JP7204952B1 - MACHINING CONDITION SEARCH DEVICE AND MACHINING CONDITION SEARCH METHOD - Google Patents

Abstract

A machining condition search device (10) learns machining conditions for electric discharge machining by an electric discharge machine (28), and based on the learning result, searches for machining conditions to be output to the electric discharge machine (28). and a machine learning parameter representing a control parameter to be searched by the machine learning part (19) and a range of values of the control parameter to be searched among the control parameters constituting the machining conditions are discharged. and a machine learning parameter determination unit (15) that determines the mode of machining based on the specified machining parameters.

Description

The present disclosure relates to a machining condition searching device and a machining condition searching method for searching for machining conditions for electric discharge machining.

An electrical discharge machine that performs die-sinking electrical discharge machining selects appropriate machining conditions based on machining parameters such as the material to be machined, the shape of the electrode, and the machining depth, and performs electrical discharge machining. It is difficult for the electric discharge machine to perform good machining unless the machining conditions are changed in accordance with changes in the machining state as the machining progresses. In addition, when the processing speed is high or when high processing accuracy is required, slight changes in the processing conditions such as changes in the state of the processing fluid or differences in the quality of the processing material for each lot of processing materials. Even small changes can change the state of the process. For this reason, the electric discharge machine needs not only to select machining conditions from preset machining conditions, but also to adjust the machining conditions according to the progress of machining or changes in the circumstances in which machining is performed.

Patent Literature 1 discloses an electrical discharge machine that observes the state of electrical discharge machining during electrical discharge machining and adjusts the machining conditions based on the observation results. The electric discharge machine according to Patent Document 1 observes the state of electric discharge machining while changing the values of the control parameters that constitute the machining conditions, and optimizes the values of the control parameters.

WO2019/202672

When adjusting the machining conditions for electric discharge machining, it is necessary to adjust the values of the control parameters for each of the plurality of control parameters. It is not easy to search for the optimum combination among all combinations of values of each control parameter for all possible values of each of the plurality of control parameters. The technique of Patent Document 1 has a problem that it is difficult to efficiently search for machining conditions because the optimal combination is searched from a huge number of combinations of values of each control parameter.

The present disclosure has been made in view of the above, and an object of the present disclosure is to obtain a machining condition searching device that enables efficient searching of machining conditions for electrical discharge machining.

In order to solve the above-described problems and achieve the object, the machining condition search device according to the present disclosure learns the machining conditions for electric discharge machining by the electric discharge machine, and based on the learning result, sends the machining conditions to the electric discharge machine. A machine learning unit that searches for a machining condition to be output, a control parameter that is part of a plurality of control parameters that make up the machining condition and is a search target, and a part of the value range of the control parameter that is a search target. and a machine learning parameter determination unit that determines a machine learning parameter indicating a target range based on a machine learning parameter determination rule . In the decision rule, among the control parameters set in advance, control parameters to be searched, control parameters to be excluded from the search, and data about values to be searched among the values of the control parameters to be searched are set in association with the machining specifications of electrical discharge machining. The machine learning unit searches for machining conditions by limiting the control parameters and the range of values of the control parameters based on the machine learning parameters.

The machining condition searching device according to the present disclosure has the effect of enabling efficient searching of machining conditions for electrical discharge machining.

1 is a diagram showing the configuration of a machining condition search device according to a first embodiment; FIG. 4 is a flow chart showing the procedure of processing executed by the machining condition search device according to the first embodiment; FIG. 4 is a diagram showing an example of a decision rule used by the machining condition search device according to the first embodiment to decide machine learning parameters; FIG. 4 is a diagram showing an example of a search range for machining conditions in the machining condition search device according to the first embodiment; 4 is a flow chart showing the procedure of learning by the learning unit of the machining condition search device according to the first embodiment; FIG. 2 is a diagram showing a configuration example of hardware of the processing condition search device according to the first embodiment;

A processing condition searching device and a processing condition searching method according to embodiments will be described below in detail with reference to the drawings.

Embodiment 1.
FIG. 1 is a diagram showing the configuration of a machining condition search device 10 according to Embodiment 1. As shown in FIG. The machining condition searching device 10 searches machining conditions for electric discharge machining by the electric discharge machine 28 . The machining condition search device 10 searches for an optimum combination of values of each control parameter for a plurality of control parameters constituting machining conditions.

The machining condition search device 10 learns the machining conditions by machine learning during machining by the electrical discharge machine 28, and searches for the machining conditions based on the learning results. The machining condition search device 10 learns machining conditions according to machine learning parameters described later.

The electrical discharge machine 28 is a machine that performs die-sinking electrical discharge machining. The electric discharge machine 28 has an electrode, a drive device for controlling the distance between the electrode and the workpiece, and a machining power source for generating electric discharge between the electrode and the workpiece. Illustration of the configuration of the electric discharge machine 28 is omitted.

The machining condition search device 10 includes a machining specification input unit 11 to which information on machining specifications is inputted, a machining condition creation unit 12 to create reference machining conditions, and a machine learning setting input unit to which machine learning setting information is entered. 13, a determination rule storage unit 14 that stores determination rules for machine learning parameters, a machine learning parameter determination unit 15 that determines machine learning parameters, a machine learning unit 19, and a processing condition adjustment that stores processing condition adjustment results. It has a result storage unit 20 and a decision rule update unit 21 for updating decision rules for machine learning parameters.

Machining specifications represent aspects of electrical discharge machining, such as machining material, machining geometry, machining depth, or critical performance. The work material is the material of the work piece. The machining shape is the shape of the electrode to be transferred to the workpiece. The machining depth is the depth of the part to be machined in the workpiece. The important performance is performance that is considered important in processing, such as processing speed or processing stability. The user of the processing condition search device 10 specifies the processing parameters by inputting information on the processing parameters into the processing parameter input unit 11 . Information on the machining specifications is input from the machining specifications input unit 11 to each of the machining condition creation unit 12 and the machine learning parameter determination unit 15 .

The reference machining conditions are machining conditions composed of initial values of control parameters, and represent initial conditions when searching is started. The machining condition creation unit 12 creates standard machining conditions based on the machining specifications. A plurality of representative machining conditions are stored in advance in the machining condition creation unit 12 . The machining condition creation unit 12 creates reference machining conditions by selecting machining conditions corresponding to the machining specifications input to the machining specification input unit 11 from among a plurality of saved machining conditions. The reference machining conditions are input from the machining condition creation unit 12 to each of the machine learning parameter determination unit 15 and the machine learning unit 19 .

The machine learning setting information is information indicating the type of machining to be searched for machining conditions by machine learning. The machining type is the type of machining targeted for search of machining conditions by machine learning, and is the type of machining such as rough machining or finishing machining. A user specifies a desired processing type by inputting machine learning setting information to the machine learning setting input unit 13 . Information of the machine learning setting information is input from the machine learning setting input unit 13 to the machine learning parameter determination unit 15 .

Further, the machine learning setting information input to the machine learning setting input unit 13 may include user designation information for the user to designate a search target or a search range. The user can specify the type of control parameter to be searched or the range of values of the control parameter to be searched by user-specified information. For example, if know-how has already been established for part of the processing conditions, the user can set arbitrary restrictions on the search range of the processing conditions by inputting user-designated information according to the know-how. For example, if there is a control parameter that the user does not want to change, the user can specify that the control parameter is excluded from search targets by entering machine learning setting information. The user may specify a combination of control parameters to be searched for using user-specified information. Input of user-specified information is optional.

The machine learning parameters include information indicating the types of control parameters to be searched by the machine learning unit 19 among the control parameters constituting the machining conditions, and information indicating the range of values of the control parameters to be searched. . That is, the machine learning parameters include information for defining a search range for machining conditions by machine learning. In addition, the machine learning parameters include information on calculation formulas for calculating evaluation points for machining conditions. Evaluation points are included in learning data in machine learning.

The machine learning parameter determination unit 15 includes a search parameter determination unit 16 that determines control parameters to be searched by the machine learning unit 19, and a search range determination unit 17 that determines the range of values of the control parameters to be searched. , and an evaluation point calculation formula determination unit 18 that determines the evaluation point calculation formula.

The determination rule storage unit 14 holds information on determination rules for determining machine learning parameters based on processing specifications. The determination rule storage unit 14 stores information on predetermined determination rules. The stored determination rule is updated by a determination rule updating unit 21, which will be described later.

The search parameter determination unit 16 determines control parameters to be searched based on the processing specifications, the determination rule, and the processing type indicated in the machine learning setting information. When user-designated information about the type of control parameter is included in the machine learning setting information, the search parameter determining unit 16 further determines control parameters to be searched based on the user-designated information.

The search range determination unit 17 determines a range of control parameter values to be searched based on the processing specifications, the determination rule, and the processing type indicated in the machine learning setting information. Further, the search range determination unit 17 adjusts the range of values of the control parameters based on the standard machining conditions. If the machine learning setting information includes user-specified information about the value range of the control parameter, the search range determining unit 17 further determines the value range of the control parameter based on the user-specified information.

The evaluation point calculation formula determination unit 18 determines the evaluation point calculation formula based on the processing specifications, the determination rule, and the processing type indicated in the machine learning setting information. Specifically, the evaluation point calculation formula determination unit 18 determines values of variables used in the calculation formula.

Thus, the machine-learning parameter determination unit 15 determines the machine-learning parameters based on the machining specifications, the determination rule, the machine-learning setting information, and the standard machining conditions. The machine learning parameter determination unit 15 outputs machine learning parameters including information indicating the type of parameter to be searched, information indicating the range of values of the parameter to be searched, and information on the determined calculation formula.

The machine learning unit 19 learns machining conditions for electric discharge machining by the electric discharge machine 28, and searches for machining conditions to be output to the electric discharge machine 28 based on the learning result. The machine learning unit 19 receives input of the reference machining conditions, machine learning parameters, and state data. The state data is data indicating the state of electric discharge machining in the electric discharge machine 28 . The state data includes processing depth data, processing speed data, the number of discharge occurrences, pulse state data, and the like. The pulse state data is data indicating whether the state of the current pulse or voltage pulse is good or bad. State data is input from the electric discharge machine 28 to the machine learning unit 19 .

The machine learning section 19 has a learning section 22 , a state observation section 23 , a machining condition adjustment section 24 and an evaluation point calculation section 25 . The state observation unit 23 obtains information indicating the type of control parameter to be searched among the machine learning parameters, information indicating the range of values of the control parameter to be searched, and information about the state of electrical discharge machining. A certain evaluation point is observed as a state variable. The learning unit 22 learns the optimum combination of control parameter values according to the data set created based on the state variables. The learning unit 22 has a reward calculator 26 and a function updater 27 . The reward calculator 26 and the function updater 27 will be described later.

The machining condition adjustment unit 24 adjusts machining conditions to be output to the electric discharge machine 28 based on the learning result of the learning unit 22 . The evaluation point calculator 25 calculates the evaluation points for the machining conditions output to the electric discharge machine 28 based on the state data.

The machining condition search device 10 outputs the machining conditions adjusted by the machining condition adjustment unit 24 to the electric discharge machine 28 . The electric discharge machine 28 processes the workpiece according to the machining conditions input from the machining condition searching device 10 .

The machining conditions adjusted by the machining condition adjustment unit 24 are input to the machining condition adjustment result storage unit 20 . The machining condition adjustment result storage unit 20 holds the machining conditions adjusted by the machining condition adjustment unit 24 .

Although FIG. 1 shows the machining condition searching device 10 as a device external to the electric discharge machine 28, the machining condition searching device 10 may be built in the electric discharge machine 28. FIG. The machining condition search device 10 may be built in a numerical controller that controls the electric discharge machine 28 .

The decision rule updating unit 21 reads the processing conditions from the processing condition adjustment result storage unit 20 . The determination rule updating unit 21 updates the determination rule of the machine learning parameter based on the adjustment result of the machining condition by the machining condition adjustment result storage unit 20 .

Next, processing executed by the processing condition search device 10 will be described. FIG. 2 is a flow chart showing the procedure of processing executed by the processing condition search device 10 according to the first embodiment.

The user inputs information on the processing specifications to the processing specifications input unit 11 . Also, the user inputs machine learning setting information to the machine learning setting input unit 13 . In step S1, the machining condition search device 10 receives machining specification information and machine learning setting information. The machine learning setting information includes information indicating the processing type. Alternatively, the machine learning setting information includes information indicating the processing type and user specified information.

In step S<b>2 , the machining condition generating unit 12 of the machining condition search device 10 creates standard machining conditions corresponding to the machining parameters. The machining condition creation unit 12 creates reference machining conditions by selecting machining conditions corresponding to the machining specifications input to the machining specification input unit 11 from a plurality of machining conditions stored in advance. .

In step S3, the machine learning parameter determination unit 15 of the processing condition search device 10 determines the machine learning parameters based on the processing specifications, the machine learning parameter determination rule, the machine learning setting information, and the reference processing conditions. decide. The machine learning parameter determination unit 15 determines a machine learning parameter including information indicating the type of parameter to be searched, information indicating the range of values of the parameter to be searched, and information of a calculation formula.

Here, the rules for determining machine learning parameters will be described. FIG. 3 is a diagram showing an example of determination rules used by the processing condition search device 10 according to the first embodiment to determine machine learning parameters. A decision rule is represented by a correspondence between machining specifications and machine learning parameters. The determination rule storage unit 14 holds a database including information on processing specifications and information on machine learning parameters associated with the processing specifications.

In FIG. 3, the column of "machining specifications" shows three items of "machining type", "machining shape" and "important performance". The processing type information is information included in the machine learning setting information, but in FIG. 3, the processing type is treated as one of the items of the processing parameters. In FIG. 3, for example, in the row where the "machining type" is "rough machining", the "machining shape" is "round", and the "important performance" is "speed", "rough machining", "round" and " Machine learning parameter information corresponding to the machining parameter "velocity" is shown.

In the "search parameter" column of FIG. 3, the items "pause time", "jump parameter 1", "jump parameter 2", "servo parameter 1" and "servo parameter 2" represent control parameters. "Pause time" is a control parameter that indicates the time between discharges. Each of "jump parameter 1" and "jump parameter 2" is a control parameter for the electrode jump operation. Each of "servo parameter 1" and "servo parameter 2" is a parameter representing the number of discharges per unit time or the distance between the electrode and the workpiece.

In FIG. 3, "○" in the "search parameter" column indicates a control parameter to be searched. A "-" in the "search parameter" column indicates that the control parameter is not a search target. In FIG. 3, for example, for the machining specifications with "rough machining" for "machining type", "round" for "machining shape", and "speed" for "important performance", "pause time", "jump parameter 1 ”, “jump parameter 2” and “servo parameter 1” are searched.

In the column of "Search range" in FIG. "step size" and each item. In FIG. 3, for example, the “minimum value” of 3, the “maximum value” of 7, and the “interval” of 1 in the “pause time” column indicate that the control parameter “pause time” value search range is from 3 to 7. and indicates that the search is performed in the range at intervals of 1. In FIG. 3, the columns of "jump parameter 2", "servo parameter 1" and "servo parameter 2" in the column of "search range" are omitted. 1” and “servo parameter 2” are also set with data about values to be searched, as in the case of “pause time” and “jump parameter 1” shown in FIG. A "-" in the "search range" column indicates that the data for the value to be searched is not set.

The evaluation point calculator 25 acquires data necessary for evaluating the processing conditions from the state data. For example, the evaluation point calculator 25 obtains the machining speed, machining stability, and pulse state data from the state data. The evaluation point calculator 25 calculates an evaluation point by substituting the machining speed, the machining stability, and the pulse state data into the evaluation formula. In the column of "variables for calculating evaluation points" in FIG. 3, the items "machining speed", "machining stability" and "pulse state" represent the values of variables used in the evaluation point calculation formula. That is, the variable for each item represents weighting in evaluation of processing conditions.

For example, for the machining conditions corresponding to the machining specifications where the "machining type" is "rough machining", the "machining shape" is "round", and the "important performance" is "speed", the machining speed is emphasized. , a high value of "0.8" is set to the variable of "machining speed". Further, in the machining parameters, the machining stability may be ignored, and the variable of "machining stability" is set to "0". In addition, in view of the fact that even if the machining speed is highly evaluated, if the defect rate of the current pulse or voltage pulse is high, the machining speed will decrease. "0.2" is set to take into consideration the state.

In FIG. 3, the items of "machining specifications" in the decision rule are three, "processing type", "machining shape" and "important performance", but the items of "machining specifications" are limited to these not a thing The item of "processing specifications" can be set arbitrarily.

In FIG. 3, in the "search parameter" of the decision rule, the control parameters for which whether or not to search are set are "pause time", "jump parameter 1", "jump parameter 2", and "servo parameter 1". and "servo parameter 2", but the number is not limited to this. A control parameter for setting whether or not to search in the "search parameter" can be set arbitrarily.

In FIG. 3, there are three items, "machining speed", "machining stability" and "pulse state", for which "variables for calculating evaluation points" can be set in the decision rule. Items are not limited to these. The item of "Variables for calculating evaluation points" can be set arbitrarily.

The machine learning parameter determination unit 15 sets "search parameters" and " Each information of "search range" and "variable for calculating evaluation points" is read out from the database of the determination rule storage unit 14. FIG.

The search parameter determination unit 16 acquires information on "search parameters". When user-designated information about the type of control parameter is included in the machine learning setting information, the search parameter determination unit 16 adjusts the type of control parameter indicated in "search parameter" based on the user-designated information. Thereby, the search parameter determination unit 16 determines control parameters to be searched.

The search range determination unit 17 acquires information on the “search range”. The search range determining unit 17 adjusts the range of values indicated in the "search range" based on the standard processing conditions. For example, the search range determining unit 17 adjusts the values of the control parameters so as to shorten the pause time and jump operation time as compared to the standard machining conditions. Furthermore, when user-specified information about the value range of the control parameter is included in the machine learning setting information, the search range determination unit 17 adjusts the value range indicated by the “search range” based on the user-specified information. Thereby, the search range determination unit 17 determines the range of values to be searched for for each control parameter.

The evaluation point calculation formula determination unit 18 acquires information on "variables for evaluation point calculation". Accordingly, the evaluation point calculation formula determination unit 18 determines variables used in the evaluation point calculation formula. That is, the evaluation point calculation formula determination unit 18 determines the evaluation point calculation formula. Machine learning parameters including information indicating the type of parameter to be searched, information indicating the range of values of the parameter to be searched, and information on the determined calculation formula are input from the machine learning parameter determination unit 15 to the machine learning unit 19. be done.

The machining condition search device 10 inputs machining specifications to the machining specification input unit 11 and machine learning setting information including information on the type of machining to the machine learning setting input unit 13. Determining machine learning parameters corresponding to specifications and processing types. The machining condition search device 10 can determine machine learning parameters corresponding to arbitrary machining specifications and machining types specified by the user. By inputting machine learning setting information including user-specified information to the machine learning setting input unit 13, the processing condition search device 10 further performs processing based on the user's specification of the type of control parameter and the value range of the control parameter. , the machine learning parameters can be determined.

After the machine learning parameters are input to the machine learning section 19, the machining condition search device 10 starts outputting the machining conditions from the machining condition adjusting section 24 to the electric discharge machine 28 in step S4 shown in FIG. The electric discharge machine 28 processes the workpiece according to the machining conditions input from the machining condition search device 10 .

A reference machining condition is input to the machining condition adjusting unit 24 . At the start of machining by the electric discharge machine 28, the machining condition adjustment unit 24 outputs the reference machining condition as the machining condition. The electric discharge machine 28 starts machining according to the standard machining conditions.

After the machining by the electric discharge machine 28 is started, the machining condition search device 10 performs machine learning by the learning unit 22 in step S5. Further, the machining condition searching device 10 adjusts the machining conditions in the machining condition adjustment unit 24 based on the result of machine learning.

State data from the electric discharge machine 28 is input to the machine learning unit 19 while the electric discharge machine 28 is performing machining. The evaluation point calculator 25 obtains the machining speed, the machining stability, and the pulse state data from the state data. For example, the evaluation point calculator 25 estimates the machining speed from the amount of displacement of the machining depth. The evaluation point calculation unit 25 estimates the machining depth based on the amount of movement of the axis for moving the electrode and data on the occurrence of electric discharge. The discharge occurrence status is, for example, the number of discharge occurrences. The evaluation point calculation unit 25 obtains the pulse state data based on the result of the determination of whether the discharge is good or bad. Pulse state data is, for example, a value that indicates the percentage of bad pulses in a current or voltage pulse. Machining stability represents the variation in the number of discharge occurrences at regular intervals, for example, at intervals of several milliseconds. The evaluation point calculator 25 obtains the machining stability based on the number of discharge occurrences for each fixed period. The greater the variation in the number of discharge occurrences, the lower the machining stability. For example, if machining is not stable due to a momentary increase in the number of electrical discharge occurrences, it is assumed that there is a problem in machining control or that there is an abnormality in the surroundings of the electrical discharge position. The more uniform the number of discharge occurrences per fixed period, the higher the machining stability.

The evaluation point calculation unit 25 acquires information on the calculation formula input from the machine learning parameter determination unit 15 to the machine learning unit 19 . The evaluation point calculator 25 calculates an evaluation point by substituting the machining speed, the machining stability, and the pulse state data into the evaluation formula.

The evaluation points calculated by the evaluation point calculation unit 25 and the machining conditions to be evaluated are input to the state observation unit 23 . Among the machine learning parameters, information indicating the type of parameter to be searched and information indicating the range of values of the parameter to be searched are input to state observing section 23 . The state observation unit 23 observes the evaluation points, the processing conditions to be evaluated, information indicating the types of parameters to be searched, and information indicating the range of values of the parameters to be searched as state variables.

The learning unit 22 learns the optimum combination of control parameter values according to the data set created based on the state variables. The learning unit 22 generates a learning model for searching for the optimum combination of control parameter values among the types of parameters to be searched and the range of values of the parameters to be searched. Details of machine learning by the learning unit 22 will be described later.

The learning model generated by the learning section 22 is input to the machining condition adjusting section 24 . The machining condition adjustment unit 24 adjusts the machining conditions by inferring a combination of control parameter values to be searched based on the learning model. The machining condition search device 10 outputs the machining conditions adjusted by the machining condition adjustment unit 24 to the electric discharge machine 28 .

In this manner, the machine learning unit 19 learns machining conditions, and searches for machining conditions to be output to the electric discharge machine 28 based on the learning results. For example, when the electric discharge machine 28 performs rough machining according to the machining conditions input from the machining condition search device 10, the machine learning unit 19 performs machining under the optimum machining conditions until the electric discharge machine 28 finishes rough machining. Always search for machining conditions to meet the conditions.

In step S<b>6 , the machining condition search device 10 stores the adjustment result of the machining condition by the machining condition adjustment unit 24 in the machining condition adjustment result storage unit 20 . The machining condition adjustment result storage unit 20 holds all machining conditions adjusted by the machining condition adjustment unit 24 .

In step S<b>7 , the processing condition search device 10 updates the decision rule of the machine learning parameter in the decision rule update unit 21 . The determination rule updating unit 21 reads the processing conditions from the processing condition adjustment result storage unit 20 and updates the determination rule based on the adjustment result of the processing conditions. By updating the determination rule, the machining condition search device 10 can search for machining conditions more efficiently in future machining that is the same as or similar to the machining for which the machining conditions have been searched. With the above, the processing condition search device 10 ends the processing according to the procedure shown in FIG.

FIG. 4 is a diagram showing an example of a search range for machining conditions in the machining condition search device 10 according to the first embodiment. FIG. 4 shows an example of "minimum value", "maximum value" and "step size" for each of the control parameters "pause time", "jump parameter 1" and "jump parameter 2". The left part of FIG. 4 shows the value range of each control parameter before the search range is defined according to the first embodiment. The right part of FIG. 4 shows the search range defined according to the first embodiment.

Before the search range is defined, each of "pause time", "jump parameter 1" and "jump parameter 2" can take any of 16 values. In this case, there are 4,096 combinations of the values of "pause time," "jump parameter 1," and "jump parameter 2."

On the other hand, by defining the search range according to the first embodiment, the number of possible values of the "pause time" is reduced from 16 to 5. The number of possible values for each of "jump parameter 1" and "jump parameter 2" is reduced from sixteen to six. As a result, the number of combinations of the values of "pause time", "jump parameter 1" and "jump parameter 2" is reduced from 4096 to 180.

The machining condition search device 10 can search for machining conditions by limiting the search range of the machining conditions, thereby reducing the number of trials, which is the number of repetitions of observation of the machining state while adjusting the machining conditions. can be done. The machining condition search device 10 can search for optimum machining conditions accurately and efficiently.

Note that when the same processing specifications as those input during past processing are input, the learning unit 22 updates the learning model generated during the past processing through learning. It is good to do. As a result, the machining condition search device 10 can obtain a more accurate learning model, and can perform a highly accurate search for machining conditions.

Next, details of machine learning by the learning unit 22 will be described. The learning unit 22 creates a data set in which the state variables input from the state observation unit 23 are put together. The learning unit 22 learns, according to the data set, the optimum combination of control parameter values in the types of parameters to be searched and the range of values of the parameters to be searched. Any learning algorithm may be used by the learning unit 22 . As an example, a case where reinforcement learning is applied to the learning algorithm used by the learning unit 22 will be described.

Reinforcement learning is that an action subject, an agent, in an environment observes the current state and decides what action to take. Agents obtain rewards from the environment by selecting actions, and learn policies that maximize rewards through a series of actions. As representative methods of reinforcement learning, Q-learning and TD-learning are known. For example, in the case of Q-learning, the action-value table, which is a general update formula for the action-value function Q(s, a), is expressed by the following formula (1). The action value function Q(s, a) represents the action value Q, which is the action value of selecting the action 'a' under the environment 's'.

In equation (1) above, "s _t+1 " represents the environment at time "t". "a _t " represents an action at time "t". The action 'a _t ' changes the environment to 's _t+1 '. “r _t+1 ” represents the reward obtained by changing the environment. “γ” represents the discount rate. “α” represents a learning coefficient. When Q-learning is applied, the processing condition, that is, the combination of the control parameter values becomes the action "a _t ".

The update formula represented by the above formula (1) is that if the action value of the best action "a" at time "t+1" is greater than the action value Q of action "a" executed at time "t" , the action value Q is increased, and vice versa, the action value Q is decreased. In other words, the action value function Q(s, a) is updated so that the action value Q of action "a" at time "t" approaches the best action value at time "t+1". As a result, the best behavioral value in one environment propagates to the behavioral value in the previous environment in sequence.

The reward calculator 26 calculates a reward based on the state variables. The function updating unit 27 updates the function for determining the machining condition, which is a combination of control parameter values, according to the reward calculated by the reward calculating unit 26 .

The remuneration calculation unit 26 calculates a remuneration "r" based on the evaluation points, which are the evaluation results of the electric discharge machining state. For example, when the evaluation point increases as a result of changing the value of the control parameter, the remuneration calculator 26 increases the remuneration "r". The reward calculator 26 increases the reward "r" by giving "1", which is the value of the reward. Note that the reward value is not limited to "1". Further, when the evaluation point is lowered as a result of changing the value of the control parameter, the reward calculation unit 26 reduces the reward "r". The reward calculation unit 26 reduces the reward "r" by giving "-1", which is the value of the reward. Note that the reward value is not limited to "-1".

The function updater 27 updates the function for determining the processing conditions according to the reward calculated by the reward calculator 26 . Updating the function can be done according to the data set, for example by updating the action value table. The action value table is a data set in which arbitrary actions and their action values are associated and stored in the form of a table. For example, in the case of Q-learning, the action value function Q(s _t , a _t ) expressed by Equation (1) above is used as a function for calculating a combination of control parameter values.

FIG. 5 is a flow chart showing the procedure of learning by the learning unit 22 of the processing condition search device 10 according to the first embodiment. A reinforcement learning method for updating the action-value function Q(s, a) will be described with reference to the flowchart of FIG.

In step S11, the learning unit 22 acquires a data set created based on the state variables. In step S12, the learning unit 22 calculates a reward based on the evaluation points. In step S13, the learning unit 22 updates the action-value function Q(s, a) based on the reward. In step S14, the learning unit 22 determines whether or not the action-value function Q(s, a) has converged. The learning unit 22 determines that the action-value function Q(s, a) has converged by not updating the action-value function Q(s, a) in step S13.

When it is determined that the action-value function Q(s, a) has not converged (step S14, No), the learning unit 22 returns the procedure to step S11. When it is determined that the action-value function Q(s, a) has converged (step S14, Yes), the learning unit 22 ends learning according to the procedure shown in FIG. Note that the learning unit 22 may continue learning by returning the procedure from step S13 to step S11 without making the determination in step S14.

The learning unit 22 outputs the learning model, which is the generated action-value function Q(s, a), to the processing condition adjustment unit 24 . The machining condition adjustment unit 24 adjusts the machining conditions by inferring a combination of control parameter values to be searched based on the learning model.

Although the case where reinforcement learning is applied to the learning algorithm used by the learning unit 22 has been described in the first embodiment, learning other than reinforcement learning may be applied to the learning algorithm. The learning unit 22 may perform machine learning using known learning algorithms other than reinforcement learning, such as deep learning, neural networks, genetic programming, inductive logic programming, or support vector machines. good.

The learning unit 22 is not limited to being built in the processing condition searching device 10 . The learning unit 22 may be implemented by a device external to the processing condition searching device 10 . In this case, the device functioning as the learning unit 22 may be a device that can be connected to the processing condition search device 10 via a network. A device that functions as the learning unit 22 may be a device that exists on a cloud server.

The learning unit 22 may learn the processing conditions according to data sets created for a plurality of processing condition searching devices 10 . The learning unit 22 may create a data set based on data obtained from a plurality of processing condition search devices 10 used at the same site, or a plurality of processing condition search devices used at different sites. A data set may be created based on the data obtained from 10. The data sets may be collected from a plurality of processing condition search devices 10 that operate independently of each other at a plurality of sites. After starting to collect data sets from a plurality of processing condition searching devices 10, a new processing condition searching device 10 may be added to the targets from which data sets are collected. Moreover, after starting to collect data sets from the plurality of processing condition searching devices 10, some of the plurality of processing condition searching devices 10 may be excluded from the targets for which the data sets are collected.

The learning unit 22 that has learned about one processing condition searching device 10 may also learn about other processing condition searching devices 10 other than the processing condition searching device 10 concerned. The learning unit 22 that performs learning for the other processing condition searching device 10 can update the learning model by re-learning in the other processing condition searching device 10 .

Next, the hardware configuration of the processing condition search device 10 will be described. FIG. 6 is a diagram showing a hardware configuration example of the processing condition search device 10 according to the first embodiment. The configuration example shown in FIG. 6 is a configuration example in which the main part of the processing condition search device 10 is realized by a processing circuit 30 having a processor 32 and a memory 33. In FIG. The machining condition creating unit 12, the machine learning parameter determining unit 15, the machine learning unit 19, and the decision rule updating unit 21 shown in FIG.

The processor 32 is a CPU (Central Processing Unit). Processor 32 may be an arithmetic unit, microprocessor, microcomputer, or DSP (Digital Signal Processor). The memory 33 is, for example, RAM (Random Access Memory), ROM (Read Only Memory), flash memory, EPROM (Erasable Programmable Read Only Memory), EEPROM (registered trademark) (Electrically Erasable Programmable Read Only Memory), and the like.

The memory 33 stores a machining condition search program, which is a program for operating as a main part of the machining condition search device 10 . A main part of the machining condition search device 10 can be realized by having the processor 32 read out and execute the machining condition search program. The processing condition search program stored in the memory 33 may be provided in a state written in a storage medium such as a CD (Compact Disc)-ROM, a DVD (Digital Versatile Disc)-ROM, etc. Better, it may be in the form of being provided via a communication line.

A main part of the processing condition search device 10 may be realized by a dedicated processing circuit instead of the processor 32 and the memory 33 . A dedicated processing circuit may be a single circuit, a composite circuit, an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA), or a combination thereof. As for the main part of the processing condition search device 10, part of the functions may be realized by the processor 32 and the memory 33, and the remaining functions may be realized by a dedicated processing circuit.

The input unit 31 is a circuit that receives an input signal to the machining condition search device 10 from the outside. State data from the electrical discharge machine 28 is input to the input unit 31 . Input unit 31 also includes an input device that receives an input operation. The machining parameter input unit 11 and the machine learning setting input unit 13 are implemented by an input device.

The output unit 35 is a circuit that outputs the signal generated by the processing condition searching device 10 to the outside of the processing condition searching device 10 . The output unit 35 outputs the machining conditions adjusted by the machining condition search device 10 to the electric discharge machine 28 .

The storage device 34 stores various information. The storage device 34 is an HDD (Hard Disk Drive), an SSD (Solid State Drive), or the like. The determination rule storage unit 14 and the processing condition adjustment result storage unit 20 are implemented by the storage device 34 .

According to Embodiment 1, the machining condition search device 10 learns machining conditions for electric discharge machining, and searches for machining conditions to be output to the electric discharge machine 28 based on the learning result. The machining condition search device 10 determines machine learning parameters representing control parameters to be searched and the range of values of the control parameters to be searched, based on the machining parameters. The machining condition search device 10 determines machine learning parameters and searches for machining conditions by learning based on the determined machine learning parameters, thereby efficiently searching for machining conditions suitable for the state of machining. becomes. As described above, the machining condition search device 10 has the effect of being able to efficiently search for machining conditions for electrical discharge machining.

The configuration shown in the above embodiment shows an example of the content of the present disclosure. The configuration of the embodiment can be combined with another known technique. A part of the configuration of the embodiment can be omitted or changed without departing from the gist of the present disclosure.

10 machining condition search device 11 machining specification input unit 12 machining condition creation unit 13 machine learning setting input unit 14 determination rule storage unit 15 machine learning parameter determination unit 16 search parameter determination unit 17 search range determination unit , 18 evaluation point calculation formula determination unit, 19 machine learning unit, 20 processing condition adjustment result storage unit, 21 decision rule update unit, 22 learning unit, 23 state observation unit, 24 processing condition adjustment unit, 25 evaluation point calculation unit, 26 Remuneration calculation unit 27 function update unit 28 electric discharge machine 30 processing circuit 31 input unit 32 processor 33 memory 34 storage device 35 output unit.

Claims (11)

a machine learning unit that learns machining conditions for electric discharge machining by an electric discharge machine and searches for machining conditions to be output to the electric discharge machine based on the learning result;
A machine learning parameter that is part of a plurality of control parameters that make up the machining conditions and is a search target control parameter, and a part of the value range of the control parameter that indicates the search target range, a machine learning parameter determination unit that determines based on the machine learning parameter determination rule ;
with
In the decision rule, control parameters to be searched among preset control parameters, control parameters to be excluded from the search target, and values to be searched among the control parameter values to be searched. data is set in association with the machining specifications of the electric discharge machining,
A machining condition searching apparatus, wherein the machine learning unit searches for the machining condition by limiting a control parameter and a range of values of the control parameter based on the machine learning parameter. The machine learning unit
Information indicating the type of control parameter to be searched among the machine learning parameters, information indicating the range of values of the control parameter to be searched, and information about the state of the electrical discharge machining are observed as state variables. a state observation unit;
a learning unit that learns an optimal combination of control parameter values according to a data set created based on the state variables;
a machining condition adjustment unit that adjusts machining conditions to be output to the electric discharge machine based on the learning result of the learning unit;
The machining condition searching device according to claim 1, characterized by comprising: The machine learning unit has an evaluation point calculation unit that calculates an evaluation point for the machining conditions output to the electric discharge machine based on state data indicating the state of the electric discharge machining,
3. The machining condition searching apparatus according to claim 2, wherein the state observation unit observes the evaluation point, which is information about the state of the electrical discharge machining, as the state variable. In the determination rule, values of variables used in calculation formulas for calculating the evaluation points are set in association with the machining parameters,
4. The processing condition search apparatus according to claim 3, wherein the machine learning parameter determination unit determines the machine learning parameter including information of the calculation formula based on the determination rule . 5. The processing condition search device according to claim 1 , further comprising a decision rule storage unit that holds the decision rule. 3. The processing condition searching apparatus according to claim 2 , further comprising a determination rule updating unit that updates the determination rule based on a result of adjustment of the processing conditions by the processing condition adjustment unit. 7. The machining condition searching apparatus according to claim 1, further comprising a machining parameter input unit into which information on said machining parameters is inputted. a machining condition creation unit that creates a reference machining condition, which is a machining condition composed of initial values of control parameters, based on the machining parameters;
8. The machining condition searching apparatus according to claim 7, wherein the machine learning parameter determination unit determines the machine learning parameter based on the machining specifications and the reference machining conditions. A machine learning setting input unit for inputting machine learning setting information indicating the type of processing to be searched for processing conditions,
8. The machining condition search apparatus according to claim 7, wherein the machine learning parameter determination unit determines the machine learning parameter based on the machining specifications and the machine learning setting information. The machine learning setting information includes information for a user of the machining condition searching device to specify a type of control parameter to be searched or a range of values of the control parameter to be searched. Item 10. The processing condition search device according to Item 9. A machining condition searching method in which a machining condition searching device searches for machining conditions for electric discharge machining by an electric discharge machine,
A machine learning parameter that is part of a plurality of control parameters that make up the machining conditions and is a search target control parameter, and a part of the value range of the control parameter that indicates the search target range, determining based on the machine learning parameter determination rule ;
a step of performing learning for searching for machining conditions by limiting the control parameters and the range of values of the control parameters based on the machine learning parameters;
adjusting the machining conditions to be output to the electric discharge machine based on the learning result;
including
In the decision rule, control parameters to be searched among preset control parameters, control parameters to be excluded from the search target, and values to be searched among the control parameter values to be searched. data is set in association with the machining specifications of the electric discharge machining .

Images