PROCESSING CONDITIONS SEARCH DEVICE AND PROCESSING CONDITIONS SEARCH PROCEDURE
The machining condition search apparatus dynamically predicts optimal machining conditions by adjusting to state changes, overcoming inefficiencies in existing systems by constructing and weighting models, thus optimizing machining results without extensive data collection.
Patent Information
- Application Number
- DE112020006553
- Authority / Receiving Office
- DE · DE
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2020-03-17
- Publication Date
- 2024-11-28
- Estimated Expiration
- 2040-03-17
AI Technical Summary
Existing machining systems face challenges in predicting optimal machining conditions due to changes in machining states, requiring extensive data collection and model construction to achieve desired results, which is inefficient and resource-intensive.
A machining condition search apparatus and method that generates, evaluates, and predicts machining conditions using a dynamic evaluation value prediction model, adjusting to changes in machining states without requiring large data collections, by constructing and weighting models based on state changes.
Enables efficient search for optimal machining conditions by predicting evaluation values for untested conditions, reducing the need for extensive data collection and model construction, while adapting to changing machining states.
Smart Images

Figure 00000000_0000_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present disclosure relates to a machining condition search apparatus and a machining condition search method for searching for a machining condition. STATE OF THE ART
[0002] Generally, different machining results are obtained depending on the machining conditions established in a machining device. Even when the same machining condition is established, different machining results are obtained depending on a change in a machining state, which indicates the state of the machining device and the state of a workpiece. Therefore, to obtain a machining result that meets the required machining specifications, it is necessary to search for a suitable machining condition based on the machining state.
[0003] For example, Patent Literature 1 describes a control device that predicts the machining result of a workpiece by a machining device. This control device predicts the influence of a change in a state quantity on the machining result using a learning model that predicts the correlation between a state variable representing a change in the state quantity indicating the machining state and determination data indicating the machining result. LIST OF CITIONSPATENT LITERATURE
[0004] Patent Literature 1: JP 2019-32649 A SUMMARY OF THE INVENTIONTECHNICAL PROBLEM
[0005] When the machining state changes, different machining results are obtained even if the same machining condition is established in the machining device, and the relationship between the machining condition and an evaluation value of machining performed under the machining condition also changes. Consequently, there is a problem that in order to search for a machining condition that meets the required machining specifications using the control device described in Patent Literature 1, it is necessary to collect a large amount of training data in advance and construct a learning model that can accurately predict the influence of a change in the machining state on the machining result.
[0006] The present disclosure addresses the above problem, and its objects are to obtain a machining condition search apparatus and a machining condition search method with which it is possible to search for a machining condition without constructing a learning model by previously collecting a large amount of data. SOLUTION TO THE PROBLEM
[0007] A machining condition search device according to the present disclosure includes: a machining condition generation unit for executing generation of a machining condition to be established in a machining device; a machining state acquisition unit for executing acquisition of a machining state indicating each of a state of the machining device that performs machining according to the machining condition, a state of a workpiece to be machined, and a state of an installation environment of the machining device; a machining result acquisition unit for executing acquisition of a machining result of the machining performed according to the machining condition; a machining result evaluation unit for executing calculation of an evaluation value of the machining condition according to the machining condition based on the machining result,collected by the processing result collection unit; at least one evaluation value prediction model for performing a prediction of the evaluation value corresponding to the untested processing condition based on the processing condition generated by the generation performed by the processing condition generation unit and the evaluation value calculated by the calculation performed by the processing result evaluation unit; and a model construction unit for performing a construction of the at least one evaluation value prediction model in a case where a degree of change in a processing state is equal to or greater than a threshold value,and performing weighting based on the machining state on the at least one evaluation value prediction model. The machining state generation unit generates the next machining state to be tested using a prediction value of the evaluation value obtained by the prediction performed by the at least one evaluation value prediction model, and the machining condition search device repeatedly executes each of the processes performed by the machining state acquisition unit, the machining result acquisition unit, the machining result evaluation unit, the at least one evaluation value prediction model, and the model construction unit until the machining condition generation unit determines to end the search for the machining condition. ADVANTAGEOUS EFFECTS OF THE INVENTION
[0008] According to the present disclosure, the evaluation value corresponding to the untested machining condition is predicted based on the machining condition established in the machining device, the machining state, and the evaluation value of the machining. The evaluation value prediction model is constructed based on a degree of change in the relationship between the machining condition and the evaluation value, and weighting is performed on the evaluation value prediction model based on the machining state. Even if the relationship between the machining condition and the evaluation value changes due to a change in the machining state, the evaluation value prediction model is reconstructed based on the degree of change, and weighting is performed on the evaluation value prediction model based on the machining state.Consequently, the machining condition search device according to the present disclosure can search for a machining condition without constructing a learning model by previously collecting a large amount of data. BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a block diagram illustrating a configuration of a machining condition search device according to a first embodiment. Fig. 2 is a flowchart illustrating a machining condition search method according to the first embodiment. Fig. 3 is a diagram illustrating a relationship between a processing condition, a prediction value of an evaluation value corresponding to the processing condition, and an index indicating the uncertainty of the prediction of the evaluation value. Fig. 4A is a block diagram illustrating a hardware configuration for implementing functions of the machining condition search device according to the first embodiment, and Fig. 4B is a block diagram illustrating a hardware configuration for executing software that implements the functions of the machining condition search device according to the first embodiment. DESCRIPTION OF EMBODIMENTSFirst Embodiment
[0009] Fig. 1 is a block diagram illustrating the configuration of a machining condition search device 1 according to the first embodiment. The machining condition search device 1 searches for an optimal machining condition from a large number of machining conditions that can be established in a machining device 2, and sets the machining condition resulting from the search in the machining device 2. The optimal machining condition is, for example, a machining condition under which a machining result meeting required machining specifications is achieved. Furthermore, a display unit 3 displays the machining condition searched by the machining condition search device 1, etc.For example, the display unit 3 displays a machining condition set in the machining device 2 and an evaluation value of machining performed by the machining device 2 according to the machining condition.
[0010] The machining device 2 is an industrial device that performs machining according to a machining condition, and machines a metal workpiece into a desired shape, for example, by cutting or polishing the workpiece or removing unnecessary portions using electricity or other energy. The workpiece is not limited to metal and may be ceramic, glass, or wood. Examples of the machining device 2 include a laser beam machine, an electric discharge machining machine, a cutting machine, a grinding machine, an electrochemical machine, an ultrasonic machining device, and an electron beam machine. An example in which the machining device 2 is an electric discharge machining machine, specifically, an electric discharge machining machine, will be described below.
[0011] The machining condition is constructed by combining a variety of control parameters used to control the machining device 2. Generally, different machining conditions result in different machining results. Moreover, even if the same machining condition is established in the machining device 2, different machining results may be achieved depending on a change in the machining state, which indicates a state of the machining device 2 performing machining, a state of the machined workpiece, and a state of an installation environment of the machining device 2. That is, in order to obtain an optimal machining result that meets the required machining specifications, it is necessary to search for a suitable machining condition based on the machining state.
[0012] For example, in a case where there are three control parameters that can be adjusted when machining by an electric discharge die-sinking EDM machine, and where the value of each control parameter can be selected on a scale of 1 to 10, 10 3 = 1000 machining conditions formed by a combination of control parameters. For this purpose, the machining condition search device 1 does not sequentially search for a large number of machining conditions that can be set up in the machining device 2, but rather limits the number of machining conditions to be searched for depending on a change in the machining state, thus reducing the number of attempts to search for machining conditions.
[0013] In Fig. 1, the machining condition search device 1 includes a machining condition generation unit 11, a machining state acquisition unit 12, a machining result acquisition unit 13, a machining result evaluation unit 14, a model construction unit 15, and an evaluation value prediction model 16. The machining condition search device 1 also includes a machining state storage unit 17A, a search result storage unit 17B, a model storage unit 17C, a prediction result storage unit 17D, and an uncertainty storage unit 17E. All or some of the storage units 17A to 17E may be provided in an external device provided separately from the machining condition search device 1.
[0014] The machining condition generation unit 11 generates a machining condition and sets the generated machining condition in the machining device 2. The machining condition generation unit 11 includes a machining condition calculation unit 11a, a current machining instruction unit 11b, and a seeker end determination unit 11c.
[0015] The machining condition calculation unit 11a calculates a machining condition to be established in the machining device 2. For example, the machining condition calculation unit 11a selects a combination corresponding to the machining detail from among combinations of a plurality of control parameters of the machining device 2 and possible values of these control parameters, and calculates the machining condition from the selected combination. The control parameters include, for example, laser power, cutting speed, beam magnification, focus position, and gas pressure.
[0016] The current machining instruction unit 11b causes the machining device 2 to execute machining based on the machining condition calculated by the machining condition calculation unit 11a. For example, the current machining instruction unit 11b generates an instruction for operating the machining device 2 based on the machining condition calculated by the machining condition calculation unit 11a and outputs the generated instruction to the machining device 2.
[0017] The searcher determination unit 11c determines whether or not to stop searching for the machining condition based on data stored in the prediction result storage unit 17D or the uncertainty storage unit 17E. If the searcher determination unit 11c determines that it is necessary to additionally search for machining conditions, the machining condition calculation unit 11a generates a machining condition under which machining is to be performed next. If it is determined that it is not necessary to additionally search for machining conditions, the machining condition determination unit 11a determines the machining condition predicted to have the highest evaluation value as an optimal machining condition using the prediction result of the evaluation value stored in the prediction result storage unit 17D.
[0018] The machining state acquisition unit 12 acquires data indicating a machining state of machining performed by the machining device 2. The machining state includes a state of the machining device 2, a state of the machined workpiece, and a state of an installation environment of the machining device 2. For example, the machining state indicates a condition that affects the machining result but cannot be controlled by a user, such as the temperature of the machining device 2, the temperature of the workpiece, and the temperature or humidity in a room that is the installation environment of the machining device 2. The data indicating the machining state indicates a state quantity obtained by quantifying each state indicated by the machining state.The data collected by the machining state acquisition unit 12 is stored in the machining state storage unit 17A.
[0019] The machining result acquisition unit 13 acquires, from the machining device 2, a machining result of the machining performed according to the machining conditions. For example, the machining result acquisition unit 13 acquires machining-related information obtained during or after machining. Examples of machining-related information include sound or light detection data observed during machining, the number of discharge pulses, and the surface condition of the workpiece after machining.
[0020] The machining result evaluation unit 14 calculates an evaluation value of the machining performed by the machining device 2 according to the machining condition based on the machining result acquired by the machining result acquisition unit 13. The evaluation value indicates the quality of the machining and is, for example, a value ranging from 0 to 1. The larger the evaluation value, the better the machining result. The evaluation value is 1 when the best machining is performed, and the evaluation value is 0 when the worst machining is performed. The machining result evaluation unit 14 stores a combination of the machining condition and the evaluation value as a search result in the search result storage unit 17B. The search result storage unit 17B stores the search result.
[0021] The model construction unit 15 constructs the evaluation value prediction model 16 based on a degree of change in the relationship between the machining condition and the evaluation value, and performs weighting based on the machining state on the evaluation value prediction model 16. For example, the model construction unit 15 calculates a degree of change in the relationship between the machining condition and the evaluation value based on the information indicating the machining state stored in the machining state storage unit 17A and the evaluation value stored in the search result storage unit 17B.When the degree of change in the relationship between the machining condition and the evaluation value is equal to or greater than a threshold value, the model construction unit 15 determines that a change in the machining condition has occurred for newly developing the evaluation value prediction model, and develops the evaluation value prediction model.
[0022] Furthermore, the model construction unit 15 performs weighting of the evaluation value prediction model 16 based on the processing state. For example, the prediction values of the evaluation values obtained by a plurality of evaluation value prediction models 16 are integrated using weights corresponding to the magnitudes of state quantities obtained by quantifying the processing states according to the evaluation value prediction models 16. In the integrated prediction value of the evaluation value, the prediction result of the evaluation value prediction model 16 having a larger weight is considered more important.
[0023] The evaluation value prediction model 16 predicts an evaluation value corresponding to an untested machining condition based on the machining condition 11 generated by the machining condition generation unit, the machining condition acquired by the machining condition acquisition unit 12, and the evaluation value calculated by the machining result evaluation unit 14. Fig. 1, the valuation value prediction model 16 comprises a valuation value prediction unit 16a and an uncertainty assessment unit 16b.
[0024] The evaluation value prediction unit 16a predicts an evaluation value corresponding to an untested machining condition (under which machining is not performed) based on the machining condition stored in the search result storage unit 17B and the evaluation value corresponding to this machining condition. Furthermore, the evaluation value prediction unit 16a stores the machining condition and a prediction value of the evaluation value corresponding to this machining condition in the prediction result storage unit 17D. The prediction result storage unit 17D stores an untested machining condition and a prediction value of an evaluation value corresponding to the untested machining condition in association with each other.
[0025] The uncertainty evaluation unit 16b calculates an index indicating the uncertainty of the prediction of the evaluation value by the evaluation value prediction unit 16a. For example, the uncertainty evaluation unit 16b calculates an index indicating the uncertainty of the prediction value of the evaluation value, that is, the probability of deviation from the prediction, using the search result stored in the search result storage unit 17B. The uncertainty evaluation unit 16b stores uncertainty information including the calculated index value and the machining condition in the uncertainty storage unit 17E. The uncertainty storage unit 17E stores the untested machining condition and an index value indicating the uncertainty of a prediction of the evaluation value corresponding to the untested machining condition in association with each other.
[0026] Fig. 2 is a flowchart illustrating the machining condition search method according to the first embodiment, and illustrating a series of processing until an optimal machining condition is found by the machining condition search device 1. When the processing for searching for an optimal machining condition is started, the machining condition generation unit 11 generates an initial machining condition (step ST1). The machining condition calculation unit 11a selects a predetermined number of combinations as initial machining conditions from among all the combinations of control parameters that can be set in the machining device 2. As the method for selecting the initial machining condition, an experimental design method, an optimal design method, and random sampling can be adopted.When the user has an idea of the optimal machining condition based on the records of the last use of the machining device 2, the machining condition determined by the user can be selected as the initial machining condition.
[0027] For example, if three control parameters constitute the machining condition and the value to be set in the machining device 2 is selected from 10 values for each control parameter, the total number of machining condition combinations is 1000. The machining condition calculation unit 11a selects a specific number of machining conditions from 1000 machining conditions. In the following description, the specific number is, for example, five, and the machining condition calculation unit 11a selects five machining conditions from 1000 machining conditions.
[0028] Subsequently, the machining condition generation unit 11 sets the initial machining conditions in the machining device 2 and causes the machining device 2 to perform machining according to the initial machining conditions (step ST2). For example, the machining condition calculation unit 11a selects one initial machining condition from five initial machining conditions and outputs the selected initial machining condition to the current machining instruction unit 11b. The current machining instruction unit 11b generates an instruction for operating the machining device 2 based on the initial machining condition output by the machining condition calculation unit 11a and outputs the generated instruction to the machining device 2. The machining device 2 executes machining according to the initial machining condition.In the following description, processing according to the initial processing condition is also referred to as “initial processing”.
[0029] The machining state acquisition unit 12 acquires data indicating a machining state of the machining performed by the machining device 2 according to the initial machining condition (step ST3). The machining state acquisition unit 12 stores the data indicating the machining state acquired by the machining device 2 in the machining state storage unit 17A. As a result, the machining state storage unit 17A stores the initial machining condition and the machining state in association with each other.
[0030] The machining result acquisition unit 13 acquires data indicating a machining result of machining performed by the machining device 2 according to the initial machining condition (step ST4). The data acquired by the machining result acquisition unit 13 is output to the machining result evaluation unit 14.
[0031] The machining result evaluation unit 14 calculates an evaluation value of the machining performed by the machining device 2 according to the initial machining condition based on the machining result acquired by the machining result acquisition unit 13 (step ST5). The machining result evaluation unit 14 calculates an evaluation value indicating the quality of the machining by quantifying the machining result acquired by the machining result acquisition unit 13.For example, the machining result evaluation unit 14 measures detection data of sound or light or the number of discharge pulses observed during machining according to the initial machining condition, and converts the measured value into an evaluation value indicating the quality of the machining. The evaluation value is a continuous value or a discrete value of a plurality of levels (e.g., 10 levels). The search result storage unit 17B stores a combination of the machining condition and the evaluation value.
[0032] The machining condition calculation unit 11a checks whether or not the initial machining has been completed for all the machining conditions selected as the initial machining conditions (step ST6). If there is an initial machining condition under which initial machining has not been completed (NO in step ST6), the processes from step ST1 to step ST5 are sequentially executed for the initial machining condition under which the initial machining has not been completed. As a result, all the initial machining conditions (for example, five initial machining conditions) and the data indicating machining states are stored in association with each other in the machining state storage unit 17A. Further, the search result storage unit 17B stores all the initial machining conditions and the evaluation values in association with each other.
[0033] If the initial machining has been completed under all initial machining conditions (YES in step ST6), the machining condition calculation unit 11a checks whether the machining has been completed according to the optimal machining condition resulting from the search (step ST7). If the machining device 2 completes the machining according to the optimal machining condition (YES in step ST7), the series of processes in Fig. 2.
[0034] If machining has not been completed according to the optimal machining condition (NO in step ST7), the model construction unit 15 determines, using the data indicating the machining condition stored in the machining condition storage unit 17A, whether the value of an index indicating a degree of change in the relationship between the machining condition and the evaluation value is equal to or greater than the threshold value (step ST8). A method for obtaining the degree of change includes change point detection. The change point detection uses an abnormality degree as an index indicating the degree of change. The abnormality degree is, for example, a value obtained by squaring a difference between a forecast value of the machining condition and a current measured value of the machining condition stored in the machining condition storage unit 17A.
[0035] For example, an autoregressive model is used to calculate the prediction value of the machining state. In the autoregressive model, the prediction value at y(t) of the machining state at a time t one step ahead is calculated using the following equation (1). In the following equation (1), y(t - 1) is a current measurement value of the machining state at a current time t - 1, y(t - 2) is a current measurement value of the machining state at a previous time t - 2, and y(t - 3) is a current measurement value of the machining state at a previous time t - 3. The coefficient α1 is a coefficient for the current measurement value y(t - 1), the coefficient α2 is a coefficient for the current measurement value y(t - 2), and the coefficient α3 is a coefficient for the current measurement value y(t - 3).Note that the model construction unit 15 may evaluate the degree of change in the relationship between the processing state and the evaluation value using supervised learning such as a multiple regression model, a decision tree, or a neural network instead of the autoregressive model. y^(t)=α1y(t−1)+α2y(t−2)+α3y(t−3)
[0036] If it is determined that the value of the index indicating the degree of change in the relationship between the machining condition and the evaluation value is equal to or greater than the threshold value (YES in step ST8), the model construction unit 15 determines that a change has occurred in the machining condition for which the evaluation value prediction model 16 is to be reconstructed, and constructs the evaluation value prediction model 16 according to the changed machining condition (step ST9). The evaluation value prediction model 16 can be constructed using supervised learning, for example, an autoregressive model, a multiple regression model, a decision tree, or a neural network.
[0037] When the model construction unit 15 determines that the value of the index indicating the degree of change in the relationship between the machining condition and the evaluation value is smaller than the threshold value (NO in step ST8), or when the process of ST9 is completed, the model construction unit 15 performs weighting for each of the already constructed evaluation value prediction models 16 based on the machining state stored in the machining state storage unit 17A (step ST10). The weighting method includes linear interpolation and gain. Furthermore, the model construction unit 15 may select one evaluation value prediction model 16 and perform weighting for the selected evaluation value prediction model 16, instead of performing weighting for the plurality of evaluation value prediction models 16.
[0038] For example, the model construction unit 15 performs weighting of the evaluation value prediction model 16 using linear interpolation. Using a machining state s1 at the previous time t-2 and a machining state s2 at the previous time t-3 stored in the machining state storage unit 17A, and a machining state s at the current time t-1 acquired by the machining state acquisition unit 12, the model construction unit 15 calculates a weight β1(s) for a function f1 determined by the evaluation value prediction model 16 corresponding to the machining state s1, and a weight β2(s) for a function f2 determined by the evaluation value prediction model 16 corresponding to the machining state s2 according to the following equation (2).An output f(x, s) of the evaluation value prediction model 16 for the machining condition x and the machining state s is calculated using the weights according to the following equation (3). Note that although the weight values are determined for the two evaluation value prediction models 16, the model construction unit 15 can calculate the weight values for three or more evaluation value prediction models 16. β1(s)=(s2−s) / (s2−s1)β2(s)=(s−s1) / (s2−s1) f(x,s)=β1(s)f1(x)+β2(s)f2(x)
[0039] When machining is completed according to, for example, the five initial machining conditions, the evaluation value prediction unit 16a predicts evaluation values in the evaluation value prediction model 16 using the machining conditions stored in the search result storage unit 17B and the corresponding evaluation values for all of the 1000 machining conditions described above (step ST11). The evaluation values predicted by the evaluation value prediction unit 16a are stored in the prediction result storage unit 17D (step ST12).
[0040] As a method for predicting the evaluation value by the evaluation value prediction unit 16a, for example, Gaussian process regression can be used. In Gaussian process regression, the evaluation value prediction unit 16a is a probabilistic model of the machining condition corresponding to the evaluation value, which is constructed under the assumption that the evaluation value corresponding to the machining condition is a random variable. When a vector in which the machining conditions stored in the search result storage unit 17B and their corresponding kernel values are arranged is k, and a vector in which the evaluation values stored in the search result storage unit 17B are arranged is t, the evaluation value prediction unit 16a can calculate a prediction value m(x) of the evaluation value for the machining condition x according to the following equation (4). In equation (4), C is Na Gram matrix. Note that the rating value prediction unit 16a may not use Gaussian process regression and may predict the rating value using supervised learning, such as a decision tree, linear regression, boosting, or a neural network. m(x)=kT⋅(CN−1)⋅t
[0041] For example, when machining is completed according to the five initial machining conditions, the uncertainty evaluation unit 16b calculates, in the evaluation value prediction model 16, indexes expressing the uncertainty of the prediction of the evaluation values for all of the above-described 1000 machining conditions using the machining conditions stored in the search result storage unit 17B and the evaluation values corresponding to them (step ST13). The values of the indexes calculated by the uncertainty evaluation unit 16b are stored in the uncertainty storage unit 17E (step ST14).
[0042] As a method for calculating the indices by the uncertainty evaluation unit 16b, for example, Gaussian process regression can be used. In the Gaussian process regression, when a vector in which the processing conditions stored in the search result storage unit 17B and values of kernels corresponding to them are arranged is k, and a scalar value obtained by adding the accuracy parameter of the evaluation value prediction unit 16a to the value of the kernel between the processing conditions x is c, the uncertainty evaluation unit 16b can calculate an index σ 2 (x), which indicates the uncertainty of the prediction of the evaluation value for the untested machining condition according to the following equation (5). In equation (5), C Na Gram matrix. The uncertainty assessment unit 16b may not use Gaussian process regression, but may calculate the index by regression using density estimation, a mixed density network, and Kullback-Leibler divergence. σ2(x)=c−kT⋅(CN−1)k
[0043] Fig. Figure 3 is a diagram illustrating a relationship between a processing condition, a prediction value of an evaluation value corresponding to the processing condition, and an index indicating the uncertainty of the prediction of the evaluation value. The evaluation value prediction model 16 performs a prediction based on the premise that, for example, a Gaussian process regression is used and the evaluation value follows a Gaussian distribution. Fig. 3 illustrates black dots indicate processing conditions and evaluation values stored in the search result storage unit 17B. Fig. 3 statistically indicates that in a case where the prediction value of the evaluation value is set as an average m(x) of the Gaussian distribution and the index indicating the uncertainty of the prediction of the evaluation value is set as a standard deviation σ(x) of the Gaussian distribution, the black points fall within a range of equal to or less than m(x) + 2σ(x) and equal to or greater than m(x) - 2σ(x) with a probability of about 95% even if the prediction of the evaluation value is wrong.
[0044] The search end determination unit 11c included in the machining condition generation unit 11 determines whether or not to terminate the search for the machining condition using the prediction value of the machining condition stored in the prediction result storage unit 17D and the index indicating the uncertainty of the prediction of the evaluation value stored in the uncertainty storage unit 17E (step ST15). For example, the search end determination unit 11c compares the values of the indexes stored in the uncertainty storage unit 17E and indicates the uncertainty of the prediction of the evaluation values of all the machining conditions searched so far with a threshold value. In a case where the values of the indexes are equal to or less than the threshold value, the search end determination unit 11c determines that the optimal machining condition has been searched and terminates the search for the machining condition.
[0045] Furthermore, if the number of processing conditions under which the prediction of the evaluation values may differ significantly is equal to or less than a specified number as a result of a comparison between the indexes indicating the uncertainty of the prediction of the evaluation values of all processing conditions and the threshold value, the search end determination unit 11c stops the search for the processing condition. This is because if a processing condition exists under which the current measured value of the evaluation value and the prediction value of the evaluation value differ significantly, the prediction value changes significantly in a wide range of the search space of the processing condition.
[0046] For example, the searcher determination unit 11c uses the processing condition x, the prediction value m(x) of the evaluation value for the processing condition x, and an index σ(x) (standard deviation) indicating the uncertainty of the prediction of the evaluation value, and can determine that as the value of m(x) + κσ(x) increases, the importance of searching for the processing condition increases. Note that κ is a parameter determined before starting the search for the processing condition. When the value of κ is smaller, a processing condition with a higher prediction value of the evaluation value is selected, and when the value of κ is larger, a processing condition with a higher probability of a large deviation from the prediction of the evaluation value is selected. The value of κ can be continuously used with the same value, or the value can be changed during processing.
[0047] If it is determined that the search for the machining condition is terminated (YES in step ST15), the searcher determination unit 11c extracts the machining condition predicted to have the highest evaluation value from the prediction values of the evaluation values of all the machining conditions stored in the prediction result storage unit 17D, and outputs the extracted machining condition to the current machining instruction unit 11b. The current machining instruction unit 11b outputs an instruction including the machining condition output by the searcher determination unit 11c to the machining device 2 and sets the machining condition in the machining device 2 (step ST16).
[0048] If it is determined that it is necessary to additionally search for a machining condition (NO in step ST15), the search end determination unit 11c outputs the continuation of the search to the machining condition calculation unit 11a. When the machining condition calculation unit 11a receives an instruction to continue the search from the search end determination unit 11c, it generates a next machining condition to be tested by using the prediction value of the evaluation value of the machining condition stored in the prediction result storage unit 17D (step ST17). The machining condition to be tested next, calculated by the machining condition calculation unit 11a, is output to the current machining instruction unit 11b.The current machining instruction unit 11b outputs an instruction including the next machining condition to be tested to the machining device 2 and sets the machining condition in the machining device 2.
[0049] When the optimal machining condition is established in step ST16 or the next machining condition to be tested is established in step ST17, the machining device 2 executes machining (step ST18). During machining by the machining device 2, the machining condition acquisition unit 12 acquires data indicating the machining condition and stores the machining condition and the machining condition in association with each other in the machining condition storage unit 17A. The machining result acquisition unit 13 acquires data indicating the machining result and outputs the data to the machining result evaluation unit 14. The machining result evaluation unit 14 calculates an evaluation value of the machining performed by the machining device 2 based on the machining result acquired by the machining result acquisition unit 13 (step ST19).Then, the processing proceeds to the process of step ST7, and the above-described processing is executed.
[0050] The display unit 3 displays the machining condition and the evaluation value corresponding to the machining condition obtained during the machining condition search by the machining condition search device 1. Furthermore, the display unit 3 displays the machining condition and the prediction value of the evaluation value corresponding to the machining condition or the optimal machining condition that is the search result. That is, the display unit 3 displays at least one of: the machining condition read from the search result storage unit 17B and the evaluation value corresponding to this machining condition; the machining condition read from the prediction result storage unit 17D and the prediction value of the evaluation value corresponding to this machining condition; or the optimal machining condition that is the search result output from the machining condition calculation unit 11a.As a result, a machining worker can recognize the search situation and the search result of the machining condition by referring to the information displayed on the display unit 3.
[0051] A hardware configuration for implementing the functions of the machining condition search device 1 is as follows.
[0052] The functions of the machining condition generation unit 11, the machining state acquisition unit 12, the machining result acquisition unit 13, the machining result evaluation unit 14, the model construction unit 15, and the evaluation value prediction model 16 in the machining condition search device 1 are implemented by a processing circuit. That is, the machining condition search device 1 includes a processing circuit that performs processes from step ST1 to step ST19 in Fig. 2. The processing circuitry may be dedicated hardware or a central processing unit (CPU) that executes a program stored in memory.
[0053] Fig. 4A is a block diagram illustrating a hardware configuration for implementing the function of the machining condition search device 1. Fig. 4B is a block diagram illustrating a hardware configuration for executing software that implements the function of the machining condition search device 1. Fig. 4A and Fig. 4B, an input interface 100 transmits data indicating the machining state and the machining result output from the machining device 2 to the machining condition search device 1, and transmits storage data output from each of the storage units 17A to 17E to the machining condition search device 1. An output interface 101 transmits 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 each of the storage units 17A to 17E.
[0054] If the processing circuit is a processing circuit 102 which is dedicated hardware, as in Fig. As illustrated in FIG. 4A, the processing circuit 102 is, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination of some of these circuits. The functions of the machining condition generation unit 11, the machining state acquisition unit 12, the machining result acquisition unit 13, the machining result evaluation unit 14, the model construction unit 15, and the evaluation value prediction model 16 in the machining condition search device 1 may be implemented by separate processing circuits, or these functions may be implemented collectively by a single processing circuit.
[0055] If the processing circuit is in Fig. 4B, the functions of the machining condition generation unit 11, the machining state acquisition unit 12, the machining result acquisition unit 13, the machining result evaluation unit 14, the model construction unit 15, and the evaluation value prediction model 16 in the machining condition search device 1 are implemented by software, firmware, or a combination of software and firmware. Note that software or firmware is described as a program and stored in a memory 104.
[0056] The processor 103 executes the functions of the machining condition generation unit 11, the machining state acquisition unit 12, the machining result acquisition unit 13, the machining result evaluation unit 14, the model construction unit 15, and the evaluation value prediction model 16 in the machining condition search device 1 by reading and executing the program stored in the memory 104. For example, the machining condition search device 1 includes the memory 104 for storing programs for ultimately executing the processes from step ST1 to step ST19 in the Fig.2 when the processing in the flowchart is executed by the processor 103. These programs cause a computer to execute operations or methods executed by the machining condition generation unit 11, the machining state acquisition unit 12, the machining result acquisition unit 13, the machining 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 in which a program that causes a computer to function as the machining condition generation unit 11, the machining state acquisition unit 12, the machining result acquisition unit 13, the machining result evaluation unit 14, the model construction unit 15, and the evaluation value prediction model 16 is stored.
[0057] The memory 104 is, for example, a non-volatile or volatile semiconductor memory, such as a random access memory (RAM), a read-only memory (ROM), a flash memory, an erasable programmable ROM (EPROM) or an electrically erasable programmable ROM (EPROM), a magnetic disk, a flexible disk, an optical disk, a compact disc, a minidisk, or a digital versatile disc (DVD).
[0058] Some of the functions of the machining condition generation unit 11, the machining state acquisition unit 12, the machining result acquisition unit 13, the machining result evaluation unit 14, the model construction unit 15, and the evaluation value prediction model 16 in the machining condition search device 1 may be implemented by dedicated hardware, and some may be implemented by software or firmware. For example, the functions of the machining condition generation unit 11, the machining state acquisition unit 12, the machining result acquisition unit 13, the machining result evaluation unit 14, and the model construction unit 15 are implemented by the processing circuit 102, which is dedicated hardware, and the function of the evaluation value prediction model 16 is implemented by the processor 103, which reads and executes the program stored in the memory 104.As described above, the processing circuitry may implement the above-mentioned functions through hardware, software, firmware, or a combination thereof.
[0059] As described above, in the machining condition search device 1 according to the first embodiment, the evaluation value corresponding to the untested machining condition is predicted based on the machining condition established in the machining device 2, the machining state, and the evaluation value of the machining, the evaluation value prediction model 16 is constructed based on a degree of change in the relationship between the machining condition and the evaluation value, and weighting based on the machining state is performed on the evaluation value prediction model 16.Even if the relationship between the machining condition and the evaluation value changes due to a change in the machining state, the evaluation value prediction model 16 is reconstructed based on the degree of change, and weighting is performed on the evaluation value prediction model 16 based on the machining state. Thus, the machining condition search device 1 can search for the machining condition by collecting a large amount of data in advance without constructing a learning model, even if the relationship between the machining condition and its evaluation value changes due to a change in the machining state.In addition, the machining condition search device 1 does not sequentially search for a large number of machining conditions that can be set in the machining device 2, but limits the number of machining conditions to be searched depending on a change in the machining state, so that the number of attempts to search for machining conditions is reduced.
[0060] It should be noted that any components in the embodiment may be modified or omitted. INDUSTRIAL APPLICABILITY
[0061] The machining condition search device according to the present disclosure can be used, for example, for searching machining conditions of an electric discharge machining machine. LIST OF REFERENCE SYMBOLS
[0062] 1: Machining condition search unit, 2: Machining device, 3: Display unit, 11: Machining condition generation unit, 11a: Machining condition calculation unit, 11b: Current machining instruction unit, 11c: Search end determination unit, 12: Machining state acquisition unit, 13: Machining result acquisition unit, 14: Machining result evaluation unit, 15: Model construction unit, 16: Evaluation value prediction model, 16a: Evaluation value prediction unit, 16b: Uncertainty evaluation unit, 17A: Machining state storage unit, 17B: Search result storage unit, 17C: Model storage unit, 17D: Prediction result storage unit, 17E: Uncertainty storage unit, 100: Input interface, 101: Output interface, 102: Processing circuit, 103: Processor, 104: Memory
Claims
[1] Machining condition search device (1), comprising: a machining condition generation unit (11) for executing generation of a machining condition to be set in a machining device (2); a machining state acquisition unit (12) for executing a machining state acquisition indicating each of a state of the machining device (2) that executes machining according to the machining condition, a state of a workpiece to be machined, and a state of an installation environment of the machining device (2); a machining result acquisition unit (13) for executing an acquisition of a machining result of the machining performed according to the machining condition; a machining result evaluation unit (14) for executing a calculation of an evaluation value of the machining performed according to the machining condition on the basis of the machining result obtained by the survey performed by the machining result survey unit (13); at least one evaluation value prediction model (16) for performing a prediction of the evaluation value corresponding to the machining condition that has not been tested based on the machining condition generated by the generation performed by the machining condition generation unit (11), wherein the machining state was acquired by the acquisition performed by the machining state acquisition unit (12) and the evaluation value was calculated by the calculation performed by the machining result evaluation unit (14); and a model construction unit (15) for carrying out a construction of the at least one evaluation value prediction model (16) in a case where a degree of change in a processing state is equal to or greater than a threshold value, and carrying out a weighting based on the processing state on the at least one evaluation value prediction model, wherein the machining condition generation unit (11) generates the next machining condition to be tested using a prediction value of the evaluation value obtained by the prediction executed by the at least one evaluation value prediction model (16), and the machining condition search device (1) repeatedly executes each of the processes executed by the machining state acquisition unit (12), the machining result acquisition unit (13), the machining result evaluation unit (14), the at least one evaluation value prediction model (16), and the model construction unit (15) until the machining condition generation unit (11) determines to end the search for the machining condition. [2] Machining condition search device (1) according to claim 1, wherein the at least one evaluation value prediction model (16) calculates an index indicating an uncertainty of the prediction of the evaluation value based on the processing condition and the processing state, and the machining condition generation unit (11) generates the next machining condition to be tested based on the prediction value of the evaluation value and the index indicating the uncertainty of the prediction of the evaluation value. [3] The machining condition search device (1) according to claim 2, wherein the machining condition generation unit (11) determines whether or not to stop the search using the prediction value of the evaluation value and the index indicating the uncertainty of the prediction of the evaluation value, and when determining to stop the search, predicts the machining condition that is optimal based on only the prediction value of the evaluation value without considering the uncertainty of the prediction of the evaluation value. [4] The machining condition search device (1) according to claim 2, wherein the at least one evaluation value prediction model (16) is a probabilistic model of the machining condition corresponding to the evaluation value constructed on the assumption that the evaluation value corresponding to the machining condition is a random variable, and the at least one evaluation value prediction model calculates the prediction value of the evaluation value and the index indicating the uncertainty of the prediction of the evaluation value. [5] The machining condition search device (1) according to claim 1, further comprising a display unit (3) that displays 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 that is a result of the search. [6] The machining condition search device (1) according to claim 1, wherein the model construction unit (15) constructs an evaluation value prediction model by performing linear interpolation on a plurality of evaluation value prediction models included in the at least one evaluation value prediction model. [7] Machining condition search method, comprising: a step of executing a generation of a machining condition to be set in a machining device (2) by a machining condition generation unit (11); a step of performing a machining state acquisition indicating each of a state of the machining device (2) performing machining according to the machining condition, a state of a workpiece to be machined, and a state of an installation environment of the machining device (2) by a machining state acquisition unit (12); a step of executing a collection of a processing result of the processing executed according to the processing condition by a processing result collection unit (13); a step of executing, by a machining result evaluation unit (14), a calculation of an evaluation value of the machining performed according to the machining condition based on the machining result obtained by the survey executed by the machining result survey unit (13); a step of performing a prediction of the evaluation value corresponding to the machining condition that has not been tested based on the machining condition generated by the generation performed by the machining condition generation unit (11), wherein the machining state was acquired by the acquisition performed by the machining state acquisition unit (12) and the evaluation value was calculated by the calculation performed by the machining result evaluation unit (14), by at least one evaluation value prediction model (16); and a step of executing a construction of the at least one evaluation value prediction model (16) in a case where a degree of change in a processing state is equal to or greater than a threshold value, and of executing a weighting based on the processing state on the at least one evaluation value prediction model by a model construction unit (15), wherein the machining condition generation unit (11) generates the next machining condition to be tested using a prediction value of the evaluation value obtained by the prediction executed by the at least one evaluation value prediction model (16), and each of the steps executed by the machining state acquisition unit (12), the machining result acquisition unit (13), the machining result evaluation unit (14), the at least one evaluation value prediction model (16) and the model construction unit (15) is repeatedly executed until the machining condition generation unit (11) determines to end the search for the machining condition.
Citation Information
Patent Citations
Control device and machine learning device
JP2019032649A
Machine learning device, electric discharge machine, and machine learning method
WO2019202672A1
JP002019032649A
