Device and system for setting machining conditions

Abstract

Machining condition setting device (1) for setting a machining condition of a machine tool for machining a workpiece, wherein the machining condition setting device (1) comprises: a data acquisition unit (34) for obtaining at least one data set indicating a state of machining including a machining type in the machine tool; a priority condition storage unit (56) which stores priority condition data in which the machining type in the machine tool is linked to a priority condition for the machining type; a preprocessing unit (36) which generates data for use in machine learning based on data acquired by the data acquisition unit (34) and the priority condition according to the processing type, which is contained in the data and stored in the priority condition storage unit (56); and a machine learning device (100) which performs a machine learning process with respect to a machining condition and / or a machining parameter for machining by the machine tool in an environment in which the workpiece is machined by the machine tool, based on data generated by the preprocessing unit (36), wherein the machine learning device (100) contains: a learning model storage unit (130) which stores multiple learning models generated for each machining operation in the machine tool, and a learning model selection unit (105) which selects a learning model from the several learning models stored in the learning model storage unit (130) for use in the machine learning process based on the processing type contained in the data which the preprocessing unit (36) has generated.

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention

[0001] The invention relates to a machining condition setting device and a machining condition setting system. 2. State of the art

[0002] The optimization of molding conditions in an injection molding machine using machine learning is known in the prior art (see JP 2017 - 30 152 A as an example). When a mold is manufactured for a new molded product, the corresponding injection molding machine must be adjusted to optimal injection molding conditions. The technology described in the prior art according to JP 2017 - 30 152 A and similar publications is capable of adjusting the injection molding conditions for the specific injection molding process using a machine learning device.

[0003] When a workpiece is machined with a cutting operation in a so-called machining center, the machining center performs several types of machining, such as rough machining, drilling, tapping, and finishing. In such a machine tool, the details required for each type of machining vary. Rough machining generally requires short cycle times and high machining efficiency, while allowing for minor errors. In finishing, on the other hand, shape precision is paramount, even if the cycle time is longer.

[0004] When machining with machine tools, the parameters required for each type of machining are therefore different. Optimization can also depend on the operator. Optimal settings also vary depending on factors such as the installation environment of the machine tool used for machining.

[0005] Further prior art is disclosed in DE 10 2018 007 642 A1. BRIEF DESCRIPTION OF THE INVENTION

[0006] The object is therefore to provide a machining condition setting device and a machining condition setting system that adjust the machining conditions and / or machining parameters while taking into account the details required according to the type of machining of a workpiece in a machine tool. This object is achieved by a machining condition setting device having the features of patent claim 1 and by a machining condition setting system having the features of patent claim 7.

[0007] A variant of the invention is a machining condition setting device that sets a machining condition and / or a machining parameter for a machine tool for machining a workpiece. The machining condition setting device includes: a data acquisition unit that acquires at least one data set indicating a machining state including a machining type in the machine tool; a priority condition storage unit that stores priority condition data in which the machining type in the machine tool is linked to a priority condition for the machining type; a preprocessing unit that generates data for use in machine learning based on data acquired by the data acquisition unit and the priority condition corresponding to the machining type contained in the data and stored in the priority condition storage unit;and a machine learning device that performs the machine learning process regarding a machining condition and / or a machining parameter for machining by the machine tool in an environment where the workpiece is machined by the machine tool, based on data generated by the preprocessing unit. The machine learning device includes: a learning model storage unit that stores a plurality of learning models generated for each machining type in the machine tool; and a learning model selection unit that selects a learning model for use in machine learning from the plurality of learning models stored in the learning model storage unit, based on the machining type contained in the data generated by the preprocessing unit.

[0008] A variant of the invention is a machining condition setting system having a plurality of devices connected to each other via a network, wherein the plurality of devices include a machining condition setting device having at least one learning unit.

[0009] According to a variant of the invention, the machining condition can be set according to a machining type for a workpiece in the machine tool and taking into account the conditions required for the respective machining type. BRIEF DESCRIPTION OF THE CHARACTERS

[0010] The above and other objects and features of the invention will become even clearer from the following description of embodiments with reference to the figures: Fig. 1 schematically shows the device structure of a machining condition setting device according to an embodiment; Fig. 2 is a schematic functional block diagram of a machining condition setting apparatus according to a first embodiment; Fig. 3 describes priority condition data; Fig. 4 is a schematic functional block diagram of a machining condition setting apparatus according to a second embodiment; Fig. 5 is a schematic functional block diagram of a machining condition setting device according to a third embodiment; Fig. 6 is a schematic functional block diagram of a machining condition setting apparatus according to a fourth embodiment; Fig. 7 is a schematic functional block diagram of a machining condition setting apparatus according to a fifth embodiment; Fig. Figure 8 shows an example of a system with a three-layer structure including a cloud server, a fog computer, and an edge computer; Fig. 9 schematically shows the device structure of a machining condition setting device in an implementation in a computer; Fig. 10 schematically shows the configuration of a machining condition setting system according to a sixth embodiment; Fig. 11 schematically shows the configuration of a machining condition setting system according to a seventh embodiment; and Fig. 12 schematically shows the configuration of a machining condition setting system according to an eighth embodiment. DESCRIPTION OF PREFERRED EMBODIMENTS IN DETAIL

[0011] In the following, embodiments of the invention are described in more detail with reference to the figures.

[0012] Fig. 1 schematically shows the hardware configuration for explaining essential parts of the machining condition setting device including a machine learning device according to an embodiment. A machining condition setting device 1 according to the embodiment can be implemented in a control device that controls, for example, a machine tool. The machining condition setting device 1 according to the embodiment can be implemented in a personal computer in conjunction with a control device that controls a machine tool, or in a computer, such as a data center, an edge computer, a fog computer, or a cloud server, which are connected to the control device via a wired / wireless network. In the embodiment, the machining condition setting device 1 is implemented in a control device that controls a machine tool.

[0013] A CPU (central processing unit) 11 is included in the machining condition setting device 1 and, in the present embodiment, is a processor that controls the machining condition setting device 1 as a whole. The CPU 11 reads system programs stored in a ROM (read-only memory) 12 via a bus 20 and controls the entire machining condition setting device 1 according to the system programs. Temporarily generated calculation data, display data, various types of data input by an operator via an input unit (not shown), etc., are temporarily stored in a RAM (random access memory) 13.

[0014] A non-volatile memory 14 is configured as a memory whose storage state is maintained by a battery (not shown) or the like even when the machining condition setting device 1 is turned off. Programs read from external devices 72 via an interface 15 or programs input via a display / MDI unit 70 are stored in the non-volatile memory 14.The non-volatile memory 14 stores various types of data (tool information such as tool types; cutting condition information such as spindle speed, feed rate, and cutting depth; workpiece information such as workpiece materials and shapes; information regarding power consumed by each motor; dimensional values ​​and surface qualities of portions of the machined workpiece; and temperatures of portions of the machine tool as measured by sensors 3 or the like) obtained from units of the machining condition setting device 1, the machine tool, the sensors 3, or the like. The programs or various types of data stored in the non-volatile memory 14 can be loaded into the RAM 13 at the time of program execution or data use.Various system programs (including system programs for controlling interaction with the machine learning device 100, which will be described in more detail below), such as well-known analysis programs, are pre-written in the ROM 12.

[0015] The interface 15 connects the machining condition setting device 1 to the external equipment 72, such as an adapter. Programs or various parameters are read from the external equipment 72. The programs or various parameters edited by the machining condition setting device can be stored in external storage devices via the external equipment 72. A programmable machine controller (PMC) 16 inputs and outputs signals via an input / output (I / O) unit 17, respectively, to a machine tool, a robot, and devices, and the sensors 3 installed on or in the machine tool or robot, to execute control according to a sequence program stored in the machining condition setting device 1.

[0016] The machining condition setting device 1 is connected to sensors 3, such as a contact distance sensor or a non-contact distance sensor, an image pickup device, and a surface roughness measuring instrument, each of which is used when machining a workpiece with the machine tool. The sensors 3 are used to measure design data errors, profile errors, or the like in areas of the workpiece machined by the machine tool.

[0017] The display / MDI unit 70 has manual data input, including a display, a keyboard, or the like. An interface 18 receives commands or data from the keyboard of the display / MDI unit 70 and transmits the commands or data to the CPU 11. The interface 19 is connected to an operation panel 71 including a manual pulse generator or the like, which is used when axes are moved manually.

[0018] Axis control circuits 30 control the axes of the machine tool and receive movement distance commands for the axes from the CPU 11 and issue commands regarding the axes to servo amplifiers 40. The servo amplifiers 40 receive the commands and drive servo motors 50, which move the axes of the machine tool. The servo motors 50 for the axes contain position / speed detectors, output position / speed signals from the detectors back to the axis control circuits 30, and thus perform feedback control (regulation) regarding positions / speeds. Although the axis control circuits 30, the servo amplifiers 40, and the servo motors 50 are each only one-piece in the hardware configuration according to Fig. 1, however, in fact, the number of axis control circuits 30, servo amplifier 40 and servo motor 50 each corresponds to the number of axes in the machine tool to be controlled (three each for a machine tool with three linear axes or five for a five-axis machine tool, as examples).

[0019] A spindle control circuit 60 receives a spindle rotation command for a spindle of the machine tool and outputs a spindle speed signal to the spindle amplifier 61. The spindle amplifier 61 receives the spindle speed signal, rotates a spindle motor 61 for the spindle at a rotation speed according to the command, and thus drives a tool. The position encoder 63 is coupled to the spindle motor 62 and outputs feedback pulses synchronized with the rotation of the spindle. The feedback pulses are read by the CPU 11.

[0020] An interface 21 connects the machining condition setting device 1 and the machine learning device 100. The machine learning device 100 includes a processor 101 that controls the machine learning device 100 as a whole, a ROM 102 that stores system programs and the like, a RAM 103 that stores data temporarily generated in processes related to machine learning, and a non-volatile memory 104 that stores a learning model or the like.The machine learning device 100 is capable of processing information acquired in the machining condition setting device 1 and input via the interface 61 (information regarding tools such as the types of tools; information regarding cutting conditions such as spindle speed, spindle feed, and cutting depth; information regarding the workpiece such as the materials and shapes of the workpiece, the power consumed by the respective motors; information regarding the dimensional values ​​and surface quality of the portions of the machined workpiece; and the temperatures of portions of the machine tool as measured by the sensors or the like, each as examples).The machining condition setting device 1 receives information output from the machine learning device 100 and executes control of the machine tool based on data from the display / MDI unit 70, transmission of information from other devices via a network (not shown), or the like.

[0021] Fig. Figure 2 is a functional block diagram for explaining the machining condition setting device 1 and the machine learning device 100 according to a first embodiment. The machining condition setting device 1 of this embodiment includes a configuration for a learning device as required for the machine learning device 100 using so-called reinforcement learning. The functional blocks according to Fig. 1 are implemented by executing respective system programs and by controlling the operation of the units of the machining condition setting device 1 and the machine learning device 100 by means of the CPU 11 in the machining condition setting device 1 according to Fig. 1 and by the processor 101 of the machine learning device 100.

[0022] The machining condition setting device 1 of this embodiment includes a control unit 32, a machining type determination unit 33, a data acquisition unit 34, a preprocessing unit 36 ​​and a priority condition setting unit 37, and the machine learning device 100 in the machining condition setting device 1 includes a learning model selection unit 105, a learning unit 110 and a decision unit 122. In the non-volatile memory 14 according to Fig. 1, an acquired data storage unit 52 is provided, in which data acquired from the machine tool 2, the sensors 3, and the like are stored, and a priority condition storage unit 56 is provided, in which priority condition data set by the priority condition setting unit 37 is stored. In the non-volatile memory 104 of the machine learning device 100 according to Fig. 1, a learning model storage unit 130 is arranged, which stores a learning model formed by machine learning by the learning unit 110.

[0023] The control of the machine tool 2 by means of the control unit 32 is carried out by executing system programs which are read from the ROM 12 and executed by the CPU 11 which is installed in the machining condition setting device 1 according to Fig. 1, further by calculation processes mainly by the CPU 11 using the RAM 13 and the non-volatile memory 14 and by controlling the machine tool 2 and the sensors 3 via the axis control circuits 30, the spindle control circuit 60 and the PMC 16. The control unit 32 controls the machining operation by the machine tool 2 and the measuring operation by the sensors 3 on the basis of a control program 54 which is stored in the non-volatile memory 14 according to Fig. 1 is stored. The control unit 32 has functions for the overall control as required for the individual areas of the machine tool 2, such as the output of movement commands in each control cycle to the servo motors 50 ( Fig. 1), which move the axes of the machine tool 2, and to the spindle motor 62 ( Fig. 1) according to the control program 54. Furthermore, the control unit 32 outputs commands for executing the measuring operation by the sensors 3. Furthermore, the control unit 32 receives data regarding a state and a result of machining by the machine tool 2, which data are obtained from the machine tool 2 and the sensors 3, and outputs corresponding data to the data acquisition unit 34.Examples of the data acquired by the control unit 32 from the machine tool 2 and the sensors 3 and supplied to the data acquisition unit 34 include information regarding tools such as the types of tools, information regarding cutting conditions such as spindle speed, feed rate and cutting depth, information regarding machining parameters, information regarding workpieces such as their materials and shapes, information regarding the power consumed by the respective motors and information regarding the temperatures of portions of the machine tool, information regarding machining results such as dimensions and shape errors of machined workpiece portions and the like, each by way of example.

[0024] When machining conditions and / or machining parameters are output from the machine learning device 100, the control unit 32 controls the machining operation in the machine tool 2 using the machining conditions or machining parameters output from the machine learning device 100, rather than based on the machining conditions or machining parameters according to the instructions of the control program 54 or the like.

[0025] The determination of a machining type by the machining type determination unit 33 is carried out by executing system programs executed by the CPU 11 in the machining condition setting device 1 according to Fig. 1 are read from the ROM 12, and based on arithmetic operations mainly by the CPU 11 using the RAM 13 and the non-volatile memory 14. The machining type determining unit 33 determines the machining type of the machining being performed by the machine tool 2 according to the command from the control unit 32. The machining type determining unit 33 can analyze the control program 54 currently being executed by the control unit 32, and thereby determine, for example, the machining type of the machining being performed based on commands currently being executed, commands executed before or after, etc.Thus, the machining type can be determined based on the type of tool currently being used (tapping when using a tap, drilling when using a drill, or the like), further based on a machining condition (control code in the command, rough machining at a cutting depth equal to or greater than a given threshold, finishing at a cutting depth less than the threshold, or the like), or a comment in the control program 54 or the like, each for example. On the other hand, the machining type can also be determined by embedding a code for specifying the machining type in the control program and reading the code.Furthermore, the machining type currently being performed during machining can also be determined by designing the control program 54 such that it is different for each machining type, or by analyzing CAD / CAM data or the like according to the control program 54.

[0026] The acquisition of data by the data acquisition unit 34 is supported by the execution of the system programs read from the ROM 12 by the CPU 11 and by the arithmetic processing of data mainly by the CPU 11 using the RAM 13 and the non-volatile memory 14. The data acquisition unit 34 stores in the acquired data storage unit 52 the data regarding the state and result of machining by the machine tool 2 as inputted by the control unit 32, the data regarding the result of machining as indicated by an operator via the display / MDI unit 70, the machining type as designated by the machining type designation unit 33, and the like.The data acquisition unit 34 links the data regarding the state and the result of machining by the machine tool 2 as inputted by the control unit 32, the data regarding the result of machining as inputted by the operator, the machining type as determined by the machining type determining unit 33, and the like, and stores the thus linked data as acquired data in the acquired data storage unit 52.

[0027] The setting by the priority condition setting unit 37 is achieved by executing system programs executed via the CPU 11 in the machining condition setting device according to Fig. 1 from the ROM 12, further by the arithmetic processes mainly executed by the CPU 11 with the RAM 13 and the non-volatile memory 14, and by the process control via the display / MDI unit 70 or the like, with data transmission via the interface 18. The priority condition setting unit 37 obtains a priority condition for each machining type in machining a workpiece and stores the priority condition in the priority condition storage unit 56.The priority condition setting unit 37 displays a UI screen for setting the priority condition with respect to each machining type on the display / MDI unit 70, further obtains the priority condition for each machining type set by the operator via the UI screen, generates priority condition data, that is, data in which the machining type is linked to the priority condition, and stores the priority condition data in the priority condition storage unit 56.

[0028] Fig. Figure 3 shows an example of priority condition data stored in the priority condition storage unit 56. The machining type can be defined as a machining process for performing machining with given purposes, such as rough machining, finishing machining, profile machining, for example. The priority conditions can be defined by utilizing the required features in machining, such as "high cycle time" (reductions in machining time in the cycle), "energy saving" (reduction of power consumption, cutting fluid consumption, air consumption, or lubricant consumption, etc.), "high-quality machining" (improvement of surface quality, reduction of machining patterns, etc.), "machining accuracy" (improvement of machining precision), "extension of the service life of machine tool components" (reduction of component wear, etc.).such as on a feed shaft or bearings, reducing heavy loads, etc.), "extending tool life" (reducing wear and preventing breakage on tools, reducing excessive loads, etc.), "extending the life of peripheral devices" (setting a minimum speed, setting operating frequencies, or the number of operations), "reducing the maximum peak power of the machine tool", "improving the production rate of machined products", "optimizing the shape and dimensions of the chips", "reducing vibrations, noise, electromagnetic signals, and heat generated by the machine tool and peripheral devices", or "reducing heat generation by the machine tool",each as examples. With such a configuration, condition expressions for all required details can be stored in advance in the priority condition storage unit 56, and these condition expressions can be used when determining the required properties. A condition expression such as "Slope error < Err, pit “ regarding a machining condition, a machining parameter, measurement data, data regarding the result of machining or the like can be directly defined as a priority condition.

[0029] The priority condition data can be data in which a plurality of priority conditions are linked to a single processing type. On the other hand, the priority condition data can also be data in which priorities are assigned to a plurality of priority conditions. In the example according to Fig. 3, two priority conditions, "shape precision" and "cycle time," are linked to the machining type "finishing," and the priority of the "shape precision" condition is specified with respect to the "cycle time" condition. The priorities of the priority conditions can be defined by weights or the like, and the weights can be indicated by numerical values ​​or by describing the weights using a graph or the like. In addition to setting priority conditions via the display / MDI unit 70, the priority condition setting unit 37 can also be configured to define the priority condition for each machining type in the control program 54 and set the priority condition for each machining type according to the reading of the control program 54.Furthermore, the priority condition setting unit 37 may be configured to set the priority condition for each machining type by obtaining the priority condition from another device or the like connected via a network (not shown).

[0030] The preprocessing by the preprocessing unit 36 ​​is carried out by means of system programs executed by the CPU 11 in the machining condition setting device 1 according to Fig. 1 read from the ROM 12, and by the computing processes mainly of the CPU 11 using the RAM 13 and the non-volatile memory 14. The preprocessing unit 36 ​​generates learning data for use in machine learning by the machine learning device 100 based on data acquired by the data acquisition unit 34 and the priority condition data stored in the priority condition storage unit 56. The preprocessing unit 36 ​​generates the learning data in which the data acquired by the data acquisition unit 34 (and the data stored in the acquired data storage unit 52) ​​have undergone conversion (such as digitization or sampling) into a unified format processable by the machine learning device 100, and the generated learning data is provided to the machine learning device 100 along with the processing type.For example, when the machine learning device 100 performs reinforcement learning, the preprocessing unit 36 ​​generates a set of state data S and determination data D in given formats for the learning process, which are then the learning data.

[0031] The status data S generated by the preprocessing unit 36 ​​according to the exemplary embodiment contain at least tool data S1 with information regarding the tools to be used when machining a workpiece with the machine tool 2, and / or machining condition data S2 including information regarding the machining conditions when machining the workpiece by the machine tool 2 and / or machining parameter data S3 including parameter information regarding the machining of the workpiece with the machine tool 2.

[0032] The tool data S1 are defined as data strings that specify the types and materials of the tools to be used when machining the workpiece with the machine tool 2. The tool types can be classified into cutting tools, milling tools, drilling tools, or the like, each as examples, or according to the shapes of the tools used in machining, or in the form of numerical values ​​that allow for unique identification. The materials of the tools, such as high-speed steel and cemented carbide, can also be expressed in the form of numerical values ​​that allow for unique identification.The tool data S1 can be generated by obtaining information regarding the tools as set via the machining condition setting device 1 and the machine tool 2 by an operator, and also based on information obtained regarding the tools.

[0033] The machining condition data S2 is defined as a data string including machining conditions such as spindle speed, feed rate, and cutting depth based on settings or commands related to machining the workpiece by the machine tool 2. Numerical values ​​can be used for the spindle speed, feed rate, cutting depth, and the like, and the values ​​of the respective machining conditions are expressed using given units. The values ​​for the respective machining conditions are set with commands from the control program 54 or as default values ​​for the controller, thus they can be generated by obtaining the commands or by default values.

[0034] The machining parameter data S3 are defined as data chains including control parameters for the machine, which are used for machining the workpiece by the machine tool 2. The control parameters are parameters such as the control time constants for motors for controlling the machine tool 2, parameters related to controlling peripheral devices or the like, etc. The parameters that are set during machining can be used as the machining parameter data S3.

[0035] For the determination data D generated by the pre-processing unit 36 ​​according to this embodiment, data obtained from the machine tool 2 and the sensors 3 and relating to the priority condition associated with the machining type in the machining state in which the above-mentioned state data S is obtained can be used. If the priority condition data according to the example according to Fig. 3 is stored in the priority condition storage unit 56, and the condition data S is generated based on data acquired by the machine tool 2 (and the sensors 3) during the drilling process, for example, the preprocessing unit 36 ​​generates the determination data D corresponding to the condition data S based on data regarding the pitch error. If the condition data S is generated based on data acquired by the machine tool 2 (and the sensors 3) during profile machining, the preprocessing unit 36 ​​generates the determination data D corresponding to the condition data S based on data for determining the surface quality or the like as measured by the sensors 3.

[0036] The selection of the learning model by the learning model selection unit 105 is carried out by executing system programs executed by the processor 101 in the machining condition setting device 1 according to Fig. 1 from the ROM 102, and by the arithmetic processes mainly by the processor 101 using the RAM 103 and the non-volatile memory 104. The learning model selection unit 105 according to this embodiment selects a learning model from the learning model storage unit 130 according to the processing type input from the pre-processing unit 36 ​​and causes the selected learning model to be used for the learning process by the learning unit 110 and the decision process by the decision unit 122. If a learning model corresponding to the processing type input from the pre-processing unit 36 ​​is not stored in the learning model storage unit 130, the learning model selection unit 105 may newly generate a learning model according to the processing type, and the learning model may be stored in the learning model storage unit 130.

[0037] The learning process by the learning unit 110 is carried out by means of system programs executed by the processor 101 in the machining condition setting device 1 according to Fig. 1 is read from the ROM 102, and by the computing processes mainly by the processor 101 using the RAM 103 and the non-volatile memory 104. The learning unit 110 according to the embodiment performs machine learning with learning data generated by the preprocessing unit 36. The learning unit 110 updates the learning model selected by the learning model selection unit 105 to learn an adaptive behavior regarding the machining conditions and / or machining parameters with respect to the state and result of machining by the machine tool 2 according to well-known reinforcement learning techniques, and stores the updated learning model in the learning model storage unit 130.Reinforcement learning is a technique in which a cycle is iteratively executed according to trial and error, monitoring a current state (the input) of an environment in which a learning object exists, executing a given behavior (the output) in the current state, and awarding a specific reward for the behavior. In this technique, an action (the setting behavior regarding the processing conditions and / or processing parameters in the machine learning device 100 according to the respective application) is learned as an optimal solution that maximizes the total of the rewards. Techniques for such reinforcement learning, which are performed by the learning unit 110, include Q-learning and the like.

[0038] During Q-learning by the learning unit 110, the reward R can be determined based on the priority condition stored with the processing type in the priority condition storage unit 56. If priority condition data are stored according to the example of Fig. 3 is stored in the priority condition storage unit 56 and the state data S as the current learning object is also data-based, which is obtained, for example, during drilling, the reward value R can be set positive (plus) if the slope error (as the determination data) is smaller than the threshold value Err pit which is stored in the priority condition storage unit 56, or it can be set negative (minus) if the slope error is equal to or greater than the threshold value Err pitIf the data are treated according to gradual values, the reward R can be set as a positive (plus) or a negative (minus) value according to the gradual differences. In the example of the gradient error explained above, the size of the positive or negative reward can be set according to the respective deviation of the gradient error from the threshold value Err. pit .

[0039] If a plurality of priority conditions are associated with a processing type in the priority condition storage unit 56, the reward R can be calculated using an expression in which the plurality of priority conditions are combined. If the priority condition data are stored according to the example of Fig. 3 is stored in the priority condition storage unit 56 and, during final processing, for example, the state data S as the current learning object is also data-based, a given formula can be used for the reward calculation, in which data regarding the shape precision and data regarding the cycle time are used, which are defined in advance, and the reward R can then be calculated using this formula for the reward calculation. If priorities are set under the priority conditions, dominant data for calculating the reward can be changed according to such priorities. For example, if the data according to Fig. 3 a profile machining, the formula for the reward calculation can be defined as R = b1 × f (x) (“x” is data regarding the shape precision) + b2 × g (y) (y is data regarding the cycle time) (where f(x) and g(y) are given functions for the reward calculation) and where coefficients b1 and b2 are adjustable with respect to a given reference so that b1 > b2.

[0040] The learning unit 110 can use a neural network (learning model) for a value function Q, and it can be configured to input the state data S and the behavior a as input data into the neural network such that a value (result y) is outputted regarding the behavior a in the respective state. In such a configuration, a three-layer neural network consisting of an input layer, an intermediate layer, and an output layer can be used as the learning model. The learning model can be configured to perform more efficient learning and reasoning using a so-called deep learning technique using a neural network with three or more layers.The learning model updated by the learning unit 110 is stored in the learning model storage unit 130 in the non-volatile memory 104 and used to determine the adaptation behavior with respect to the machining conditions and / or the machining parameters by means of the decision unit 122.

[0041] The learning unit 110 is an essential component in the learning process, but is no longer an essential component once the learning process regarding the adaptation behavior for the processing conditions and / or the processing parameters is completed by the learning unit 110. If a product for which the machine learning device 100 has completed the learning process is to be delivered to a customer, the machine learning device 100 can, for example, be removed from the learning unit 110.

[0042] The decision process by the decision unit 122 is carried out by means of system programs read by the processor 101 in the machining condition setting device 1 from the ROM 102 according to Fig. 1, and by the arithmetic operations mainly by the processor 101 using the RAM 103 and the non-volatile memory 104. The decision unit 122 determines the optimal solution with respect to the adaptation behavior for the machining conditions and / or the machining parameters using the learning model as selected by the learning model selection unit 105 on the basis of the state data S input by the pre-processing unit 36 ​​and outputs the adaptation behavior thus determined with respect to the machining conditions and / or the machining parameters.The decision unit 122 of the embodiment inputs the state data S (the tool data S1, the machining condition data S2, and the machining parameter data S3) input from the preprocessing unit 36 ​​and the adjustment behavior for the machining conditions and / or the machining parameters (a combination for setting the feed rate, setting the spindle speed, and the like, or a change in the setting of the parameters) as input data to the updated learning model (with the determined parameters) in the course of reinforcement learning by the learning unit 110, so that the reward for executing the behavior in question in the current state is calculated.The reward calculation in the decision unit 122 is performed based on the adaptation behavior with respect to the currently valid machining conditions and / or machining parameters. By comparing the plurality of calculated rewards, the adaptation behavior with respect to the machining conditions and / or machining parameters that yields the greatest reward in the calculation is determined as the optimal solution. The optimal solution of the adaptation behavior for the machining conditions and / or machining parameters as determined by the decision unit 122 is input to the control unit 32 and then applied to the machining conditions and / or machining parameters during actual machining.The optimal solution regarding the adaptation behavior can additionally also be used for display on the display / MDI unit 70 or it can be transmitted for output via a wired / wireless network (not shown) to a fog computer, a cloud computer or the like.

[0043] The machining condition setting device 1 can appropriately set the machining conditions and / or the machining parameters according to requirements required by an operator during machining of a workpiece by the machine tool 2.

[0044] In a modification of the machining condition setting device 1 according to the embodiment, the preprocessing unit 36 ​​may further generate workpiece data S4 indicating information related to the workpiece to be machined by the machine tool 2 as state data in addition to the tool data S1, the machining condition data S2, and the machining parameter data S3. The workpiece data S4 is defined as data strings indicating materials of the workpiece to be machined by the machine tool 2. The materials of the workpiece, such as aluminum or iron, may be expressed as numerical values ​​that allow unique identification. The workpiece data S4 may be generated based on information related to the workpiece input to the machining condition setting device 1 or the machine tool 2 by the operator or the like.

[0045] Fig. Fig. 4 is a schematic functional block diagram for explaining the machining condition setting device 1 and the machine learning device 100 according to a second embodiment. The machining condition setting device 1 according to this embodiment includes a configuration for learning a learning process as required for a machine learning device 100 that performs supervised learning. The functions of the Fig. 4 are executed by the corresponding system programs and the control of the operation of the units of the machining condition setting device 1 and the machine learning device 100 according to a control by the CPU 11 in the machining condition setting device 1, as shown in Fig. 1, and by the processor 101 of the machine learning device 100.

[0046] The machining condition setting device 1 of this embodiment includes a control unit 32, the machining type determination unit 33, the data acquisition unit 34, the preprocessing unit 36, and the priority condition setting unit 37, while the machine learning device 100 in the machining condition setting device 1 includes the learning model selection unit 105 and the learning unit 110. In the non-volatile memory 14 according to Fig. 1, data of the storage unit 52 for acquired data, which is acquired from the machine tool 2, the sensors 3, etc., is stored, and the priority condition storage unit 56 is provided, in which priority condition data set by the priority condition setting unit 37 is stored. The non-volatile memory 104 of the machine learning device 100 according to Fig. 1 also provides the learning model storage unit 130 in which a learning model constructed by the machine learning of the learning unit 110 is stored.

[0047] The control unit 32, the machining type determination unit 33, the data acquisition unit 34, the priority condition setting unit 37, and the learning model selection unit 105 according to the second embodiment have functions similar to the functions of the control unit 32, the machining type determination unit 33, the data acquisition unit 34, the priority condition setting unit 37, and the learning model selection unit 105 according to the first embodiment.

[0048] The preprocessing unit 36 ​​according to the present embodiment generates learning data for use in machine learning by the machine learning device 100 based on the data acquired by the data acquisition unit 34 and the priority condition data stored in the priority condition storage unit 56. The preprocessing unit 36 ​​generates the learning data in which the data acquired by the data acquisition unit 34 (and stored in the acquired data storage unit 52) ​​has been subjected to a conversion (such as digitization or scanning), wherein the conversion is carried out into a uniform format that can be processed in the machine learning device 100, and wherein the generated learning data is provided to the machine learning device 100 together with the processing type.For example, when the machine learning device 100 performs supervised learning, the preprocessing unit 36 ​​generates a set of state data S and labeled data (label data) L in the formats for learning, each as learning data.

[0049] The preprocessing unit 36 ​​according to this embodiment generates the set of state data S and labeled data L as learning data based on only those acquired data that satisfy a priority condition associated with the machining type in a state in which the data were respectively acquired, wherein the selection is made among the data acquired by the data acquisition unit 34 (stored in the acquired data storage unit 52). In the case of data acquired only during drilling and under the assumption that the priority condition data according to the example of Fig. 3 are stored in the priority condition storage unit 56 (as an example), the set of state data S and labeled data L as learning data is derived only from the data for which the slope error is smaller than the threshold value Err pgt and only these data are passed to the machine learning device 100. Therefore, only those learning data that satisfy the priority condition for the respective processing type are passed from the preprocessing unit 36 ​​to the machine learning device 100. Therefore, the learning model that has been subjected to the learning process using these learning data yields at least either the processing conditions or the processing parameters for the processing while maintaining the priority conditions.

[0050] In this embodiment, the status data S generated by the preprocessing unit 36 ​​contains at least the tool data S1 with information regarding the tools used in machining a workpiece with the machine tool 2. The tool data S1 are defined as data strings that indicate the types and materials of the tools to be used for machining the workpiece with the machine tool 2. The types of tools can be classified into cutting tools, milling tools, drilling tools, etc., each as examples, depending on the shapes of the tools or the use in machining, and the classification can be expressed with numerical values ​​that enable unique identification. The materials of the tools, such as high-speed steel and cemented carbide, can also be expressed in the form of numerical values ​​that enable unique identification.The tool data S1 can be generated by obtaining information regarding the tools used for the machining condition setting device 1 and the machine tool 2 by an operator, and based on the information obtained regarding the tools.

[0051] The label data L generated by the preprocessing unit 36 ​​according to this embodiment includes at least machining condition label data L1 provided with information regarding machining conditions when machining a workpiece by the machine tool 2 in a machining state in which the state data S was obtained, or machining parameter label data L2 with parameter information regarding the machining of the workpiece by the machine tool 2.

[0052] The machining condition label data L1 is defined as a data string containing the machining conditions such as the spindle speed, feed rate, and cutting depth according to the settings or instructions for machining the workpiece with the machine tool 2. Numerical values ​​can be used for the spindle speed, feed rate, cutting depth, and the like, in which values ​​related to the respective machining conditions are expressed using given units. The values ​​for the respective machining conditions are set by instructions specified by the control program 54 or as default values ​​for the controller, and thus can be generated by acquiring the instructions or based on default values.

[0053] The machining parameter label data L2 is defined as data strings including control parameters for the machine, which are referred to for machining the workpiece by the machine tool 2. The control parameters include, for example, control time constants for the motors of the machine tool 2, parameters related to the control of peripheral devices, or the like, etc. As the machining parameter label data L2, parameters that are set during machining can be obtained.

[0054] The learning by the learning unit 110 according to this embodiment is carried out by the system programs read out from the ROM 102 by the processor 101 included in the machining condition setting device 1 according to Fig. 1, and by the arithmetic operations mainly by the processor 101 using the RAM 103 and the non-volatile memory 104. The learning unit 110 according to the embodiment performs machine learning using the learning data generated by the preprocessing unit 36. The learning unit 110 updates the learning model selected by the learning model selection unit 105 so as to learn at least the machining conditions or the machining parameters that satisfy the priority condition with respect to the machining state of the machine tool 2, using well-known supervised learning techniques, and stores the updated learning model in the learning model storage unit 130.For the supervised learning technique executed by the learning unit 110, for example, the multi-layer perceptron technique, the recurrent neural network technique, the long short-term memory technique, the convolutional neural network (CNN) technique, and the like are considered.

[0055] The learning unit 110 according to this embodiment updates the learning model to learn the machining conditions and machining parameters that satisfy the priority condition according to the machining type in conjunction with the machining state of the machine tool 2. The learning model generated by the learning unit 110 according to this embodiment in this manner can be used to estimate the machining conditions and / or machining parameters that satisfy the priority condition depending on the machining type when the corresponding machining state of the machine tool 2 exists.

[0056] The learning unit 110 is an essential component during the learning stage, but not necessarily an essential component after completion of the learning process of at least the machining conditions and / or the machining parameters that satisfy the priority condition according to the machining type, wherein the learning process is performed by the learning unit 110. For example, if the machine learning device 100 is delivered to a customer after the learning process has been completed, the delivery may include a machine learning device 100 from which the learning unit 110 has been removed.

[0057] The machining condition setting device 1 with the above configuration according to the embodiment generates a plurality of learning models in which at least the machining conditions and / or the machining parameters that, in association, satisfy the priority condition according to the machining type and the machining state of the machine tool 2 are learned. Using the plurality of learning models thus generated, an estimation unit 120, described in more detail below, can perform an estimation process based on the state data S obtained from the machine tool 2, which is necessary for determining at least the machining conditions and / or the machining parameters that are more suitable and correspond to the machining type in the obtained state.

[0058] In a modification of the machining condition setting device 1 according to this embodiment, the preprocessing unit 36 ​​may further generate, in addition to the tool data S1, workpiece data S4 indicating information regarding the workpiece to be machined by the machine tool 2. The workpiece data S4 is defined as data strings containing materials of the workpiece to be machined by the machine tool 2. The materials of the workpiece, such as aluminum or iron, may be selected as numerical values ​​each enabling unique identification. The workpiece data S4 may be generated based on information regarding the workpiece set for the machining condition setting device 1 or the machine tool 2 by, for example, the operator.

[0059] Fig. 5 illustrates, with a schematic functional block diagram, the machining condition setting device 1 and the machine learning device 100 according to a third embodiment. The machining condition setting device 1 according to this embodiment includes an estimation configuration, which is required when the machine learning device 100 estimates at least the machining conditions and / or the machining parameters for machining with the respective machine tool. The functional blocks according to Fig. 5 are implemented by executing respective system programs and controlling the operation of the units of the machining condition setting device 1 and the machine learning device 100 by means of the CPU 11 in the machining condition setting device 1 according to Fig. 1 and the processor 101 of the machine learning device 100.

[0060] The machining condition setting device 1 of this embodiment includes the control unit 32, the machining type determination unit 33, the data acquisition unit 34, the preprocessing unit 36 ​​and the priority condition setting unit 37, and the machine learning device 100 in the machining condition setting device 100 includes the learning model selection unit 105 and an estimation unit 120. In the non-volatile memory 14 according to Fig. 1, the acquisition data storage unit 52, in which data acquired by the machine tool 2, data acquired by the sensors 3, and so on are stored, and the priority condition storage unit 56, in which priority condition data set by the priority condition setting unit 37 are stored, are provided. The non-volatile memory 104 of the machine learning device 100 according to Fig. 1 also provides the learning model storage unit 130, in which learning models are stored that are generated by machine learning by means of the learning unit 110, as described with reference to the second embodiment.

[0061] The control unit 32, the machining type determination unit 33, the data acquisition unit 34, the priority condition setting unit 37, and the learning model selection unit 105 according to the third embodiment have functions corresponding to the functions of the control unit 32, the machining type determination unit 33, the data acquisition unit 34, the priority condition setting unit 37, and the learning model selection unit 105 according to the first embodiment.

[0062] In this embodiment, when estimating the machining conditions and / or machining parameters that satisfy the priority condition using the learning model by the machine learning device 100, the preprocessing unit 36 ​​performs conversion (such as digitization or scanning) into a unified format processable in the machine learning device 100 based on the data acquired by the data acquisition unit 34 and the priority condition data stored in the priority condition storage unit 56. The state data S in a predetermined format for use for estimation in the machine learning device 100 is generated from the converted data and the generated state data S together with the machining type and is provided to the machine learning device 100.For example, the preprocessing unit 36 ​​generates the tool data S1 based on the data obtained by the data acquisition unit 34.

[0063] Estimation by the estimation unit 120 is performed by system programs read by the processor from the ROM 102 and by arithmetic operations mainly by the processor 101 using the RAM 103 and the non-volatile memory 104. The estimation unit 120 estimates the machining conditions and / or machining parameters that satisfy the priority condition using the learning model selected by the learning model selection unit 105 based on the state data S generated by the preprocessing unit 36. In the estimation unit 120 of this embodiment, the state data S obtained by the preprocessing unit 36 ​​is input into the learning model generated (with certain parameters) by the learning unit 110, and the machining conditions and / or machining parameters that satisfy the priority condition associated with the machining type are thus estimated and output.A result estimated by the estimation unit 120 is input to the control unit 32.

[0064] The machining condition setting device 1 of the above configuration according to the embodiment is capable of estimating at least one of the above-mentioned determinations, that is, estimating the machining conditions and the machining parameters which correspond to the priority condition according to each machining type, and controlling the machining operation for a workpiece by the machine tool 2 based on at least one of these estimated determinations, that is, based on the machining conditions and / or the machining parameters.

[0065] In a modification of the machining condition setting device 1 according to this embodiment, the preprocessing unit 36 ​​may further generate, in addition to the tool data S1 as state data, the workpiece data S4 indicating information regarding the workpiece to be machined by the machine tool 2. The workpiece data S4 is defined as data strings indicating the materials of the workpieces to be machined by the machine tool 2. The materials of the workpieces, such as aluminum and iron, may be expressed with numerical values ​​that enable each to be uniquely identified. The workpiece data S4 may be generated based on information regarding the workpieces set for the machining condition setting device 1 or for the machine tool 2 by the operator or otherwise.

[0066] Fig. 6 is a schematic functional block diagram for explaining the machining condition setting device 1 and the machine learning device 100 according to a fourth embodiment. The machining condition setting device 1 of this embodiment includes a configuration for a machine learning device 100 that performs unsupervised learning. The functional blocks according to Fig. 6 are implemented by executing the respective system programs and by controlling the operation of the units of the machining condition setting device 1 and the machine learning device 100 by means of the CPU 11 in the machining condition setting device 1 according to Fig. 1 and by the processor 101 of the machine learning device 100.

[0067] The machining condition setting device 1 of this embodiment includes the control unit 32, the machining type determination unit 33, the data acquisition unit 34, the preprocessing unit 36 ​​and the priority condition setting unit 37, and the machine learning device 100 of the machining condition setting device 1 includes the learning model selection unit 105 and the learning unit 110. The non-volatile memory 14 according to Fig. 1 provides the acquired data storage unit 52, in which data acquired from the machine tool 2, the sensors 3, and the like are stored, and also the priority condition storage unit 56, in which priority condition data set by the priority condition setting unit 37 are stored. The non-volatile memory 104 of the machine learning device 100 according to Fig. 1 provides the learning model storage unit 130, which stores a learning model formed by machine learning of the learning unit 110.

[0068] The control unit 32, the machining type determination unit 33, the data acquisition unit 34, the priority condition setting unit 37, and the learning model selection unit 105 according to the fourth embodiment have functions corresponding to the functions of the control unit 32, the machining type determination unit 33, the data acquisition unit 34, the priority condition setting unit 37, and the learning model selection unit 105 according to the first embodiment, respectively.

[0069] The preprocessing unit 36 ​​according to this embodiment generates learning data for use in machine learning by the machine learning device 100 based on data acquired by the data acquisition unit 34 and the priority condition data stored in the priority condition storage unit 56. The preprocessing unit 36 ​​generates the learning data by converting the data acquired by the data acquisition unit 34 and stored in the acquired data storage unit 52, such as digitizing or scanning, into a unified format that can be processed by the machine learning device 100, and then transmitting the generated learning data, along with the processing type, to the machine learning device 100.For example, when the machine learning device 100 performs unsupervised learning, the preprocessing unit 36 ​​generates state data S in a predetermined format during learning, and this is the learning data.

[0070] The preprocessing unit of this embodiment generates the state data S as learning data only from the acquired data that satisfy a priority condition in connection with the machining type in a state in which the data was acquired; thus, a selection is made from the data acquired by the data acquisition unit 34 (and stored in the acquired data storage unit 52). For the data acquired during drilling, for example, when the priority condition data according to the example of Fig. 3 are stored in the priority condition storage unit 56, the state data S is generated as learning data only from those data for which the slope error is smaller than the threshold value Err pet and this data is passed to the machine learning device 100. The preprocessing unit 36 ​​thus passes to the machine learning device 100 only those learning data generated based on the data that satisfy the priority condition for the respective machining type. Therefore, the learning model that has performed the learning process using this learning data represents a distribution for the machining conditions and / or the machining parameters for the machining that satisfies the priority conditions.

[0071] The status data S, which are generated by the preprocessing unit 36 ​​according to this exemplary embodiment, contain at least the tool data S1 with information regarding the tools to be used when machining workpieces by the machine tool 2 and the machining condition data S2 with information regarding the machining conditions when machining the workpieces by the machine tool 2 and / or the machining parameter data S3 with parameter information regarding the machining of the workpiece with the machine tool 2.

[0072] The tool data S1 are given as data strings that indicate the types and materials of the tools to be used for machining the workpieces with the machine tool 2. The types of tools can be classified, for example, into cutting tools, milling tools, drilling tools, and the like according to the shapes of the tools or the use in machining, and the classification can be made with numerical values ​​that allow each to be uniquely identified. The materials of the tools, such as high-speed steel and cemented carbide, can be expressed in the form of numerical values ​​that allow uniquely identification. The tool data S1 can be generated by obtaining information regarding the tools set for the machining condition setting device 1 and the machine tool 2 by an operator, and based on the information obtained regarding the tools.

[0073] The machining condition data S2 is defined as a data string containing the machining conditions, such as spindle speed, feed rate, and cutting depth, according to settings or instructions for machining the workpiece by the machine tool 2. Numerical values ​​can be used for the spindle speed, feed rate, cutting depth, and the like, and the values ​​are each assigned the appropriate units for the respective machining conditions. The values ​​related to the respective machining conditions are set according to instructions from the control program 54 or as default values ​​for the controller.

[0074] The machining parameter data S3 are defined as data chains containing control parameters for the machine, which are accessed for machining the workpiece with the machine tool 2. The control parameters include, for example, control time constants for motors controlling the machine tool 2, parameters related to the control of peripheral devices, or the like. Parameters set during machining can be used for the machining parameter data S3.

[0075] The learning by the learning unit 110 according to this embodiment is carried out by system programs read out from the ROM 102 by means of the processor 101 in the machining condition setting device 1 according to Fig. 1 and by the arithmetic processes mainly performed by the processor 101 using the RAM 103 and the non-volatile memory 104. The learning unit 110 according to this embodiment executes the machine learning process using learning data generated by the preprocessing unit 36. The learning unit 110 updates the learning model selected by the learning model selection unit 105 so as to learn the distribution of the machining conditions and / or the machining parameters that satisfy the priority condition in machining by the machine tool 2, performing a well-known unsupervised learning, and the updated learning model is stored in the learning model storage unit 130.For the unsupervised learning technique performed by the learning unit 110, the so-called autoencoder technique, the k-means technique and the like can be used, for example.

[0076] The learning unit 110 according to this embodiment updates the learning model to learn the distribution of the machining conditions and / or the machining parameters that satisfy the priority condition according to the machining type during machining by the machine tool 2. The learning model thus generated by the learning unit 110 according to the embodiment can be used to estimate the machining conditions and / or the machining parameters that satisfy the priorities according to the machining type during machining by the machine tool 2.

[0077] The learning unit 110 is an essential component during the learning stage, but not necessarily an essential component after the completion of the learning process, with respect to the distribution of the machining conditions and / or machining parameters that satisfy the priority condition according to the machining type. For example, once the machine learning device 100 has completed the learning process and is then delivered to the user, the learning unit 110 may be removed from the machine learning device 100.

[0078] The machining condition setting device 1 with the above configuration according to the embodiment generates a plurality of learning models in which the distribution of the machining conditions and the machining parameters that satisfy the priority condition according to the machining type are learned. Using the plurality of learning models thus generated, the estimation unit 120, which will be described in more detail below, is capable of performing an estimation process based on the state data S obtained from the machine tool 2, which is necessary for determining the machining conditions and / or the machining parameters that become more appropriate and correspond to the machining type in the obtained state.

[0079] According to a modification of the machining condition setting device 1 according to this embodiment, the preprocessing unit 36 ​​may further generate the workpiece data S4 indicating information of the workpiece to be machined by the machine tool 2, in addition to the tool data S1, the machining condition data S2, and the machining parameter data S3. The workpiece data S4 is represented by data strings indicating the materials of the workpiece to be machined by the machine tool 2. The materials of the workpieces, such as aluminum or iron, may be expressed with numerical values, each of which enables unique identification. The workpiece data S4 may be generated based on information regarding the workpiece set by the machining condition setting device 1 or the machine tool 2 by the operator or the like.

[0080] Fig. Figure 7 is a schematic functional block diagram for explaining the machining condition setting device 1 and the machine learning device 100 according to a fifth embodiment. The machining condition setting device 1 of this embodiment includes a configuration required when the machine learning device 100 estimates the machining conditions and / or machining parameters for machining with a machine tool. The functions of the blocks according to Fig. 7 are executed by the respective system programs and the control of the operation of the units of the machining condition setting device 1 and the machine learning device 100 by means of the CPU 11 in the machining condition setting device 1 according to Fig. 1 and the processor 101 of the machine learning device 100.

[0081] The machining condition setting device 1 of this embodiment includes the control unit 32, the machining type determination unit 33, the data acquisition unit 34, the preprocessing unit 36 ​​and the priority condition setting unit 37, and the machine learning device 100 in the machining condition setting device 1 and the learning unit includes the learning model selection unit 105 and the estimation unit 120. The Fig. The non-volatile memory 14 shown in Figure 1 represents the acquired data storage unit 52, in which data acquired by the machine tool 2, the sensors 3, and possibly other units are stored, and the priority condition storage unit 56, in which priority condition data set by the priority condition setting unit 37 is stored. The non-volatile memory 104 of the machine learning device 100 according to Fig. 1 illustrates the learning model storage unit 130, which stores a learning model formed by the machine learning of the learning unit 110 according to the second embodiment.

[0082] The control unit 32, the machining type determination unit 33, the data acquisition unit 34, the priority condition setting unit 37, and the learning model selection unit 105 according to the fifth embodiment have functions corresponding to the functions of the control unit 32, the machining type determination unit 33, the data acquisition unit 34, the priority condition setting unit 37, and the learning model selection unit 105 according to the first embodiment.

[0083] At the stage of estimating the machining conditions and / or machining parameters that satisfy the priority condition using the learning model by the machine learning device 100, the preprocessing unit 36 ​​according to this embodiment performs conversion (such as digitization or scanning) into the unified format processable in the machine learning device 100 based on the data acquired by the data acquisition unit 34 and the priority condition data stored in the priority condition storage unit 56. The state data S in the format used by the machine learning device 100 in the estimation is generated from the converted data, and the generated state data S, along with the machining type, is provided to the machine learning device 100.For example, the preprocessing unit 36 ​​generates at least the tool data S1 and, if applicable, the machining condition data S2 and / or the machining parameters S3 on the basis of the data obtained by the data acquisition unit 34.

[0084] The estimation by the estimation unit 120 is carried out via the system programs read out from the ROM 102 by the processor 101, wherein the processor 101 in the machining condition setting device 1 is Fig. 1, and by the arithmetic operations mainly of the processor 101 using the RAM 103 and the non-volatile memory 104. The estimation unit 120 estimates the machining conditions and / or the machining parameters that satisfy the priority condition using the learning model selected by the learning model selection unit 105 based on the state data S generated by the preprocessing unit 36. In the estimation unit 120 of this embodiment, the machining conditions and / or the machining parameters that satisfy the priority condition are estimated and output based on the arrangement of the state data S input from the preprocessing unit 36 ​​in a distribution of data in the learning model generated by the learning unit 110. The result estimated by the estimation unit 120 is given to the control unit 32.

[0085] The estimation unit 120 calculates the distance between the respective data group (“cluster”) in the data distribution in the learning model generated by the learning unit 110 and the arrangement (position) of state data S input from the preprocessing unit 36 ​​and arrives at the estimation result that the current machining conditions or machining parameters in the current machining type satisfy the priority condition if the distance between the position of the state data S as input from the preprocessing unit 36 ​​and the nearest data set Cl n is equal to or shorter than a given threshold Dist th1. The estimation unit 120 comes to the conclusion that the priority condition in the current machining mode is not fulfilled with regard to the current machining conditions and / or the current machining parameters if the distance between the position of the state data S input by the preprocessing unit 36 ​​and the nearest data set Cl n is greater than the specified threshold Dist th1 . The estimation unit 120 sets a machining condition or a machining parameter or a plurality of working conditions or a plurality of machining parameters within the state data input by the preprocessing unit 36 ​​according to a predetermined rule so that the distance to the data set Cl n is equal to or less than the specified threshold Dist th1The specified rule for this setting can be a rule that, for example, permanently sets a given processing condition or a given processing parameter. The rule can stipulate that the setting is made in such a way that the distance to the data set Cl n is equal to or less than the specified threshold Dist th1 based on the smallest possible adjustment amount. A rule can be provided that excludes a specific machining condition or a specific machining parameter from the settings. Thus, the estimation unit 120 estimates the machining conditions and / or machining parameters and outputs those that satisfy the priority condition based on the current state data S and the learning model.

[0086] The estimation unit 120 may issue a command to stop processing when the distance between the position of the state data S input from the preprocessing unit 36 ​​and the nearest data set Cl n is greater than a specified threshold Dist th2 (Threshold Dist th2 > Threshold Dist th1 ). An emergency stop command can be issued if the distance is greater than a predefined threshold Dist th3 (Threshold Dist th3 > Threshold Dist th2 ). This technique makes it possible to estimate whether the machining condition is abnormal when the operation is very different from normal operation, and this can be communicated to the operator.

[0087] The machining condition setting device 1 having the above-described configuration according to the embodiment is configured to estimate the machining conditions and / or the machining parameters that satisfy the priority condition according to each machining type, and thus the machining operation of a workpiece by the machine tool 2 can be controlled based on the machining conditions and / or the machining parameters as estimated.

[0088] In a modification of the machining condition setting device 1 according to this embodiment, the preprocessing unit 36 ​​may further generate the workpiece data S4 indicating information of the workpiece to be machined by the machine tool 2, which is then state data in addition to the tool data S1, the machining condition data S2, and the machining parameter data S3. The workpiece data S4 is given by data strings indicating the materials of the workpieces to be machined by the machine tool 2. The materials of the workpieces, such as aluminum and iron, may be expressed in the form of numerical values, each of which enables unique identification. The workpiece data S4 may be generated based on information regarding the workpiece that is set for the machining condition setting device 1 or the machine tool 2 by an operator or otherwise.

[0089] As sixth to eighth embodiments, examples will be explained in more detail below in which the machining condition setting devices 1 according to the first to fifth embodiments are implemented as parts of systems that are each connected via a wired / wireless network to a plurality of devices, such as a cloud server, a host computer, a fog computer, or an edge computer (as part of a robot controller, a control device, or the like). In the sixth to eighth embodiments, exemplified in Fig. As shown in Figure 8, a system is configured to be logically divided into three layers, with the plurality of devices each connected to a network. The three layers are: a layer including a cloud server 6 or the like, a layer including fog computers 7 or the like, and a layer including edge computers 8 or the like (a robot controller, control device, and the like, each contained in cells 9). In such a system, the machining condition setting device 1 according to a variant of the invention can be implemented in any one of the cloud server 6, the fog computer 7, or the edge computer 8.The machining condition setting device 1 is configured to perform distributed learning by sharing data used in the machine learning processes with the plurality of devices via the network, particularly for large-scale analysis, with the collection of the generated learning models in the fog computers 7 or the cloud server 6, mutual use of the generated learning models, and the like. In the exemplary system, accordingly. Fig. 8, multiple cells 9 are provided, each located in factories in different regions, and the fog computers 7 of a higher layer (level) separately manage the cells 9 in given units (each unit consisting of a single factory or a collection of multiple factories of the same manufacturer, or the like). Data collected and analyzed by the fog computers 7 is further collected and analyzed by the cloud server 6 at a next higher level, so that the resulting information can be used for control in each of the edge computers 8 or the like.

[0090] Fig. 9 schematically shows a hardware configuration in which the machining condition setting devices are implemented in computers such as the cloud server or the fog computers.

[0091] A CPU 311 in the machining condition setting device 1', which is implemented in a computer according to the embodiment, is a processor for overall control of the machining condition setting device 1'. The CPU 311 reads system programs stored in a ROM 312 via a bus 21 and controls the entire machining condition setting device 1' according to the system programs. Temporarily required calculation data, display data, various types of data input by an operator via an unillustrated input device, and the like are temporarily stored in a RAM 313.

[0092] A non-volatile memory 314 maintains its storage state with backup from a battery (not shown) or the like, for example, even when the machining condition setting device 1' is turned off. Programs input via an input device 371 and various types of data obtained by units of the machining condition setting device 1' or through a network 5 from a machine tool 2' (or sensors 3) or the like are stored in the non-volatile memory 314. The programs or other types of data stored in the non-volatile memory 314 can be transferred to the bus RAM 313 upon execution / use.Various system programs (including system programs for controlling interaction with the machine learning device 100, as described in more detail below), such as well-known analysis programs, are pre-written in the RAM 312.

[0093] The machining condition setting device 1' is connected to a wired / wireless network 5 via an interface 319. At least one machine tool 2' (machine tool with a controller), other machining condition setting devices 1, the edge computers 8, the fog computers 7, the cloud server 6, and the like are connected to the network 5 so that data can be exchanged with the machining condition setting device 1'.

[0094] A display device 370 displays data obtained via an interface 317, which is stored in a memory or obtained as a result of executing programs or the like. The input device 371 has a keyboard, a pointing device, or the like for transferring commands, data, or the like by an operator, with the data being supplied to the CPU 311 via an interface 318.

[0095] An interface 321 connects the machining condition setting device 1' and the machine learning device 100. The machine learning device 100 has a configuration corresponding to the configuration shown in Fig. 1 was described.

[0096] In a configuration in which the machining condition setting device 1' is implemented in a computer such as a cloud server or a fog computer, the functions of the machining condition setting device 1' correspond to the functions described with reference to the first to third embodiments, except that the acquisition of information from the machine tool 2' and the sensors 3 and the commands to the machine tool 2 for setting the machining conditions are carried out via the network 5. Here, the machine tool 2' includes the control device, and thus the control unit 32 in the machining condition setting device 1' does not control the machine tool 2', but acquires information about the machining state in the machine tool 2' from the control device in the machine tool 2'.The control unit 32 in the machining condition setting device 1' indirectly controls the sensors 3 via the control device in the machine tool 2' and obtains measured values ​​from the sensors 3 via the control device in the machine tool 2'.

[0097] Fig. 10 schematically shows the configuration of a machining condition setting system according to a sixth embodiment, including the machining condition setting device 1'. The machining condition setting system 500 includes a plurality of machining condition setting devices 1, 1', a plurality of machine tools 2', and a network 5 connecting the machining condition setting devices 1, 1' and the machine tools 2'.

[0098] In the machining condition setting system 500, the machining condition setting device 1' with the machine learning device 100 determines machining conditions that satisfy the priority condition based on the machining state of the machine tool 2' using the learning results of the learning unit 110. At least one machining condition setting device 1' is configured to learn the machining conditions that satisfy the priority conditions corresponding to the machining state of the machine tools 2, 2', which are common to all the machining condition setting devices 1, 1', based on the state data S and the labeled data L or the determination data D obtained by each of the plurality of other machining condition setting devices 1, 1', so as to share the learning results with other machining condition setting devices 1, 1'.Accordingly, with the machining condition setting system 500, the speed and reliability of the learning process can be improved by using various amounts of data (including the state data S and the labeled data L or the destination data D) as inputs.

[0099] Fig. 11 schematically shows the configuration of a system according to a seventh embodiment, in which the machine learning device and the machining condition setting devices are implemented in different devices. The machining condition setting system 500' includes at least one machine learning device 100 implemented as part of a computer, such as a cloud server, a host computer, or a fog computer ( Fig. 11 shows an example of an implementation as part of a fog computer 7), a plurality of machining condition setting devices 1", wherein a network 5 establishes connections among the machining condition setting devices 1" and with the computer. In the hardware configuration of the computer, hardware components such as the CPU 311, the RAM 313 and the non-volatile memory 314 are connected in a common computer via the bus 320, corresponding to the schematically illustrated hardware configuration of the machining condition setting device 1' according to Fig. 9.

[0100] In the machining condition setting system 500' of the above configuration, the machine learning device 100 learns machining conditions corresponding to the priority conditions corresponding to the machining states of the machine tool 2 that are common to all the machining condition setting devices 1" based on the state data S and the labeled data L or the determination data D obtained for each of the plurality of machining condition setting devices 1". The machining conditions for each machine tool 2 can then be set using the results of this learning.With this configuration of the machining condition setting system 500', the plurality of machining condition setting devices 1" are each capable of connecting the required number of machining condition setting devices 1" to the machine learning device 100 at the required times, regardless of the location and operating times of the machining condition setting devices 1".

[0101] Fig. 12 schematically shows the configuration of a machining condition setting system 500" according to an eighth embodiment with the machine learning devices 100' and the machining condition setting devices 1. The machining condition setting system 500" has at least one machine learning device 100' implemented in a computer such as an edge computer, a fog computer, a host computer, or a cloud server ( Fig.12 shows an example with an implementation as part of a fog computer 7), further comprising a plurality of machining condition setting devices 1 and a wired / wireless network 5 which establishes connections between the machining condition setting devices 1 and the computer.

[0102] In the machining condition setting system 500" of the above configuration, the fog computer 7 including the machine learning device 100' acquires, from each of the machining condition setting devices 1, a learning model obtained as a result of machine learning by the machine learning device 100 in each of the machining condition setting devices 1. The machine learning device 100' in the fog computer 7 newly generates an optimized learning model by optimizing or fitting based on the plurality of learning models and outputs the thus generated learning model to the machining condition setting devices 1.

[0103] As an example of the optimization of the learning model by the machine learning device 100', the generation of a distilled model based on the plurality of learning models obtained by the machining condition setting devices 1 can be cited. In this example, the machine learning device 100' generates initial data for input to the learning models and regenerates the distilled model as a learning model by relearning using initial data obtained as a result of input to each of the learning models. As described above, the distilled model thus generated is distributed to the machining condition setting devices 1 or to other computers via an external storage medium or the network 5.

[0104] Another example of the optimization of the learning model by the machine learning device 100' in the distillation of the multiple learning models obtained by the multiple machining condition setting devices 1 may be to analyze a distribution of output data from each learning model in response to the input data using a common statistical method, exclude outliers in the sets of input data and output data, and perform distillation on the sets of input data and output data without the excluded outliers. With such a process, exceptional estimation results can be excluded from the input data and output data obtained by each learning model, and the distilled model can be generated using the sets of input data and output data in which unusual estimation results are excluded.The model distilled in this way can be used as a more versatile learning model compared to the learning models generated with the multiple machining condition setting devices 1.

[0105] Another well-known technique for optimizing a learning model can also be used (such as analyzing each learning model and optimizing hyperparameters of the learning model based on the results of the analysis).

[0106] When using the machining condition setting system 500" according to this embodiment, for example, the machine learning device 100' may be provided in the fog computer 7 used as an edge computer for the plurality of machining condition setting devices 1, and the learning models generated in each of the machining condition setting devices 1 may be collected and stored in the fog computer 7. After optimization based on the plurality of stored learning models, the optimized learning model may be distributed back to the machining condition setting devices 1 as needed.

[0107] In the machining condition setting system 500" according to this embodiment, for example, the learning models collected and stored in the fog computer 7 and the learning model optimized in the fog computer can be collected in a host computer or a cloud server at a higher level, and the learning models collected in the server can be used for analysis in a factory or at a manufacturer of the machining condition setting devices 1. Examples include designing and distributing more adaptive learning models to and from a higher-level server, supporting maintenance work based on the results of the learning model analysis, investigating the performance of each machining condition setting device 1, using them for developing new devices, or the like.

[0108] Although embodiments of the invention have been described in more detail above, the invention is not limited to such examples and can be implemented with various modifications.

[0109] The machining condition setting device 1 and the machine learning device 100 are described above with different CPUs (processors), but the machine learning device 100 can also be implemented by the CPU 11 provided in the machining condition setting device 1 and the system programs stored in the ROM 12.

Claims

[1] Machining condition setting device (1) for setting a machining condition of a machine tool for machining a workpiece, the machining condition setting device (1) comprising: a data acquisition unit (34) for obtaining at least one data set indicating a state of machining including a machining type in the machine tool; a priority condition storage unit (56) which stores priority condition data in which the machining type in the machine tool is linked to a priority condition for the machining type; a preprocessing unit (36) which generates data for use in machine learning based on data obtained by the data acquisition unit (34) and the priority condition corresponding to the processing type contained in the data and stored in the priority condition storage unit (56); and a machine learning device (100) which performs a machine learning process with respect to a machining condition and / or a machining parameter for machining by the machine tool in an environment in which the workpiece is machined by the machine tool, on the basis of data generated by the preprocessing unit (36), wherein the machine learning device (100) contains: a learning model storage unit (130) which stores a plurality of learning models generated for each machining type in the machine tool, and a learning model selection unit (105) which selects a learning model from the plurality of learning models stored in the learning model storage unit (130) for use in the machine learning process based on the processing type included in the data generated by the preprocessing unit (36). [2] Machining condition setting device (1) according to claim 1, wherein on the basis of the data obtained by the data acquisition unit (34), the preprocessing unit (36) generates determination data (D) based on the priority condition corresponding to the machining type contained in the data and stored in the priority condition storage unit (56), and on status data (S) containing at least tool data (S1) including information of the tool used for machining the workpiece with the machine tool, and machining condition data (S2) including information relating to the machining condition during machining of the workpiece by the machine tool and / or machining parameter data (S3) including information relating to the machining parameters during machining of the workpiece by the machine tool as data used for reinforcement learning by the machine learning device (100), wherein the machine learning device (100) includes: a learning unit (110) which generates a learning model in which the state of machining by means of the machine tool is linked to adaptation behavior for the machining condition and / or the machining parameter according to the priority condition on the basis of the state data (S) and the determination data (D), and a decision unit (122) which determines the processing condition and / or the processing parameters which satisfy the priority condition, using the learning model generated by the learning unit (110) on the basis of the state data (S), and wherein the machine tool is controlled based on the machining condition and / or the machining parameters which satisfy the priority condition determined by the decision unit (122). [3] Machining condition setting device (1) according to claim 1, wherein the preprocessing unit (36) generates, only on the basis of data obtained by the data acquisition unit (34) and satisfying the priority conditions according to the machining type contained in the data and stored in the priority condition storage unit (56), state data (S) containing at least tool data (S1) with information about the tool to be used in machining the workpiece by the machine tool, and labeled data (L), which is / are labeled machining condition data (L1) labeled with the machining condition, and / or labeled machining parameter data (L2) labeled with the machining parameter, as data to be used in supervised learning by the machine learning device (100), and the machine learning device (100) contains a learning unit (110) which, on the basis of the state data (S) and the labeled data (L), generates a learning model in which the state of the machining by the machine tool is linked to the machining condition and / or the machining parameter which satisfy the priority condition. [4] Machining condition setting device (1) according to claim 1, wherein the preprocessing unit (36) generates, on the basis of the data obtained by the data acquisition unit (34), state data (S) which contain at least tool data (S1) including information about a tool to be used in the machining of the workpiece by the machine tool, as data to be used for the estimation by the machine learning device (100), the machine learning device (100) has an estimation unit (120) which, on the basis of the state data (S), estimates the processing condition and / or the processing parameter which satisfies the priority condition, using the learning model selected by the learning model selection unit (105), the learning models stored in the learning model storage unit (130) are those learning models in which the state of machining by the machine tool is linked to the machining condition and / or the machining parameter which satisfies the priority condition, and the machine tool is controlled on the basis of the machining condition and / or the machining parameter which are estimated by the estimation unit (120) as satisfying the priority condition. [5] Machining condition setting device (1) according to claim 1, wherein the preprocessing unit (36) generates status data (S) only from data obtained by the data acquisition unit (34) and satisfying the priority condition according to the machining type contained in the data and the storage in the priority condition storage unit (56), which state data (S) contain at least tool data (S1) with information about the tool to be used in machining the workpiece by the machine tool and machining condition data (S2) including information about the machining condition when machining the workpiece by the machine tool and / or machining parameter data (S3) including information about the machining parameter to be used in machining the workpiece by the machine tool as data for use in unsupervised learning by the machine learning device (100), and the machine learning device (100) contains a learning unit (110) which, on the basis of the state data (S), generates a learning model which represents a distribution of the machining conditions and / or the machining parameters which satisfy the priority condition during machining by the machine tool. [6] Machining condition setting device (1) according to claim 1, wherein the preprocessing unit (36) generates, as data to be used in the estimation by the machine learning device (100), on the basis of the data obtained by the data acquisition unit (34), state data (S), which contain at least tool data (S1) including information about a tool to be used in the machining of the workpiece by the machine tool and machining condition data (S2) including information about the machining condition during the machining of the workpiece by the machine tool or machining parameter data (S3) including information about the machining parameter during the machining of the workpiece by the machine tool, the machine learning device (100) includes an estimation unit (120) which estimates the processing condition and / or the processing parameter which satisfy the priority condition, wherein the estimation is carried out on the basis of a relationship between the state data (S) and the learning model selected by the learning model selection unit (105), the learning models stored in the learning model storage unit (130) are those learning models which represent a distribution of the machining conditions and / or the machining parameters which satisfy the priority condition during machining by the machine tool, and the machine tool is controlled based on the machining condition and / or the machining parameter which satisfy the priority condition as estimated by the estimation unit (120). [7] A machining condition setting system (500) comprising a plurality of devices connected via a network, wherein the plurality of devices include a first machining condition setting device (1) which is a machining condition setting device (1) according to at least one of claims 1 to 6. [8] Machining condition setting system (500) according to claim 7, wherein the plurality of devices include a computer including a machine learning device (100), the computer obtains at least one of the learning models for the first machining condition setting device (1), and the machine learning device (100) contained in the computer carries out an optimization or adaptation based on the obtained learning model. [9] Machining condition setting system (500) according to claim 7, wherein the plurality of devices comprise a second machining condition setting device (1) different from the first machining condition setting device (1), and a result of learning by the first machining condition setting device (1) is shared with the second machining condition setting device (1). [10] Machining condition setting system (500) according to claim 7, wherein the plurality of devices include a second machining condition setting device (1) different from the first machining condition setting device (1), and data determined by the second machining condition setting device (1) are accessible via a network for learning by the first machining condition setting device (1).

Images