the unreliability of the prediction of the evaluation value.

3. The processing condition exploration device according to claim 2, characterized in that: The processing condition generating unit uses the predicted value of the evaluation value and an indicator indicating the unreliability of the prediction of the evaluation value to determine whether to end the exploration. If it is determined that the exploration is to end, the unreliability of the prediction of the evaluation value is not taken into consideration, and the optimal processing condition is predicted based only on the predicted value of the evaluation value.

4. The processing condition exploration device according to claim 2, characterized in that: The evaluation value prediction model is a probability model of the processing condition corresponding to the evaluation value, which is constructed by assuming that the evaluation value corresponding to the processing condition is a probability variable, and calculates the predicted value of the evaluation value and an index representing the unreliability of the prediction of the evaluation value.

5. The processing condition exploration device according to claim 1, characterized in that: A display unit is provided for displaying at least one of the processing condition and the evaluation value corresponding to the processing condition, the processing condition and a predicted value of the evaluation value corresponding to the processing condition, or the processing condition as a result of a search.

6. The processing condition exploration device according to claim 1, characterized in that: The model construction unit constructs the evaluation value prediction model by linear interpolation of the plurality of evaluation value prediction models corresponding to different machining states.

7. A processing condition exploration method, characterized in that: The following steps are involved: The processing condition generating unit generates processing conditions to be set for the processing machine; The processing state collecting unit collects processing states indicating states of the processing machine performing processing according to the processing conditions, a workpiece to be processed, and an installation environment of the processing machine; The processing result collecting unit collects the processing results of the processing performed according to the processing conditions; a processing result evaluation unit that calculates an evaluation value of processing performed according to the processing conditions based on the processing results collected by the processing result collection unit; an evaluation value prediction model predicting the evaluation value corresponding to the untried processing condition based on the processing condition generated by the processing condition generation unit, the processing status collected by the processing status collection unit, and the evaluation value calculated by the processing result evaluation unit; as well as The model construction unit constructs the evaluation value prediction model when it is determined that the degree of change in the machining state is greater than or equal to a threshold value, and performs weighting corresponding to the machining state on the evaluation value prediction model. The processing condition generating unit generates the processing condition to be tried next using the predicted value of the evaluation value calculated by the evaluation value prediction model. The processing of the processing state collecting unit, the processing result collecting unit, the processing result evaluating unit, the evaluation value prediction model and the model building unit is repeated until the processing condition generating unit determines that the search for the processing conditions is completed.

Images