JPWO2020012581A1 - Machine learning device, numerical control machining program generating device, and machine learning method - Google Patents

Abstract

A machine learning device (50) is a numerical control including machining shape data including information on a machining finish shape of a machining product and information on a material of a material, and a plurality of machining processes for cutting a machining product from a workpiece. State observation for observing, as state variables, a numerical control machining program generation method for automatically generating a machining program, and edit contents for the first numerical control machining program generated by referring to the numerical control machining program generation method A section (51) and a learning section (52) for learning a method of generating a numerically controlled machining program according to a data set created based on state variables.

Description

The present invention relates to a machine learning device, a numerical control machining program generation device, and a machine learning method for learning a generation method of a numerical control machining program used for automatically generating a numerical control machining program for numerically controlling a machine tool.

In recent years, in the field of Numerical Control (NC) machine tools, in order to enable automatic machining of workpieces of complicated and various shapes, various types of tools are exchangeably equipped on the tool post, and turning is possible. A complex mechanization that enables various processing steps such as drilling, milling, etc. is under way.

In Patent Document 1, the details of cutting conditions corrected by an operator with respect to an automatically generated machining program are stored in a storage unit, and when another operator next creates a similar machining program, A numerical control machining program creation device capable of reusing cutting conditions is disclosed.

JP, 2002-132313, A

However, in the numerical control machining program creation device of Patent Document 1 described above, the reusable items are the number of rotations of the spindle that rotates the material to be machined, the relative cutting feed between the rotary tool and the material to be machined. Only the cutting conditions such as quantity and rounding up, it was not possible to automatically generate a processing program regarding the processing sequence and processing method.

The present invention has been made in view of the above, and can automatically learn a generation method of a numerically controlled machining program that enables a machining program to be automatically generated with respect to at least one of a machining order and a machining method. The purpose is to obtain a machine learning device.

In order to solve the above-mentioned problems and achieve the object, a machine learning apparatus according to the present invention is provided with machining shape data including information on a machining finish shape of a machined product and information on a material of a material, and cutting from a workpiece. A numerical control machining program generation method for automatically generating a numerical control machining program including a plurality of cutting processes for carving a workpiece, and a first numerical control generated by referring to the numerical control machining program generation method A state observing unit that observes the edited contents of the machining program as a state variable, and a learning unit that learns a generation method of the numerically controlled machining program according to a data set created based on the state variable.

According to the present invention, it is possible to obtain a machine learning device capable of automatically generating a machining program with respect to at least one of a machining sequence and a machining method and capable of automatically learning a method of generating a numerically controlled machining program. Play.

Block diagram showing a configuration example of a numerical control device according to a first embodiment of the present invention Diagram showing the concept of action value table The figure which shows an example of the state in which the processing knowledge and the reflection processing knowledge were memorize|stored in the function update part of the machine learning apparatus concerning Embodiment 1 of this invention. The flowchart which shows the procedure of the update process of the action value function Q(s, a) in Embodiment 1 of this invention. The flowchart which shows the detail of the procedure of NC processing program generation processing in the NC processing program generation part of the NC processing program generation apparatus concerning Embodiment 1 of this invention. 1 is a perspective view showing an example of a three-dimensional model of a processed finish shape represented by CAD (Computer-Aided Design) data according to Embodiment 1 of the present invention. 1 is a perspective view showing an example of a material shape including a processed finish shape represented by CAD data according to Embodiment 1 of the present invention. A perspective view showing an example of a machining removal shape generated from a machining finish shape and material shape represented by CAD data in the first embodiment of the present invention. The figure which shows the turning cross-sectional shape produced|generated from the raw material shape and processed shape in Embodiment 1 of this invention, and a coordinate value. The figure which shows the turning cross-sectional shape of the turning end surface process concerning Embodiment 1 of this invention. The figure which shows the turning cross-sectional shape of the turning drill process concerning Embodiment 1 of this invention. The figure which shows the turning cross-sectional shape of the turning bar material process concerning Embodiment 1 of this invention. The figure which shows the turning cross-sectional shape of the turning bar material process concerning Embodiment 1 of this invention. The figure which shows an example of the processing start point and processing end point of the turning end surface process concerning Embodiment 1 of this invention. The figure which shows an example of the processing start point and processing end point of the turning drill process concerning Embodiment 1 of this invention. The figure which shows an example of the cutting point, the processing start point, and the processing end point of the turning bar material process concerning Embodiment 1 of this invention. The figure which shows an example of the cutting point, the processing start point, and the processing end point of the turning bar material process concerning Embodiment 1 of this invention. The figure which shows the turning cross-sectional shape of the turning drill process concerning Embodiment 1 of this invention. The figure which shows the turning cross-sectional shape of the turning end face process concerning Embodiment 1 of this invention. The figure which shows the turning cross-sectional shape of the turning bar material process concerning Embodiment 1 of this invention. The figure which shows the turning cross-sectional shape of the turning bar material process concerning Embodiment 1 of this invention. The figure which shows an example of the processing start point and processing end point of the turning drill process concerning Embodiment 1 of this invention. The figure which shows an example of the processing start point and processing end point of the turning end surface process concerning Embodiment 1 of this invention. The figure which shows an example of the cutting point, the processing start point, and the processing end point of the turning bar front face process concerning Embodiment 1 of this invention. The figure which shows an example of the cutting point of a turning rod grooving process, a processing start point, and a processing end point concerning Embodiment 1 of this invention. A schematic diagram showing an example of a neural network model that the machine learning device according to the first exemplary embodiment of the present invention follows. The figure which shows the hardware constitutions of the NC processing program generation apparatus concerning Embodiment 1 of this invention.

Hereinafter, a machine learning device, a numerical control machining program generating device, and a machine learning method according to an embodiment of the present invention will be described in detail with reference to the drawings. The present invention is not limited to this embodiment.

Embodiment 1.
FIG. 1 is a block diagram showing a configuration example of a numerical control device 1 according to the first exemplary embodiment of the present invention. As shown in FIG. 1, the numerical control device 1 includes an interactive operation processing unit 10, a display unit 20, an instruction input unit 30, and an NC machining program generation device 40. The NC machining program generation device 40 includes a machine learning device 50. The machine learning device 50 includes a state observing section 51 and a learning section 52. The numerical control device 1 is mounted on or connected to a numerically controlled machine tool (not shown), and numerically controls the operation of the numerically controlled machine tool by an NC machining program.

The interactive operation processing unit 10 is an interface unit between the numerical control device 1 and the worker, and also an interface unit between the NC machining program generating device 40 and the worker. The interactive operation processing unit 10 transmits the instruction information input by the operator via the instruction input unit 30 to the NC machining program generation device 40. Further, the interactive operation processing unit 10 displays on the display unit 20 the instruction information input by the operator via the instruction input unit 30.

The display unit 20 is a display terminal such as a liquid crystal monitor, and displays the CAD data 100, the NC machining program, and the instruction information input by the operator via the instruction input unit 30. The display unit 20 also displays various kinds of information regarding the processing performed by the numerical control device 1 and the NC machining program generation device 40.

The instruction input unit 30 is configured to include input devices such as a mouse and a keyboard, receives instruction information from a worker, and transmits the instruction information to the interactive operation processing unit 10.

The NC machining program generation device 40 is used to cut out the shape of the machining product represented by the CAD data 100 input to the NC machining program generation device 40 from the outside of the numerical controller 1 from the workpiece by cutting. An NC machining program that is generated. The NC machining program is a numerical machining program for operating an NC machine tool (not shown). Further, the numerical control device 1 controls the NC machine tool according to the NC machining program, and cuts the workpiece to produce a cut product.

The NC machining program generation device 40 includes a machining shape data input unit 41, a machining shape data storage unit 42, an NC machining program generation unit 43, an NC machining program storage unit 44, an NC machining program editing unit 45, and decision making. And a section 46. Each component of the NC machining program generation device 40 can exchange information with each other.

The machining shape data input unit 41 receives the CAD data 100, which is the machining shape data input to the NC machining program generating apparatus 40 from an external device outside the numerical control apparatus 1, and transmits it to the machining shape data storage unit 42. The machined shape data includes information on the machined shape of the machined product and information on the material. The finished shape is a three-dimensional shape that must be finished by cutting to obtain a cut product. The material is a work piece whose machining finish shape represented by CAD data 100 is cut out. However, the processed shape data is not limited to the CAD data 100. The machining shape data may be data that can be interpreted by the NC machining program generation device 40.

The processed shape data storage unit 42 stores the processed shape data transmitted from the processed shape data input unit 41.

The NC machining program generation unit 43 generates an NC machining program for cutting a machining finish shape represented by the CAD data 100 from a material. Specifically, the NC machining program generation unit 43 refers to the machining shape data of the machined product and refers to the machining knowledge or the reflected machining knowledge acquired from the function updating unit 54 via the decision making unit 46 to perform a plurality of cuttings. The processing order and processing method in processing are determined. Then, the NC machining program generation unit 43 generates the first NC machining program 111, which is an NC machining program for cutting out a machined product from the workpiece, based on the determined machining order and machining method. Further, the NC machining program generation unit 43 associates the CAD data 100, which is machining shape data, the machining knowledge referred to in the generation of the first NC machining program 111, and the first NC machining program 111, with the CAD data 100 and the first NC machining program 111. The 1NC machining program 111 and machining knowledge are transmitted to and stored in the NC machining program storage unit 44.

The machining knowledge carves out a machined product from the material when the NC machining program generator 43 automatically generates an NC machining program for carving out the machining finish shape of the machined product represented by the CAD data 100 from the material. Is a determination condition referred to in order to determine at least one of a material processing order and a material processing method in a plurality of cutting processes, and is information indicating a method for generating an NC processing program.

Specifically, the processing knowledge is a condition that specifies at least one of the processing order of the material and the processing method of the material based on the dimension of the processing removal shape, which is the shape to be removed in order to cut out the machined product from the material. Is. The processing order of the material is the order of the cutting processing when a plurality of different cutting processings are combined in order to cut out a machined product from the material. The material processing method is a cutting method including a processing direction.

When the operator edits the first NC machining program 111 automatically generated by the NC machining program generation unit 43 as described later, the reflected machining knowledge that is the new machining knowledge in which the edited contents are reflected is generated. As a result, the new machining knowledge reflects and accumulates the edited contents that reflect the knowledge and experience of the operator. Therefore, the NC machining program generation unit 43 generates the NC machining program by referring to the machining knowledge, so that the NC machining program desired by the operator, which is the same as when the operator himself creates the NC machining program, is created. It is possible to quickly and automatically generate.

The NC machining program storage unit 44 refers to the CAD data 100 that is machining shape data, the first NC machining program 111 that is the NC machining program generated by the NC machining program generation unit 43, and the generation of the first NC machining program 111. The processing knowledge is stored in association with it. Further, the NC machining program storage unit 44 stores the second NC machining program 112 which is the NC machining program edited by the NC machining program editing unit 45. The first NC machining program 111 may be updated by the second NC machining program 112.

The NC machining program editing unit 45 reads the CAD data 100, the first NC machining program 111, and the machining knowledge associated with the first NC machining program 111 from the NC machining program storage unit 44, and sets the first NC machining program 111 to the interactive operation processing unit 10. Send to. The NC machining program editing unit 45 edits the first NC machining program 111 based on the editing instruction information input by the operator from the instruction input unit 30, and generates the second NC machining program 112. The NC machining program editing unit 45 transmits the generated second NC machining program 112 to the NC machining program storage unit 44 to store it therein. Further, the NC machining program editing unit 45 will describe the CAD data 100 that is machining shape data, the first NC machining program 111, the machining knowledge associated with the first NC machining program 111, and the second NC machining program 112, which will be described later. It is transmitted to the state observing section 51.

The NC machining program editing unit 45, when the first NC machining program 111 is not edited, that is, when the second NC machining program 112 is not generated, the CAD data 100, the first NC machining program 111, The machining knowledge associated with the first NC machining program 111 is transmitted to the state observation unit 51.

The decision making unit 46 searches for a plurality of different machining knowledge stored in the function updating unit 54 in response to the request from the NC machining program generating unit 43, and selects the most appropriate machining knowledge corresponding to the CAD data 100. Determine and get. The most appropriate machining knowledge corresponding to the CAD data 100 is the editing knowledge for the first NC machining program 111 after the generation of the first NC machining program 111 among the machining knowledge corresponding to the CAD data 100 stored in the function updating unit 54. Is the least processing knowledge. That is, the most appropriate machining knowledge corresponding to the CAD data 100 is, among the machining knowledge corresponding to the CAD data 100, the machining knowledge in which the knowledge and experience of the operator are more appropriately reflected.

The decision making unit 46 acquires the machining knowledge having the highest action value Q from the plurality of machining knowledge corresponding to the CAD data 100 stored in the function updating unit 54, and transmits it to the NC machining program generating unit 43. .. When only one machining knowledge corresponding to the CAD data 100 is stored in the function updating unit 54, the decision making unit 46 stores the machining knowledge stored in the function updating unit 54 in the CAD data 100. Decide as the most appropriate machining knowledge that corresponds.

The processing knowledge is associated with the cutting product by, for example, being assigned a management number common to the cutting product. The machining knowledge is associated with a combination of the machining finish shape and the material of the material represented by the CAD data 100, for example. As a result, the decision making unit 46 searches for the machining knowledge corresponding to the CAD data 100 designated by the NC machining program generating unit 43, using the control number or the information of the machining finish shape and the information of the material of the material as a search condition. Further, the processing knowledge having the highest action value Q can be selected from among them.

Any learning algorithm may be used as the learning algorithm used by the machine learning device 50 including the state observation unit 51 and the learning unit 52. As an example, a case where reinforcement learning is applied will be described.

Reinforcement learning is that an agent who is an action subject in an environment observes the current state and decides the action to be taken. Agents get rewards from the environment by choosing actions, and learn strategies to get the most rewards through a series of actions. Q-learning or TD-learning is known as a typical method of reinforcement learning. For example, in the case of Q learning, a general update formula of the action value function Q(s,a) is represented by the following formula (1).

In Equation (1), _{s t} shows a state at time t. a _t is an action executed at time t. s _t+1 indicates the state at time t+1, in other words, the state after the change as a result of the action a _t . r _t+1 indicates the reward given according to the environment that changes as a result of the action a _t . That is, r _t+1 represents the reward given by the change in state. In Expression (1), the term “max” represents the maximum value of the action value Q in the state s _t+1 , that is, the action value for the best action a _t+1 . γ is a discount rate and is set so as to satisfy the condition of 0<γ≦1. γ is, for example, 0.9 or more and 0.99 or less. α is a learning coefficient and is set so as to satisfy the condition of 0<α≦1. α is, for example, 0.05 or more and 0.2 or less.

The update expression represented by the mathematical expression (1) in Q learning is such that if the action value of the best action a _{t+1 at} the time t+1 is larger than the action value Q of the action a _t executed at the time t, the action at the time t. The value Q is increased, and in the opposite case, the action value Q at time t is decreased. In other words, the action value function Q(s _t , a _t ) is updated so that the action value Q of the action a _t at time t approaches the best action value at time t+1. As a result, the best action value in a certain environment is sequentially propagated to the action value in the previous environment.

That is, the machine learning device 50 according to the first embodiment can execute the reinforcement learning according to the known Q learning method. The action value Q assigned to a combination of a certain state variable and action a is unknown. The machine learning device 50 randomly selects and executes the action a for various state variables, and updates the action value function Q(s, a) by integrating the reward given as the result of the action a. .. In the NC machining program generation device 40 according to the first embodiment, “action a” corresponds to machining knowledge, that is, information indicating a method of generating the NC machining program. Further, the “state s” corresponds to the editing content for the first NC machining program 111.

The action value function Q(s,a) is also called an action value table. FIG. 2 is a diagram showing the concept of the action value table. The action value table is a data set in which an arbitrary action s and its action value Q are associated and stored in a table format. The data set in Q learning is a so-called Q table. When the Q table is expressed as a two-dimensional table, the row direction is the various states s and the column direction is the possible actions a, and the action value that is the evaluation value in the case of the combination of the actions a and the states s in each square. Q is stored. However, when the action value function Q(s,a) is managed by the Q table, the number of states s and actions a that can be held is limited. The action value Q is calculated by the above equation (1) without taking the form.

When the Q table is adopted in the first embodiment, the machining knowledge, which is the information indicating the method of generating the NC machining program corresponding to the action a, and the editing content of the first NC machining program 111 corresponding to the state s, are combined. The Q table associated with the processed shape data is stored for each processed shape data. That is, the data set according to the first embodiment is a set of the data of the machining shape data, the data of machining knowledge which is the information indicating the generation method of the NC machining program, and the data of the editing content for the first NC machining program 111. Is.

The numerical controller 1 has a machine learning function for learning how to generate an NC machining program. That is, the numerical control device 1 includes the machine learning device 50 that learns the method of generating the NC machining program. The machine learning device 50 includes a state observing section 51 and a learning section 52. The learning unit 52 learns the method of generating the NC machining program according to the data set created based on the state variables. That is, the learning unit 52 learns the relationship between the machining shape data, the NC machining program generation method, and the edit content for the first NC machining program 111 according to the data set created based on the state variables, and the NC machining program Learn how to generate. The learning unit 52 updates the reward calculation unit 53 that calculates the reward for learning the machining knowledge that is the NC machining program generation method and the function that relates to the learning of the machining knowledge that is the NC machining program generation method, and that is an evaluation value. And a function updating unit 54 that determines the action value Q. The machine learning device 50 may be built in the NC machining program generation device 40, or may exist in a cloud server that can communicate with the NC machining program generation device 40.

The state observing unit 51 includes CAD data 100 that is machining shape data, and machining knowledge that is information indicating a method of generating a numerically controlled machining program that the NC machining program generator 43 refers to when generating the first NC machining program 111. , Edit contents for the first NC machining program 111 are observed as state variables. That is, the state observing unit 51 processes the machining knowledge determined by the decision making unit 46 and the editing content for the first NC machining program 111, which is performed by the NC machining program editing unit 45 when the second NC machining program 112 is generated. Observation is made for each CAD data 100 which is shape data.

The state observing unit 51 reflects the edit content of the first NC machining program 111 on the machining knowledge referred to in the generation of the first NC machining program 111 to create a reflected machining knowledge. The reflected machining knowledge is a method of generating a reflected numerically controlled machining program in which the editing content of the first NC machining program 111 is reflected in the method of generating the numerically controlled machining program referred to in the generation of the first NC machining program 111. The reward r of the machining knowledge associated with the first NC machining program 111 is inherited to the reflected machining knowledge. The state observing unit 51 transmits the machining knowledge referred to in the generation of the first NC machining program 111 and the reflected machining knowledge to the reward calculating unit 53 as an observation result.

When the reflected machining knowledge is generated in the state observing unit 51, the management number or the machining finish shape information and the material quality information given to the original machining knowledge are inherited and given. The reflected machining knowledge and the machining knowledge generated based on the machining knowledge both correspond to the machining finish shape of one cutting product. That is, the reflected machining knowledge corresponds to the finished shape of the machined product corresponding to the machining knowledge on which the reflected machining knowledge is generated. Therefore, the reflected processing knowledge is given by inheriting the control number, or the information on the processing finish shape and the information on the material of the material, which is given to the processing knowledge that is the basis of generation. That is, the reflection processing knowledge is given while the association with the CAD data 100 is inherited.

The state observing unit 51 receives the CAD data 100, the first NC machining program 111, the machining knowledge associated with the first NC machining program 111, and the second NC machining program 112 from the NC machining program editing unit 45. Can extract the editing content for the first NC machining program 111. Therefore, in this case, the new machining knowledge in which the editing content of the CAD data 100 with respect to the first NC machining program 111 is reflected in the machining knowledge referred to in the generation of the first NC machining program 111 is the reflected machining knowledge.

On the other hand, when only the CAD data 100, the first NC machining program 111, and the machining knowledge associated with the first NC machining program 111 are transmitted from the NC machining program editing unit 45 and the second NC machining program 112 is not transmitted. Has no edit content for the first NC machining program 111. In this case, the machining knowledge itself referred to in the generation of the first NC machining program 111 becomes the reflected machining knowledge.

Therefore, the CAD data 100 that is the machining shape data, the machining knowledge that is information indicating the method of generating the numerically controlled machining program that the NC machining program generation unit 43 referred to when generating the first NC machining program 111, and the first NC machining. Observing the edit content for the program 111 as a state variable means extracting the edit content for the first NC machining program 111 with respect to the CAD data 100 and reflecting the machining knowledge referred to in the generation of the first NC machining program 111 and the observation result. It means that the processing knowledge is acquired for each CAD data 100.

The learning unit 52 includes CAD data 100 that is machining shape data, machining knowledge that is information indicating a method of generating a numerically controlled machining program that the NC machining program generator 43 refers to when generating the first NC machining program 111, and Learning the NC machining program generation method by learning the relationship between the machining shape data, the NC machining program generation method, and the edited content for the first NC machining program 111 based on the data set with the editing content for the first NC machining program 111. To do. That is, the learning unit 52 learns the processing knowledge based on the state variable observed by the state observation unit 51. In one embodiment, the learning unit 52 learns processing knowledge according to the concept of reinforcement learning.

The reward calculation unit 53 calculates the reward r based on the reflected processing knowledge that is the observation result transmitted from the state observation unit 51. More specifically, the reward calculation unit 53 calculates the reward r of the reflected machining knowledge based on the machining knowledge referred to in the generation of the first NC machining program 111 and the reflected machining knowledge. The reward calculation unit 53 calculates the reward r of the reflected processing knowledge based on the identity of the processing knowledge referred to in the generation of the first NC processing program 111 and the reflected processing knowledge, which is a reward giving criterion for giving the reward r. To do. The reward calculating unit 53 associates the calculated reward r with the CAD data 100 and the reflection processing knowledge, and transmits the calculated reward r to the function updating unit 54 together with the CAD data 100 and the reflection processing knowledge.

The reward calculation unit 53 increases the reward r of the reflected machining knowledge when the machining knowledge referred to in the generation of the first NC machining program 111 and the reflected machining knowledge are the same. When the machining knowledge referred to in the generation of the first NC machining program 111 and the reflected machining knowledge are the same, the reward calculation unit 53 reflects the reward r of “+1”, which is the “correct” reward r, for example. Give to knowledge. Then, the reward r of the reflected machining knowledge is added with “+1” to the reward r that is inherited from the machining knowledge referred to in the generation of the first NC machining program 111 when the compensation r is generated. Thereby, the reward r of the reflected machining knowledge is larger than the reward r inherited from the machining knowledge referred to in the generation of the first NC machining program 111.

When the machining knowledge referred to in the generation of the first NC machining program 111 and the reflected machining knowledge are different, the reward calculation unit 53 reduces the compensation r of the machining knowledge referred to in the generation of the first NC machining program 111. When the machining knowledge referred to in the generation of the first NC machining program 111 and the reflected machining knowledge are different from each other, the reward calculation unit 53 gives a reward r of “−1”, which is a “negative” reward r, to the first NC. It is given to the machining knowledge referred to in the generation of the machining program 111. For example, in the case where the first NC machining program 111 generated so that the turning end face front machining is performed before the turning drill machining is edited so that the turning drill machining is performed before the turning end face front machining. , The reward calculation unit 53 reduces the reward r, and gives the reward r of, for example, “−1” to the processing knowledge referred to in the generation of the first NC processing program 111. Then, for the reward r of the machining knowledge referred to in the generation of the first NC machining program 111, “−1” is added to the current reward r. As a result, the reward r of the machining knowledge referred to in the generation of the first NC machining program 111 is smaller than the reward r when referred to in the generation of the first NC machining program 111.

Further, when the machining knowledge referred to in the generation of the first NC machining program 111 and the reflected machining knowledge are different from each other, the reward calculation unit 53 calculates the machining knowledge referred to in the generation of the first NC machining program 111 as described above. After reducing the reward r, the reward r of the reflection processing knowledge is increased. When the machining knowledge referred to in the generation of the first NC machining program 111 and the reflected machining knowledge are different from each other, the reward calculation unit 53 sets the reward r of “+1”, which is a “positive” reward r, to the reflected machining knowledge, for example. give. Then, the reward r of the reflected machining knowledge is added with “+1” to the reward r that is inherited from the machining knowledge referred to in the generation of the first NC machining program 111 when the compensation r is generated. Thereby, the reward r of the reflected machining knowledge is larger than the reward r inherited from the machining knowledge referred to in the generation of the first NC machining program 111.

The numerical value of the reward r is not limited to “+1” and “−1”. Further, the remuneration r may be differentiated depending on the edit content of the first NC machining program 111, that is, the edit content from the machining knowledge referred to in the generation of the first NC machining program 111 to the reflected machining knowledge.

The function updating unit 54 updates the function for determining the generation method of the NC machining program according to the reward r of the reflected machining knowledge calculated by the reward calculation unit 53. For example, the function updating unit 54 updates the action value function Q(s,a) to determine the action value Q that is an evaluation value. The function updating unit 54 stores the determined action value Q in association with the reflection processing knowledge together with the reflection processing knowledge. As a result, in addition to the machining knowledge stored first, the function updating unit 54 refers to the reflected machining knowledge and the action value Q associated with the reflected machining knowledge in generating the first NC machining program 111. It is memorized as processing knowledge. FIG. 3 is a diagram showing an example of a state in which the processing knowledge and the reflected processing knowledge are stored in the function updating unit 54 of the machine learning device 50 according to the first exemplary embodiment of the present invention. FIG. 3 shows an example in which machining knowledge A121, machining knowledge B122, machining knowledge C123, reflection machining knowledge A131, reflection machining knowledge B132, and reflection machining knowledge C133 are stored. A machining knowledge storage unit that stores the machining knowledge, the reflected machining knowledge, and the action value Q associated with the reflected machining knowledge may be separately provided.

Next, a reinforcement learning method for updating the action value function Q(s,a) will be described. FIG. 4 is a flowchart showing the procedure of the updating process of the action value function Q(s,a) according to the first embodiment of the present invention.

In step ST110, the NC machining program generation unit 43 reads the CAD data 100 stored in the machining shape data storage unit 42 and stores it in the shape storage area of the storage unit (not shown) in the NC machining program generation unit 43.

Next, in step ST120, the NC machining program generation unit 43 acquires the machining knowledge corresponding to the CAD data 100 from the function updating unit 54 via the decision making unit 46 in order to generate the NC machining program. The decision making unit 46 decides machining knowledge based on the result learned by the machine learning device 50. That is, the decision making unit 46 searches for the plurality of machining knowledge stored in the function updating unit 54, and selects the action value Q from among the plurality of machining knowledge corresponding to the request from the NC machining program generating unit 43. The highest machining knowledge is determined, acquired, and transmitted to the NC machining program generation unit 43.

An example of machining knowledge transmitted to the NC machining program generation unit 43 is shown below.
(Processing knowledge A121)
Material Material: S45C
Processing order: When the end chamfering allowance is 20 mm or more, the turning drill step is performed first, and then the turning end surface step is performed. When the edge chamfering margin is other than the above-mentioned conditions, the turning end face step is performed first, and then the turning drill step is performed.
(Processing knowledge B122)
Material Material: S45C
Machining site: When "horizontal length of open part shape/longitudinal length of open part shape" is "less than 1," the processed part is the front side. When "horizontal length of open part shape/longitudinal length of open part shape" is "1 or more", the outer diameter is defined as the processed part. The open portion is a portion exposed on the surface of the raw material in the turning sectional shape t.
(Processing knowledge C123)
Material Material: S45C
Machining method: When the length of the lathe cross section in the vertical direction is 10 mm or less and the length of the lathe cross section in the lateral direction is 10 mm or less, the turning grooving step is performed. When the length in the vertical direction and the length in the horizontal direction of the turning cross-sectional shape are other than the above-mentioned conditions, the turning rod material step is performed.

Next, in step ST130, the NC machining program generation unit 43 refers to the acquired machining knowledge to generate the first NC machining program 111, which is an NC machining program.

Next, in step ST140, the NC machining program generation unit 43 associates the CAD data 100 with the machining knowledge referred to in the generation of the first NC machining program 111 and the first NC machining program 111 to associate the CAD data 100 with the first NC machining program 111. And the machining knowledge are transmitted to the NC machining program storage unit 44. The NC machining program storage unit 44 stores the CAD data 100, the first NC machining program 111, and machining knowledge.

Next, in step ST150, the NC machining program editing unit 45 reads the CAD data 100, the first NC machining program 111, and the machining knowledge associated with the first NC machining program 111 from the NC machining program storage unit 44, and reads the first NC machining program. 111 is transmitted to the interactive operation processing unit 10. The interactive operation processing unit 10 causes the display unit 20 to display the first NC machining program 111 transmitted from the NC machining program editing unit 45, and prompts the operator to confirm the first NC machining program 111.

When the first NC machining program 111 is displayed on the display unit 20, the operator confirms the first NC machining program 111 and inputs the edit instruction information of the first NC machining program 111 to the instruction input unit 30. The editing instruction information is transmitted from the instruction input unit 30 to the NC machining program editing unit 45 of the NC machining program generating device 40 via the interactive operation processing unit 10.

Next, in step ST160, the NC machining program editing unit 45 receives the editing instruction information, edits the first NC machining program 111 based on the editing instruction information, and generates the second NC machining program 112. The second NC machining program 112 is an NC machining program in which the first NC machining program 111 is edited based on the editing instruction information. Then, the NC machining program editing unit 45 transmits the CAD data 100, the first NC machining program 111, the machining knowledge associated with the first NC machining program 111, and the second NC machining program 112 to the state observing unit 51.

Next, in step ST170, the state observing unit 51 compares the first NC machining program 111 and the second NC machining program 112 transmitted from the NC machining program editing unit 45, and acquires the editing content for the first NC machining program 111. .. Thereby, the state observing unit 51 acquires, as state variables, the CAD data 100 that is the machining shape data, the machining knowledge associated with the first NC machining program 111, and the edit content for the first NC machining program 111.

Next, in step ST180, the state observing unit 51 reflects the edit content of the first NC machining program 111 on the machining knowledge referred to in the generation of the first NC machining program 111 to generate reflected machining knowledge. The state observing unit 51 transmits the machining knowledge associated with the first NC machining program 111 and the reflected machining knowledge to the reward calculating unit 53 as an observation result.

Next, in step ST190, the reward calculation unit 53 compares the machining knowledge associated with the first NC machining program 111 with the reflected machining knowledge, and determines whether the two are the same. If both are the same, Yes is obtained in step ST190 and the process proceeds to step ST200. If they are different, No is obtained in step ST190 and the process proceeds to step ST210.

In step ST200, the reward calculation unit 53 calculates the reward r that increases the reward r of the reflected processing knowledge so that the action value Q with respect to the reward r of the reflected processing knowledge increases, and the processing associated with the first NC processing program 111 is performed. The reward r of the reflection processing knowledge inherited from the knowledge is accumulated. Thereby, the reward r of the reflected machining knowledge is larger than the reward r inherited from the machining knowledge associated with the first NC machining program 111. As a result, the action value Q of the reflected processing knowledge increases.

When the machining knowledge associated with the first NC machining program 111 and the reflected machining knowledge are the same, the machining knowledge itself associated with the first NC machining program 111 is the reflected machining knowledge. Then, the reflected machining knowledge is considered to be high-quality reflected machining knowledge because the editing content for the first NC machining program 111 has not been made. That is, the machining knowledge associated with the first NC machining program 111 is considered to be high-quality machining knowledge.

For example, when the reflected machining knowledge is as shown below, since the machining knowledge associated with the first NC machining program 111 and the reflected machining knowledge are all the same, the reward r of the reflected machining knowledge is calculated in step ST200. Increase. The reflected processing knowledge A121a is the reflected processing knowledge generated based on the processing knowledge A121. The reflected machining knowledge B121b is the reflected machining knowledge generated based on the machining knowledge B122. The reflection processing knowledge C121c is reflection processing knowledge generated based on the processing knowledge C123.

(Reflective processing knowledge A121a)
Material Material: S45C
Processing order: When the end chamfering allowance is 20 mm or more, the turning drill step is performed first, and then the turning end surface step is performed. When the edge chamfering margin is other than the above-mentioned conditions, the turning end face step is performed first, and then the turning drill step is performed.
(Reflective processing knowledge B121b)
Material Material: S45C
Machining site: When "horizontal length of open part shape/longitudinal length of open part shape" is "less than 1," the processed part is the front side. When "horizontal length of open part shape/longitudinal length of open part shape" is "1 or more", the outer diameter is defined as the processed part. The open portion is a portion exposed on the surface of the raw material in the turning sectional shape t.
(Reflective processing knowledge C121c)
Material Material: S45C
Machining method: When the length of the lathe cross section in the vertical direction is 10 mm or less and the length of the lathe cross section in the lateral direction is 10 mm or less, the turning grooving step is performed. When the length in the vertical direction and the length in the horizontal direction of the turning cross-sectional shape are other than the above-mentioned conditions, the turning rod material step is performed.

On the other hand, in step ST210, the reward calculation unit 53 calculates a reward r for reducing the reward r of the processing knowledge so that the action value Q with respect to the reward r of the processing knowledge is reduced, and integrates the reward r of the processing knowledge. As a result, the reward r of the processing knowledge is smaller than the reward r when the reference is made in the generation of the first NC processing program 111. As a result, the action value Q of the processing knowledge is reduced.

For example, if the reflected machining knowledge is as shown below, the machining knowledge associated with the first NC machining program 111 and the reflected machining knowledge are all different, so that the reward r of the machining knowledge is reduced in step ST210. The reflected processing knowledge A131 is reflected processing knowledge generated based on the processing knowledge A121. The reflected processing knowledge B132 is the reflected processing knowledge generated based on the processing knowledge B122. The reflected machining knowledge C133 is the reflected machining knowledge generated based on the machining knowledge C123.

(Reflective processing knowledge A131)
Material Material: S45C
Processing order: When the end chamfering allowance is "10 mm or more", the turning drill is performed first, and then the turning end face process is performed. When the edge chamfering margin is other than the above-mentioned conditions, the turning end face step is performed first, and then the turning drill step is performed.
(Reflective processing knowledge B132)
Material Material: S45C
Processed part: When "the lateral length of the open part shape/the vertical length of the open part shape" is "1.0 or less", the processed part is the front side. When "the length of the open portion shape in the horizontal direction/the length of the open portion shape in the vertical direction" is "greater than 1.0", the processed portion is defined as the outer diameter.
(Reflective processing knowledge C133)
Material Material: S45C
Processing method: When the length of the lathe cross-section shape in the vertical direction is 10 mm or less and the length of the lathe cross-section shape in the horizontal direction is "20 mm or less", a turning grooving step is performed. When the length in the vertical direction and the length in the horizontal direction of the turning cross-sectional shape are other than the above-mentioned conditions, the turning rod material step is performed.

When the machining knowledge associated with the first NC machining program 111 and the reflected machining knowledge are different, the edit content for the first NC machining program 111 is made. Therefore, it is considered that the machining knowledge associated with the first NC machining program 111 has a lower quality than the reflected machining knowledge.

On the other hand, the reflected machining knowledge is considered to be machining knowledge of higher quality than the machining knowledge associated with the first NC machining program 111. Therefore, the reward calculation unit 53 calculates the reward r that increases the reward r of the reflected processing knowledge so that the action value Q with respect to the reward r of the reflected processing knowledge increases in step ST200 after step ST210, and the first NC processing program The reward r of the reflected processing knowledge inherited from the processing knowledge associated with 111 is added. Thereby, the reward r of the reflected machining knowledge is larger than the reward r inherited from the machining knowledge associated with the first NC machining program 111. As a result, the action value Q of the reflected processing knowledge increases.

Next, in step ST220, the function updating unit 54 updates the action value function Q(s,a) based on the reward r of the reflected processing knowledge calculated in step ST200 or step ST210, and the first NC processing program 111. An action value Q, which is an evaluation value of the machining knowledge and the reflected machining knowledge associated with, is determined. The function updating unit 54 associates the action value Q defined for the reflected machining knowledge with the reflected machining knowledge and stores it together with the reflected machining knowledge as the machining knowledge.

Further, the function updating unit 54 stores the action value Q defined for the machining knowledge associated with the first NC machining program 111, together with the machining knowledge, as the machining knowledge, in association with the machining knowledge. In this case, the function updating unit 54 has the machining knowledge associated with the first NC machining program 111, the machining knowledge in which the action value Q is determined in step ST220, and the reflected machining knowledge in which the action value Q is determined in step ST220. , With each action value Q being different, it is stored as processing knowledge corresponding to the CAD data 100. Thus, the decision making unit 46 has a relatively high action value Q as the most appropriate machining knowledge in which the knowledge and experience of the worker are appropriately reflected when determining the machining knowledge corresponding to the CAD data 100. Reflective processing knowledge, which is processing knowledge, can be determined.

In this case, the machining knowledge associated with the first NC machining program 111 having the action value Q determined in step ST220 is stored in the function updating unit 54 and referred to in the generation of the first NC machining program 111. May be updated.

After that, the processing from step ST110 to step ST220 is repeatedly executed corresponding to the newly read CAD data 100. Then, when a new reflection machining knowledge is generated, the management number of the base machining knowledge, information on the machining finish shape and information on the material of the material, and accompanying information such as the reward r are inherited.

In addition, when the processes of steps ST110 to ST220 are repeated, the reflected machining knowledge determined by the decision making unit 46 corresponds to the machining knowledge in step ST120 described above.

Further, in the processing of step ST200 and step ST220, high-quality reflected processing knowledge is stored in the function updating unit 54 in a state in which the action value Q is increased. Therefore, step ST210 can be omitted.

Next, details of the NC machining program generation processing in the NC machining program generation unit 43 will be described in detail with reference to FIG. FIG. 5 is a flowchart showing details of the procedure of the NC machining program generation processing in the NC machining program generation unit 43 of the NC machining program generation device 40 according to the first exemplary embodiment of the present invention. FIG. 6 is a perspective view showing an example of a three-dimensional model of a machined finish shape represented by CAD data 100 according to the first embodiment of the present invention. FIG. 7 is a perspective view showing an example of a material shape including the processed and finished shape represented by CAD data 100 according to the first embodiment of the present invention. FIG. 8 is a perspective view showing an example of a machining removal shape generated from the machining finish shape and the material shape represented by the CAD data 100 according to the first embodiment of the present invention. FIG. 9 is a diagram showing the material shape S, the lathe cross-sectional shape generated from the processed shape, and the coordinate values in the first embodiment of the present invention. In the following, a case will be described in which an NC machining program relating to lathe machining in which a material is rotated is machined.

In the NC machining program generation process, in step ST310, the NC machining program generation unit 43 reads the CAD data 100 stored in the machining shape data storage unit 42, and three-dimensionally represented by the CAD data 100 as shown in FIG. The machining finish shape of is generated and placed at the program origin which is the machining origin of the program coordinate system. In the program coordinate system, the Z axis of the XYZ axes is the turning axis SG that is the central axis when performing turning.

The NC machining program generation unit 43 arranges the machining finished shape at the program origin so that the rotation axis of the cylindrical surface or the conical surface having the maximum diameter in the machining finished shape coincides with the Z axis. Move and rotate the finished shape to place it. Further, the end surface of the finished shape in the Z-axis direction is moved so as to coincide with the program origin. That is, the NC machining program generation unit 43 extracts from the CAD data 100 a cylindrical surface or a conical surface having the turning axis SG and the center axis of rotation as the turning end surface. Therefore, the NC machining program generator 43 generates, from the CAD data 100, a three-dimensional machining finished shape with the turning axis as the rotation center axis.

Next, in step ST320, the NC machining program generation unit 43 generates a three-dimensional columnar shape including the machining finish shape arranged at the program origin, as shown in FIG. Then, the NC machining program generation unit 43 arranges the generated cylindrical shape as a material shape in the program coordinate system similarly to the machining finish shape. The material shape is the shape of the material that is machined from the finished shape by turning. That is, the NC machining program generator 43 generates a three-dimensional material shape based on the machining finish shape generated from the CAD data 100.

The dimension of the cylindrical shape including the machined finished shape can be obtained from the maximum value and the minimum value of each of the machined finished shape in the X-axis direction, the Y-axis direction and the Z-axis direction. However, since the turning end face processing is performed on the material shape, the dimension in the Z-axis direction is set to be 2 mm to 3 mm larger than the finished shape in accordance with the finished shape. Further, the end surface of the material shape in the Z-axis direction is arranged at a position protruding from the program origin by 2 to 3 mm in the direction away from the machined shape. Here, the material shape is such that the material outer diameter is 150 mm, the material length is 120 mm, and the material material is S45C.

Next, in step ST330, as shown in FIG. 8, the NC machining program generator 43 generates a machining removal shape, which is a shape to be removed from the material shape by turning, from the machining finish shape and the material shape. In FIG. 8, the finished shape is shown by a broken line. The machining removal shape can be obtained by a difference calculation that subtracts the machining finish solid model from the material shape solid model.

Next, in step ST340, as shown in FIG. 9, the NC machining program generation unit 43 generates a lathe cross-sectional shape t that is the cross-sectional shape of the machining removal shape. As shown in FIG. 9, the turning cross-sectional shape t is a 1/2 turning cross-sectional shape obtained by projecting the machining removal shape onto the +XZ plane which is limited to the +X region of the XZ plane. The turning cross-sectional shape t can be obtained by obtaining the cross-sectional shape by the product operation of the cross-sectional model consisting of the XZ plane and the solid model consisting of the processed shape, and limiting the range as the region of X≧0. As a result, as shown in FIG. 9, the lathe turning sectional shape t1 and the lathe turning sectional shape t2 are generated.

Next, in step ST350, the NC machining program generation unit 43 acquires the machining knowledge corresponding to the CAD data 100 from the function updating unit 54 via the decision making unit 46 and refers to it. The NC machining program generation unit 43 acquires and refers to the above-described machining knowledge A121, machining knowledge B122, and machining knowledge C123 stored in the function updating unit 54, for example.

The NC machining program generation unit 43 acquires, for example, a management number commonly assigned to the CAD data 100 and machining knowledge from the CAD data 100, and transmits it to the decision making unit 46. As a result, the decision making unit 46 can search for a desired machining knowledge designated by the NC machining program generating unit 43, using the management number as a search condition.

Here, the turning end face process is a process of performing a turning end face machining in which the protruding portion of the material end face is scraped off by a turning end face tool. The turning drilling step is a step of performing turning drilling in which a hole along the central axis is bored by a turning drill in a central region in the radial direction of the material. The turning bar material step is a step of performing a turning bar material process of turning an outer circumference, an inner circumference, a front surface or a back surface of a round bar material with a turning tool. The turning grooving step is a step of performing a turning grooving process in which the outer circumference, the inner circumference, the front surface or the back surface of the round bar material is grooved by a turning grooving tool. The end chamfering allowance is a cutting allowance in turning end face machining.

Next, in step ST360, the NC machining program generation unit 43 divides the turning cross-sectional shape t of the machining removal shape into a plurality of different machining steps in accordance with the acquired machining knowledge A121, machining knowledge B122, and machining knowledge C123, and performs the turning machining. Generate a shape. The turning shape is a turning cross-sectional shape t in each processing step when a finished shape is cut out from a material by a plurality of different processing steps.

Hereinafter, the procedure for dividing the turning sectional shape t will be specifically described. FIG. 10: is a figure which shows the turning cross-sectional shape SH1 of the turning end surface process concerning Embodiment 1 of this invention. FIG. 11: is a figure which shows the turning cross-sectional shape SH2 of the turning drill process concerning Embodiment 1 of this invention. FIG. 12 is a diagram showing a lathe cross-sectional shape SH3 in the lathe-bar manufacturing process according to the first embodiment of the present invention. FIG. 13: is a figure which shows the turning cross-sectional shape SH4 of the turning bar material process concerning Embodiment 1 of this invention.

The NC machining program generation unit 43 acquires information on the end chamfering allowance: 10 mm from the turning cross-sectional shape t1. Then, the NC machining program generation unit 43 refers to the machining knowledge A121, determines that the turning end face process is performed first because the end chamfering margin is not 20 mm or more, and as illustrated in FIG. The turning cross-sectional shape SH1 in the end face process is divided.

Next, the NC machining program generation unit 43 refers to the machining knowledge A121, determines that the turning drill process is performed after the turning end face process, and determines from the turning cross-section shape t1 to the turning cross-section of the turning drill process as illustrated in FIG. 11. The shape SH2 is divided.

Next, the NC machining program generation unit 43 divides the turning sectional shape SH3 of the turning bar material process from the turning sectional shape t1 as shown in FIG. Next, the NC machining program generation unit 43 divides the turning cross-sectional shape t2 into the turning cross-sectional shape SH4 of the turning bar material process as shown in FIG.

Next, in step ST370, the NC machining program generation unit 43 allocates a lathe cross sectional shape, which is the divided machining shape, to a lathe machining step according to the machining knowledge A121, the machining knowledge B122, and the machining knowledge C123, and generates a lathe machining step. FIG. 14 is a diagram showing an example of processing start points and processing end points in the turning end face process according to the first exemplary embodiment of the present invention. FIG. 15 is a diagram showing an example of processing start points and processing end points in the turning drill process according to the first exemplary embodiment of the present invention. FIG. 16 is a diagram showing an example of a cutting point, a machining start point, and a machining end point in the turning bar material process according to the first embodiment of the present invention. FIG. 17 is a diagram showing an example of a cutting point, a machining start point, and a machining end point in the turning bar material process according to the first embodiment of the present invention.

The NC machining program generation unit 43 acquires the coordinate value of the machining start point P1 and the coordinate value of the machining end point P2 of the turning end face process from the turning cross-sectional shape SH1 of the turning end face process, as shown in FIG. A turning end face process LC1 that is a process is generated.

Next, the NC machining program generation unit 43, from the lathe cross-sectional shape SH2 of the lathe drill process, as shown in FIG. 15, the hole diameter of the hole to be machined by lathe drilling, the coordinate value of the machining start point P3, and the machining end point P4. The coordinate value is acquired, and a turning drill process LC2, which is a turning drill process, is generated.

Next, the NC machining program generation unit 43, based on the lathe cross-section shape SH3 in the lathe bar process, as shown in FIG. 16, information on the longitudinal length of the open portion shape: 30 mm and the horizontal direction of the open portion shape. Length: Information of 30 mm is acquired. Then, the NC machining program generation unit 43 refers to the machining knowledge B122, and “the length of the open portion shape in the horizontal direction/the length of the open portion shape in the vertical direction” is “1.0”. The machined part in the turning bar process is judged to be the outer diameter. Next, the NC machining program generation unit 43 calculates the coordinate value of the cutting point P5, the coordinate value of the machining start point P6, and the coordinate value of the machining end point P7 of the lathe sectional shape SH3 from the lathe sectional shape SH3 in the lathe bar process. Is obtained, and a turning rod outer diameter step LC3, which is a turning rod step in which the processed portion has an outer diameter, is generated.

Next, the NC machining program generation unit 43, based on the turning cross-section shape SH4 in the turning bar material process, as shown in FIG. 17, the length of the turning cross-section shape SH4 in the vertical direction: 10 mm, the lateral direction of the turning cross-section shape SH4. Length: 20 mm of information is acquired. Then, the NC machining program generation unit 43 refers to the machining knowledge C123, and the turning cross-section shape SH4 has a length in the vertical direction of 10 mm or less, and the turning cross-section shape SH4 does not have a length in the lateral direction of 10 mm or less. Therefore, it is determined that the turning bar material process is performed.

Next, the NC machining program generation unit 43 acquires information on the longitudinal length of the open portion shape: 10 mm and the information on the lateral length of the open portion shape: 20 mm from the lathe cross-section shape SH4 in the lathe bar process. .. Then, the NC machining program generation unit 43 refers to the machining knowledge B122, and "the horizontal length of the open portion shape/the vertical length of the open portion shape" is "2.0" and "1. Since it is not "less than 0", the processed portion is determined to be the outer diameter. Next, the NC machining program generation unit 43 acquires the coordinate values of the cutting point P8, the machining start point P9, and the machining end point P10 of the lathe cross-section shape SH4 from the lathe cross-section shape SH4 in the lathe bar process, and the machining site Produces a turning rod outer diameter step LC4, which is a turning rod step with the outer diameter set to.

The contents of the turning process generated by the NC machining program generator 43 are shown below.
-Turning end face process LC1: Processing start point P1 (75,-10), processing end point P2 (0,0)
-Turning drill process LC2: Hole diameter 30 mm, processing start point P3 (0,0), processing end point P4 (0,110)
-Lathe turning bar outer diameter process LC3: cutting point P5 (75, 0), processing start point P6 (45, 0), processing end point P7 (45, 30)
・Lathe bar outer diameter process LC4: Cutting point P8 (75, 60), processing start point P9 (65, 60), processing end point P10 (65, 80)

By performing the above-described processing, the NC machining program generation unit 43 refers to the machining knowledge A121, the machining knowledge B122, and the machining knowledge C123, and refers to a plurality of different lathe machining processes for machining a machining finish shape from a material. Can be generated automatically. Then, by automatically generating the turning process step by referring to the machining knowledge in which the edited contents reflecting the knowledge and experience of the worker are reflected and accumulated, the case where the operator himself creates the NC machining program A similar turning process step desired by an operator can be automatically generated.

Then, the learning action by the machine learning device 50 according to the first embodiment is continuously performed every time the NC machining program is generated. Therefore, the NC machining program generation device 40 according to the first embodiment is a machining method that is an NC machining program generation method corresponding to the CAD data 100 including the information on the machining finish shape of the machined product and the information on the material quality of the material. Knowledge is continuously learned autonomously. That is, the NC machining program generation device 40 including the machine learning device 50 uses the NC machining program generated by the NC machining program generation unit 43 to edit the NC machining program edited by the NC machining program editing unit 45. Continuously and autonomously learn machining knowledge that reflects knowledge and experience and that has high machining efficiency.

Next, the details of the reflection machining knowledge generation process in which the state observation unit 51 creates the reflection machining knowledge by reflecting the editing content of the first NC machining program 111 on the machining knowledge referred to in the generation of the first NC machining program 111 explain. First, a case will be described in which the first NC machining program 111 is edited such that the turning end face process LC1 is performed after the turning drill process LC2 according to the editing instruction information input by the operator. The state observing unit 51 acquires information on the end chamfering allowance: 10 mm from the turning cross-sectional shape SH1 of the turning end face process LC1 of the second NC machining program 112 transmitted from the NC machining program editing unit 45.

Here, unlike the condition of the machining knowledge A121, the first NC machining program 111 is edited to generate the second NC machining program 112 under the condition of the end chamfering allowance: 10 mm so that the turning drill process is performed first. Therefore, the state observing unit 51 generates the reflected machining knowledge A131 in which the machining knowledge A121 is changed to the following conditions by reflecting the editing content for the first NC machining program 111.

(Reflective processing knowledge A131)
Material Material: S45C
Processing order: When the end chamfering allowance is "10 mm or more", the turning drill is performed first, and then the turning end face process is performed. When the edge chamfering margin is other than the above-mentioned conditions, the turning end face step is performed first, and then the turning drill step is performed.

Next, according to the editing instruction information input by the operator, the machining portion of the turning rod outer diameter process LC3 is changed from the outer diameter to the front face, and the turning rod outer diameter process LC3 is changed to the turning rod front face process LC31. The case where the first NC machining program 111 is edited will be described. The state observing unit 51, from the turning cross-section shape SH3 of the turning bar outer diameter process LC3 of the second NC machining program 112 transmitted from the NC machining program editing unit 45, the information of the length in the vertical direction of the open portion shape: 30 mm, Information of the lateral length of the open part shape: 30 mm is acquired.

Here, unlike the condition of the machining knowledge B122, the machining site is changed to the front face under the condition that "the horizontal length of the open portion shape/the vertical length of the open portion shape" is "1.0". As described above, the first NC machining program 111 is edited to generate the second NC machining program 112. Therefore, the state observing unit 51 reflects the editing content of the first NC machining program 111 to generate the reflected machining knowledge B132 in which the machining knowledge B122 is changed to the following conditions.

(Reflective processing knowledge B132)
Material Material: S45C
Processed part: When "the lateral length of the open part shape/the vertical length of the open part shape" is "1.0 or less", the processed part is the front side. When "the length of the open portion shape in the horizontal direction/the length of the open portion shape in the vertical direction" is "greater than 1.0", the processed portion is defined as the outer diameter.

Next, a case will be described in which the first NC machining program 111 is edited so as to change the turning rod outer diameter step LC4 to the turning grooving step LC41 according to the editing instruction information input by the operator. The state observing unit 51 transmits information from the turning cross-section shape SH4 of the turning bar outer diameter process LC4 of the second NC machining program 112 transmitted from the NC machining program editing unit 45 to the longitudinal length of the turning cross-section shape SH4: 10 mm. , Information on the lateral length of the turning cross-section shape SH4: 20 mm is acquired.

Here, unlike the condition of the machining knowledge C123, the turning grooving process LC41 is performed under the condition that "the longitudinal length of the turning cross sectional shape SH4 is 10 mm and the lateral length of the turning cross sectional shape SH4 is 20 mm". The first NC machining program 111 is edited so that the second NC machining program 112 is generated. Therefore, the state observing unit 51 generates the reflected machining knowledge C133 in which the machining knowledge C123 is changed to the following condition by reflecting the editing content for the first NC machining program 111.

(Reflective processing knowledge C133)
Material Material: S45C
Machining method: When the length of the lathe cross-section shape in the vertical direction is "10 mm or less" and the length of the lathe cross-section shape in the lateral direction is "20 mm or less", a turning grooving step is performed. When the length in the vertical direction and the length in the horizontal direction of the turning cross-sectional shape are other than the above-mentioned conditions, the turning rod material step is performed.

Next, the division procedure of the turning cross-sectional shape t when the NC machining program generation unit 43 generates the NC machining program with reference to the reflected machining knowledge A131, the reflected machining knowledge B132, and the reflected machining knowledge C133 will be specifically described. FIG. 18 is a diagram showing a turning cross-sectional shape SH5 in the turning drill process according to the first embodiment of the present invention. FIG. 19 is a diagram showing a lathe cross-sectional shape SH6 in the lathe end face process according to the first embodiment of the present invention. FIG. 20 is a diagram showing a turning cross-sectional shape SH7 in the turning rod material step according to the first embodiment of the present invention. FIG. 21 is a diagram showing a lathe cross-sectional shape SH8 in the lathe-bar manufacturing process according to the first embodiment of the present invention.

The NC machining program generation unit 43 acquires information on the material length: 120 mm and information on the material outer diameter: 150 mm from the material shape, and acquires information on the end chamfering margin: 10 mm from the turning cross-sectional shape t1. Then, the NC machining program generation unit 43 refers to the reflected machining knowledge A131, determines that the turning drill step is to be performed first because the end chamfering margin is 10 mm or more, and as shown in FIG. Then, the turning cross-sectional shape SH5 in the turning drilling process is divided.

Next, the NC machining program generation unit 43 refers to the reflected machining knowledge A131, determines that the turning end face process is to be performed after the turning drill process, and performs turning from the turning sectional shape t1 to the turning end face process as shown in FIG. The sectional shape SH6 is divided.

Next, the NC machining program generation unit 43 divides the lathe sectional shape SH7 in the lathe bar process from the lathe sectional shape t1 as shown in FIG. Next, the NC machining program generation unit 43 divides the turning cross-sectional shape t2 into the turning cross-sectional shape SH8 of the turning rod material process as shown in FIG.

Next, a procedure for assigning a turning machining step when the NC machining program generating unit 43 generates an NC machining program with reference to the reflected machining knowledge A131, the reflected machining knowledge B132, and the reflected machining knowledge C133 will be specifically described. FIG. 22 is a diagram showing an example of processing start points and processing end points in the turning drill process according to the first exemplary embodiment of the present invention. FIG. 23 is a diagram showing an example of a machining start point and a machining end point in the turning end face process according to the first embodiment of the present invention. FIG. 24 is a diagram showing an example of a cutting point, a processing start point, and a processing end point in the turning bar front surface process according to the first embodiment of the present invention. FIG. 25 is a diagram showing an example of a cutting point, a machining start point, and a machining end point in the turning rod material grooving step according to the first exemplary embodiment of the present invention.

The NC machining program generation unit 43 calculates, from the turning cross-sectional shape SH5 of the turning drilling process, the hole diameter of the hole to be machined by the turning drilling, the coordinate value of the machining start point P11, and the coordinate value of the machining end point P12 as shown in FIG. And a turning drill process LC11, which is a turning drill process, is generated.

Next, the NC machining program generation unit 43 acquires the coordinate values of the machining start point P13 and the machining end point P14 of the turning end face process from the turning cross-sectional shape SH6 of the turning end face process, as shown in FIG. , A turning end face process LC12, which is a turning end face process.

Next, the NC machining program generation unit 43, based on the lathe cross-section shape SH7 in the lathe bar process, as shown in FIG. 24, information on the longitudinal length of the open portion shape: 30 mm and the horizontal direction of the open portion shape. Length: Information of 30 mm is acquired. Then, the NC machining program generation unit 43 refers to the reflected machining knowledge B132, and "the horizontal length of the open portion shape/the vertical length of the open portion shape" is "1.0" and "1". Since it is "0 or less", the processed portion in the turning bar material process is determined to be the front surface. Next, the NC machining program generation unit 43 calculates the coordinate value of the cutting point P15, the coordinate value of the machining start point P16, and the coordinate value of the machining end point P17 of the turning sectional shape SH7 from the turning sectional shape SH7 of the turning bar material process. And a turning bar front face process LC3, which is a turning bar rod process in which the processed portion is the front face, is generated.

Next, the NC machining program generation unit 43, based on the turning cross-section shape SH8 of the turning bar material process, as shown in FIG. 25, the length of the turning cross-section shape SH8 in the vertical direction: 10 mm, the lateral direction of the turning cross-section shape SH8. Length: 20 mm of information is acquired. Then, the NC machining program generation unit 43 refers to the reflected machining knowledge C133, and the turning cross-sectional shape SH4 has a length of 10 mm or less in the vertical direction and the latitudinal cross-sectional shape SH4 has a length of 20 mm or less in the lateral direction. Therefore, it is determined that the turning grooving process is performed.

Next, the NC machining program generation unit 43 acquires information on the longitudinal length of the open portion shape: 10 mm and the information on the lateral length of the open portion shape: 20 mm from the lathe cross-section shape SH8 in the lathe bar process. .. The NC machining program generation unit 43 obtains the information of the groove depth: 10 mm from the information of the longitudinal length of the open portion shape: 10 mm, and the information of the lateral length of the open portion shape: 20 mm, the groove width: 20 mm. Get information. Then, the NC machining program generation unit 43 acquires the coordinate value of the machining start point P18 from the lathe cross sectional shape SH8 of the lathe bar process, and generates the lathe grooving process LC14.

The contents of the turning process generated by the NC machining program generation unit 43 with reference to the reflected machining knowledge A131, the reflected machining knowledge B132, and the reflected machining knowledge C133 are shown below.
-Turning drill process LC11: Hole diameter 30 mm, processing start point P11 (0,-10), processing end point P12 (0,110)
-Turning end face process LC12: processing start point P13 (75,-10), processing end point P14 (15,0)
-Lathe turning bar front process LC13: cutting point P15 (75,0), processing start point P16 (75,30), processing end point P17 (45,30)
• Turning grooving process LC14: groove width 20 mm, groove depth 10 mm, processing start point P18 (75, 60)

By performing the above-described processing, the NC machining program generation unit 43 refers to the reflected machining knowledge A131, the reflected machining knowledge B132, and the reflected machining knowledge C133 to refer to a plurality of different turnings for cutting a machining finish shape from a material. Processing steps can be automatically generated. In the case where the worker himself/herself creates the NC machining program by automatically generating the turning machining process by referring to the reflected machining knowledge in which the edited contents reflecting the knowledge and experience of the worker are reflected and accumulated. A turning processing step desired by the operator, which is similar to, can be automatically generated.

The decision making unit 46 uses the result learned by the machine learning device 50 to decide the machining knowledge or the reflected machining knowledge corresponding to the request from the NC machining program generating unit 43. By the way, the NC machining program may be edited by the NC machining program generation unit 43 after the NC machining program is generated once. Since the machine learning device 50 according to the present embodiment has already learned the reflected machining knowledge, the decision making unit 46 determines the most appropriate reflected machining knowledge corresponding to the request from the NC machining program generation unit 43 as a function. It can be determined as necessary from the processing knowledge stored in the updating unit 54.

Further, the machine learning device 50 can learn the relationship between the material, the CAD data 100, and the processing knowledge. Therefore, when it is estimated that the NC machining program is edited due to the material of the material and the CAD data 100, the machining knowledge or the reflected machining knowledge determined by the decision making unit 46 may be switched. Good.

The reflection processing knowledge may be stored in association with the CAD data 100 itself. For example, the CAD data 100 is associated with the reflected processing knowledge A131 and stored in the function updating unit 54 together with the processing knowledge A121.

Further, the reflection processing knowledge may be stored by associating information associated with the CAD data 100. For example, the information associated with the CAD data 100 is associated with the reflection processing knowledge A131 and stored together with the reflection processing knowledge A131. Examples of information associated with the CAD data 100 are an image of CAD data, a part of CAD data, and the like. The information associated with the CAD data 100 can be rephrased as the information associated with the combination of the information on the finished shape of the machined product and the information on the material of the workpiece.

As a result, the decision making unit 46 searches for and obtains the desired reflected machining knowledge from the plurality of machining knowledge stored in the function updating unit 54, using the CAD data 100 or information associated with the CAD data 100 as a search condition. It becomes possible to do.

Further, the machining knowledge and the reflected machining knowledge may be stored separately for each NC machine tool controlled by executing the NC machining program. When a plurality of NC machine tools are used, for example, NC machine tools of different models are used, even if NC machine tools of the same model are used, different optional functions are added for each NC machine tool. The function often differs for each NC machine tool due to reasons such as the fact that the NC machine tool is used. In this case, the machining knowledge also has a condition suitable for each NC machine tool.

Therefore, by storing the machining knowledge and the reflected machining knowledge separately for each NC machine tool, the edited contents reflecting the knowledge and experience of the operator suitable for each NC machine tool are reflected and accumulated in the reflected machining knowledge. To be done. When the NC machining program generation unit 43 generates the NC machining program by referring to the machining knowledge or the reflected machining knowledge managed separately for each NC machine tool, and the operator himself creates the NC machining program. Similarly to the above, it is possible to quickly and automatically generate an NC machining program suitable for each NC machine tool desired by the operator.

It is also possible to store the processing knowledge and the reflected processing knowledge in a storage unit provided on the network server.

As a modification of the machine learning device 50 according to the first embodiment, the machine learning device 50 may execute machine learning according to a neural network model. FIG. 26 is a schematic diagram showing an example of a neural network model that the machine learning device 50 according to the first exemplary embodiment of the present invention follows. The neural network is composed of, for example, an input layer including l input neurons, an intermediate layer including m neurons, and an output layer including n output neurons. FIG. 26 shows an example of a neural network model including an input layer including four input neurons, an intermediate layer including five neurons, and an output layer including one output neuron. Although only one intermediate layer is shown in FIG. 26, two or more intermediate layers may be provided.

Neural networks learn the relationship between state variables and environmental changes so that optimal behavior is intuitively learned based on human success or failure experiences. The neural network learns the relationship between the state variable and the environmental change according to the data set created based on the state variable observed by the state observation unit 51. That is, the neural network learns the relationship between state variables and environmental changes by so-called supervised learning. In this case, in the machine learning device 50, in response to the machining shape data input to the input layer of the neural network and the generation history of the NC machining program, the output layer is the machining knowledge corresponding to the machining shape data, Outputs processing knowledge that better reflects the knowledge and experience of operators. Then, the decision making unit 46 functions to decide the optimum NC machining program generation method. That is, the machine learning device 50 learns the relationship between the machining shape data, the NC machining program generation method, and the edit content of the first NC machining program 111 even when the neural network model is used, and generates the NC machining program. It is possible to learn how.

In the above description, the embodiment in which the machine learning device 50 performs machine learning using reinforcement learning or a neural network has been described. However, the machine learning device 50 can be used in another known method such as genetic programming or functional logic. Machine learning may be performed according to programming, support vector machines, etc.

Further, as an embodiment of the learning of the learning unit 52, the learning unit 52 can learn the machining knowledge applied to the NC machining program used in another numerical control device. That is, the learning unit 52, based on the machining knowledge used to generate the NC machining program generated for another numerical control device, and the content of the editing for the NC machining program, with respect to the certain machining shape data, the learning unit 52 determines another numerical control device. May be configured to learn machining knowledge applied to the NC machining program used in. The learning unit 52 may learn the machining knowledge applied to the NC machining program used in another numerical control device that operates at the same site, and may use another numerical control device that operates independently at a different site. You may learn the processing knowledge applied to the NC processing program used.

FIG. 27 is a diagram showing a hardware configuration of the NC machining program generation device 40 according to the first exemplary embodiment of the present invention. The hardware shown in FIG. 27 includes a processor 201 that performs arithmetic processing, a memory 202 used by the processor 201 as a work area, a storage device 203 that stores a program for operating as a numerical control device or a program conversion device, and work. An input device 204 which is an input interface with a worker, a display device 205 for displaying information to an operator, a communication device 206 having a communication function with a controlled device or another numerical control device, and other various devices. , Is provided. The processor 201, memory 202, storage device 203, input device 204, display device 205, and communication device 206 are connected by a data bus 207. Here, the processor 201 may be a processing device, a computing device, a microprocessor, a microcomputer, a CPU (Central Processing Unit), a DSP (Digital Signal Processor), or the like. The memory 202 is a nonvolatile or volatile semiconductor such as a RAM (Random Access Memory), a ROM (Read Only Memory), a flash memory, an EPROM (Erasable Programmable ROM), or an EEPROM (registered trademark) (Electrically EPROM). It corresponds to a memory, a magnetic disk, a flexible disk, an optical disk, a compact disk, a mini disk, a DVD (Digital Versatile Disc), or the like.

The NC machining program generation unit 43 is realized by the processor 201 executing the program stored in the memory 202 illustrated in FIG. 27, for example. Further, a plurality of processors and a plurality of memories may cooperate to realize the above function. Further, some of the functions of the NC processing program generation unit 43 may be mounted as an electronic circuit, and the other portions may be realized using the processor 201 and the memory 202.

In addition, the machining shape data input unit 41, the NC machining program editing unit 45, the decision making unit 46, the state observing unit 51, the reward calculating unit 53, and the function updating unit 54, and the program stored in the memory 202 in the processor 201. May be realized by executing. Further, a plurality of processors and a plurality of memories may cooperate to realize the above function. Further, some of the functions of the machining shape data input unit 41, the NC machining program editing unit 45, the decision making unit 46, the state observing unit 51, the reward calculating unit 53, and the function updating unit 54 are implemented as an electronic circuit, Other parts may be realized using the processor 201 and the memory 202. Further, the processor and the memory for realizing the functions of the machining shape data input unit 41, the NC machining program editing unit 45, the decision making unit 46, the state observing unit 51, the reward calculating unit 53, and the function updating unit 54 are NC machining. It may be the same as the processor and memory that implement the program generation unit 43, or may be a different processor and memory.

The NC machining program generation device 40 according to the first embodiment described above is based on the CAD data 100, which is machining shape data, and the machining contents in which the knowledge and experience of the operator are reflected and accumulated. By automatically generating the turning process with reference to the knowledge, it is possible to automatically generate the turning process desired by the operator, which is the same as when the operator himself creates the NC processing program.

Further, in the NC machining program generation device 40 including the machine learning device 50 according to the first embodiment, the NC machining corresponding to the CAD data 100 including the information on the machining finish shape of the machined product and the information on the material quality of the material. Autonomously learn machining knowledge, which is a method of program generation. That is, the NC machining program generation device 40 including the machine learning device 50 uses the NC machining program generated by the NC machining program generation unit 43 to edit the NC machining program edited by the NC machining program editing unit 45. Autonomously learn machining knowledge with high machining efficiency that reflects knowledge and experience. Thereby, the decision making unit 46 corresponds to the CAD data 100 from the machining knowledge stored in the function updating unit 54 when the CAD data 100 is input again to the NC machining program generating device 40. It is possible to determine the processing knowledge having the highest action value Q and the highest processing efficiency based on the learning result.

Therefore, in the NC machining program generation device 40 according to the first embodiment, even if the knowledge or know-how of the expert is not accumulated, the NC machining is close to the turning process desired by the operator and has high machining efficiency and high quality. Allows automatic generation of programs. Further, since it is not necessary for the worker to generate the NC machining program by trial and error, the work efficiency of creating the NC machining program is improved. As a result, the NC machining program generation device 40 according to the first embodiment can obtain the effect of reducing the time required for automatically generating the NC machining program and the effect of reducing the machining time of the workpiece.

Further, since the NC machining program generation device 40 according to the first embodiment learns the NC machining program editing content of the user and reflects it in the machining knowledge, the NC machining program editing work of the user is reduced when the next NC program is generated. be able to.

Although various embodiments of the present invention have been described above, those skilled in the art will recognize that the intended effects of the present invention can be realized by other embodiments. In particular, the constituent elements of the first embodiment described above can be deleted or replaced without departing from the scope of the present invention, or known means can be further added. Further, it is obvious to those skilled in the art that the present invention can be implemented by arbitrarily combining the features of the embodiments explicitly or implicitly disclosed in this specification. That is, the configurations shown in the above embodiments are examples of the content of the present invention, and can be combined with another known technique, and the configurations are within the scope not departing from the gist of the present invention. It is also possible to omit or change a part of.

1 Numerical Control Device, 10 Interactive Operation Processing Unit, 20 Display Unit, 30 Instruction Input Unit, 40 NC Machining Program Generation Device, 41 Machining Shape Data Input Unit, 42 Machining Shape Data Storage Unit, 43 NC Machining Program Generation Unit, 44 NC Machining program storage unit, 45 NC machining program editing unit, 46 decision making unit, 50 machine learning device, 51 state observing unit, 52 learning unit, 53 reward calculating unit, 54 function updating unit, 100 CAD data, 111 first NC machining program , 112 2nd NC machining program, 121 machining knowledge A, 122 machining knowledge B, 123 machining knowledge C, 131 reflection machining knowledge A, 132 reflection machining knowledge B, 133 reflection machining knowledge C, 201 processor, 202 memory, 203 storage device, 204 input device, 205 display device, 206 communication device, 207 data bus.

In order to solve the above-mentioned problems and achieve the object, a machine learning apparatus according to the present invention is provided with machining shape data including information on a machining finish shape of a machined product and information on a material of a material, and cutting from a workpiece. A numerical control machining program generation method for automatically generating a numerical control machining program including a plurality of cutting processes for carving a workpiece, and a first numerical control generated by referring to the numerical control machining program generation method A state observing unit that observes the edited contents of the machining program as a state variable, and a learning unit that autonomously learns the method of generating the numerically controlled machining program according to a data set created based on the state variable.

In order to solve the above-mentioned problems and achieve the object, a machine learning apparatus according to the present invention is provided with machining shape data including information on a machining finish shape of a machined product and information on a material of a material, and cutting from a workpiece. A numerical control machining program generation method for automatically generating a numerical control machining program including a plurality of cutting processes for carving a workpiece, and a first numerical control generated by referring to the numerical control machining program generation method comprising the edits for the machining program, and the state observing unit for observing the state variables, according to the data set that is created based on the state variable, and a learning unit that learns a method of generating numerical control machining program, the. The learning unit includes a reward calculation unit that calculates a reward based on the reflected numerical control machining program generation method in which the edited content is reflected in the numerical control machining program generation method, and a numerical control machining program based on the calculated reward. And a function updating unit that updates a function for determining the generation method of.

Claims (5)

To automatically generate a machining shape data including machining finish shape information and material quality information of a cutting work product, and a numerical control machining program including a plurality of cutting processes for cutting out the cutting work product from a workpiece. And a state observing section for observing, as a state variable, the numerical control machining program generation method described above, and the edit content for the first numerical control machining program generated by referring to the numerical control machining program generation method.
A learning unit for learning a method of generating the numerically controlled machining program according to a data set created based on the state variables;
A machine learning device comprising: The learning unit
A reward calculation unit that calculates a reward based on the reflected numerically controlled machining program generation method in which the edited contents are reflected in the numerically controlled machining program generation method,
A function updating unit for updating a function for determining a method of generating the numerical control machining program based on the calculated reward;
The machine learning device according to claim 1, further comprising: The reward calculator is
When the generation method of the numerical control machining program and the generation method of the reflection numerical control machining program are the same, the reward of the generation method of the reflection numerical control machining program is increased, and the generation method of the numerical control machining program and the reflection When the generation method of the numerical control machining program is different, it is configured to reduce the reward of the generation method of the numerical control machining program and increase the reward of the generation method of the reflected numerical control machining program,
The machine learning device according to claim 2. A machine learning device according to any one of claims 1 to 3,
A numerical control machining program generation unit for generating the first numerical control machining program by referring to a method of generating the numerical control machining program determined based on a result learned by the machine learning device,
A numerical control machining program editing unit for editing the first numerical control machining program to generate a second numerical control machining program;
A decision making unit that decides a method of generating a numerically controlled machining program corresponding to the machining shape data, based on a result learned by the learning unit of the machine learning device;
A numerically controlled machining program generation device comprising: A method of generating a numerically controlled machining program used in automatic generation of the numerically controlled machining program in a numerically controlled machining program generation device for generating a numerically controlled machining program including a plurality of cutting processes for cutting a machined product from a workpiece A machine learning method for learning
The machining shape data including the machining finish shape information and the material material information of the machined product, the numerical control machining program generation method, and the numerical control machining program generation method are referred to to refer to the numerical control machining program. Editing the first numerically controlled machining program generated by the generator and observing the edited contents as state variables;
Learning a method of generating the numerically controlled machining program according to a data set created based on the state variables;
A machine learning method comprising:

Images