Machine learning device, control system, and machine learning method

Abstract

The present invention provides a machine learning device, a control system, and a machine learning method, which are used to estimate the temperature at a location where a temperature sensor cannot be installed or it is difficult to install a temperature sensor in actual use. The machine learning device has: a virtual temperature model making unit, which has a virtual temperature calculation formula including a first coefficient that determines the heat generation and a second coefficient that determines the heat release, and uses the heat generation cause data to estimate the temperature of a specific location of the machine through the virtual temperature calculation formula and obtain virtual temperature data; and a thermal displacement model making unit, which uses the virtual temperature data obtained using the virtual temperature calculation formula and the actual temperature data obtained from at least one temperature sensor installed at a location other than the specific location to obtain the error between the thermal displacement estimated by the thermal displacement estimation calculation formula and the actually measured thermal displacement, and the virtual temperature model making unit performs machine learning to search for the first coefficient and the second coefficient so as to minimize the error.

Description

Technical Field

[0001] The present invention relates to a machine learning device for performing machine learning to create a virtual temperature model, a control system including the machine learning device, and a machine learning method, wherein the virtual temperature model is used to estimate the temperature of a specific part of a machine where a temperature sensor cannot be installed, or the temperature of a specific part of a machine where it is difficult to install a temperature sensor during actual operation even if a temperature sensor can be installed during data collection. Background Art

[0002] There is a problem that one of the causes of machining errors in machine tools is that thermal expansion of machine elements such as a spindle and a ball screw of a machine tool causes relative thermal displacement between a tool and a workpiece. Machine learning devices or thermal displacement correction devices that solve such problems are described in, for example, Patent Documents 1 to 3.

[0003] Patent document 1 describes a machine learning device having the following parts: a measurement data acquisition unit that acquires a measurement data group; a thermal displacement acquisition unit that acquires the actual measured value of the thermal displacement of a machine element; a storage unit that sets the measurement data group acquired by the measurement data acquisition unit as input data, sets the actual measured value of the thermal displacement of the machine element acquired by the thermal displacement acquisition unit as a label, and stores them in association with each other as training data; and a calculation formula learning unit that performs machine learning based on the measurement data group and the actual measured value of the thermal displacement of the machine element, thereby setting a thermal displacement prediction calculation formula for calculating the thermal displacement of the machine element based on the measurement data group.

[0004] Patent Document 2 describes a thermal displacement correction device as follows: a main shaft portion including a main shaft is considered as a two-dimensional model, the model is divided into 12 sections, and an area for storing the temperature of each section is reserved in a memory of a control device (the temperature of section I at time N is set to T IN ), set appropriate heat radiation coefficient, heat generation coefficient and heat transfer coefficient between adjacent sections according to the number, shape and size of the set sections, and store these coefficients in the memory of the control device. According to the thermal displacement correction device, a thermal displacement correction device for a machine is provided, which can reduce the load for calculation by capturing the main shaft part as a two-dimensional model, and can calculate the correction amount that changes moment by moment with a finer span.

[0005] Patent document 3 describes the following thermal displacement correction device: temperature information is received from temperature sensors 1 of various parts of the machine tool, a neural network operation unit is set up to use neural network constant values ​​to calculate the correction amount of thermal displacement of the tool and the workpiece as training data, and the neural network learning operation unit learns the thermal displacement and temperature information read from the training data storage unit 8, thereby adjusting the neural network constant value.

[0006] Patent Document 1: Japanese Patent Application Publication No. 2018-153901

[0007] Patent Document 2: Japanese Patent Application Publication No. 2015-199168

[0008] Patent Document 3: Japanese Patent Application Laid-Open No. 11-114776 Summary of the invention

[0009] Mechanical elements of machine tools, especially spindles and ball screws, are greatly affected by thermal displacement, and there are places where temperature sensors cannot be installed. If temperature data cannot be obtained using temperature sensors, it is impossible to create an estimation model for thermal displacement, and high-precision estimation cannot be achieved.

[0010] In addition, there are places such as the turret or the table periphery where temperature sensors can be installed during data collection, but installation is not allowed in consideration of actual operation. Therefore, the degree of freedom of expression of the estimation model is reduced.

[0011] (1) A first aspect of the present disclosure is a machine learning device comprising: a virtual temperature model creating unit having a virtual temperature calculation formula, wherein the temperature of a specific part of a machine is estimated by the virtual temperature calculation formula using heat generation cause data and virtual temperature data is obtained, wherein the virtual temperature calculation formula includes a first coefficient that determines a heat generation amount and a second coefficient that determines a heat release amount; and

[0012] a thermal displacement model preparation unit having a thermal displacement estimation calculation formula including a third coefficient, using the virtual temperature data obtained by the virtual temperature calculation formula and actual temperature data obtained from at least one temperature sensor installed at a location other than the specific location, to obtain an error between the thermal displacement estimated by the thermal displacement estimation calculation formula and the actually measured thermal displacement, and performing machine learning to search for the third coefficient to prepare an optimized thermal displacement estimation calculation formula so as to minimize the error,

[0013] The virtual temperature model preparation unit performs machine learning to search for the first coefficient and the second coefficient to prepare an optimized virtual temperature calculation formula so that an error obtained by the optimized thermal displacement estimation amount calculation formula using the virtual temperature data estimated by the virtual temperature calculation formula is minimized.

[0014] (2) A second aspect of the present disclosure is a machine learning device comprising: a virtual temperature model creating unit;

[0015] The virtual temperature model generating unit has a virtual temperature calculation formula including a first coefficient for determining a heat generation amount and a second coefficient for determining a heat release amount.

[0016] The virtual temperature model making unit uses virtual temperature data of a specific part of the machine obtained through the virtual temperature calculation formula using the heat cause data and actual temperature data obtained from at least one temperature sensor installed at the specific part, calculates the error between the virtual temperature data and the actual temperature data, and performs machine learning to search for the first coefficient and the second coefficient to make an optimized virtual temperature calculation formula so as to minimize the error.

[0017] (3) A third aspect of the present disclosure is a control system comprising: the machine learning device of (1) above;

[0018] a thermal displacement correction device that stores a virtual temperature calculation formula and a thermal displacement estimation calculation formula output from the machine learning device, obtains an estimated value of thermal displacement using the thermal displacement estimation calculation formula using virtual temperature data obtained from the virtual temperature calculation formula and actual temperature data obtained from at least one temperature sensor installed at a location other than the specific location, and obtains a thermal displacement correction amount based on the estimated value of thermal displacement; and

[0019] The numerical control device corrects a control command output to a motor control unit that controls the motor, based on the thermal displacement correction amount.

[0020] (4) A fourth aspect of the present disclosure is a machine learning method of a machine learning device, which obtains virtual temperature data by estimating the temperature of a specific part of a machine using heat generation cause data using a virtual temperature calculation formula including a first coefficient that determines a heat generation amount and a second coefficient that determines a heat release amount.

[0021] having a thermal displacement estimation calculation formula including a third coefficient, using the virtual temperature data obtained by the virtual temperature calculation formula and actual temperature data obtained from at least one temperature sensor installed at a location other than the specific location, obtaining an error between the thermal displacement estimated by the thermal displacement estimation calculation formula and the actually measured thermal displacement, performing machine learning to search for the third coefficient to create an optimized thermal displacement estimation calculation formula so as to minimize the error,

[0022] Machine learning for searching the first coefficient and the second coefficient is performed to create an optimized pseudo temperature calculation formula so that an error obtained by the optimized thermal displacement estimation amount calculation formula using the pseudo temperature data estimated by the pseudo temperature calculation formula is minimized.

[0023] (5) The fifth method of the present disclosure is a machine learning method of a machine learning device, which uses virtual temperature data calculated using heat cause data according to a virtual temperature calculation formula and actual temperature data obtained from at least one temperature sensor installed at the specific location, calculates the error between the virtual temperature data and the actual temperature data, and performs machine learning to search for the first coefficient and the second coefficient to create an optimized virtual temperature calculation formula to minimize the error, wherein the virtual temperature calculation formula includes a first coefficient that determines the heat generation and a second coefficient that determines the heat release, and estimates the temperature of a specific location of the machine.

[0024] According to each aspect of the present disclosure, even when a temperature sensor cannot be installed or it is difficult to install a temperature sensor in actual operation, the temperature can be estimated.

[0025] Furthermore, according to each aspect of the present invention, the estimated temperature can be used to determine the thermal displacement of a component such as a machine tool spindle. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 This is a block diagram showing a configuration of a control system during machine learning according to the first embodiment of the present disclosure.

[0027] Figure 2 This is a block diagram showing the configuration of the control system during operation according to the first embodiment of the present disclosure.

[0028] Figure 3 It is a block diagram showing the structure of a thermal displacement correction unit.

[0029] Figure 4 This is a structural diagram showing a model of a machine tool including a spindle.

[0030] Figure 5 This is a block diagram showing the configuration of a virtual temperature model creation unit and a thermal displacement model creation unit.

[0031] Figure 6 This is a diagram for explaining the influence of heat transfer and bearing friction in a motor.

[0032] Figure 7 This is a flowchart showing the operation of the machine learning unit 300 during machine learning.

[0033] Figure 8 This is a block diagram showing a configuration of a control system during machine learning according to a second embodiment of the present disclosure.

[0034] Fig. 9 This is a block diagram showing a configuration of a control system during operation according to the second embodiment of the present disclosure.

[0035] Fig.10This is a structural diagram showing a model of a machine tool including a turret.

[0036] Fig.11 This is a block diagram showing the configuration of a virtual temperature model creation unit and a thermal displacement model creation unit.

[0037] Fig.12 This is a flowchart showing the operation of the machine learning unit 300A during machine learning.

[0038] Fig.13 This is a structural diagram showing a portion of a machine tool including a ball screw.

[0039] Fig.14 This is a block diagram showing another configuration example of the control system. DETAILED DESCRIPTION

[0040] Hereinafter, embodiments of the present disclosure will be described in detail using the drawings.

[0041] (First embodiment)

[0042] First, a control system including the machine learning device of the present disclosure will be described.

[0043] Figure 1 This is a block diagram showing a configuration of a control system during machine learning according to the first embodiment of the present disclosure. Figure 2 This is a block diagram showing the configuration of the control system during operation according to the first embodiment of the present disclosure. Figure 3 It is a block diagram showing the structure of a thermal displacement correction unit. Figure 4 This is a structural diagram showing a model of a machine tool including a spindle. Figure 5 This is a block diagram showing the configuration of a virtual temperature model creation unit and a thermal displacement model creation unit.

[0044] like Figure 1 as well as Figure 2 As shown, the control system 10 includes a numerical control unit 100 such as a CNC (Computerized Numerical Control) device, a thermal displacement correction unit 200, a machine learning unit 300, and a motor control unit 400. The machine learning unit 300 includes a thermal displacement model creation unit 310 and a virtual temperature model creation unit 320. The thermal displacement correction unit 200 may also be included in the numerical control unit 100, and the motor control unit 400 may also be included in the numerical control unit 100.

[0045] The control system 10 may be configured as a single device including the numerical control unit 100, the thermal displacement correction unit 200, the machine learning unit 300, and the motor control unit 400. However, the numerical control unit 100, the thermal displacement correction unit 200, the machine learning unit 300, and the motor control unit 400 may be configured as a single device and connected via a network. In addition, even when the numerical control unit 100, the thermal displacement correction unit 200, the machine learning unit 300, and the motor control unit 400 are configured as a single device, they may be referred to as a numerical control device, a thermal displacement correction device, a machine learning device, and a motor control device.

[0046] The thermal displacement model creation unit 310 and the virtual temperature model creation unit 320 may be separated and each may be provided as a machine learning unit.

[0047] The motor controlled by the motor control unit 400 is provided as a part of a machine such as a machine tool, a robot, or an industrial machine. Figure 1 as well as Figure 2 In the embodiment, the motor control unit 400 controls the motor 510 of the machine tool 500, and the motor 510 drives the driven object 520. The driven object 520 is a spindle in this case. The numerical control unit 100 and the motor control unit 400 may also be provided as part of a machine such as the machine tool 500.

[0048] like Figure 1 as well as Figure 2 As shown, the numerical control unit 100 sends a control signal to the motor control unit 400. The numerical control unit 100 stores a machining program specified according to the machining content of the workpiece. The numerical control unit 100 extracts the conditions of the cutting process (for example, the frequency of spindle acceleration and deceleration, the number of revolutions, the cutting load, and the cutting time) by reading and interpreting the machining program, and outputs the position instruction data to the motor control unit 400. When the numerical control unit 100 is in operation (in operation), the thermal displacement correction amount output from the thermal displacement correction unit 200 is used to correct the cutting process conditions, and then outputs the position instruction data as the control instruction to the motor control unit 400.

[0049] like Figure 3 As shown, the thermal displacement correction unit 200 includes a virtual temperature calculation unit 201 , a temperature data storage unit 202 , a thermal displacement amount calculation unit 203 , and a correction amount calculation unit 204 .

[0050] The virtual temperature calculation unit 201 receives the virtual temperature model from the virtual temperature model creation unit 320 of the machine learning unit 300 after the machine learning of the control system 10, and stores it. The virtual temperature calculation unit 201 receives the heat cause data (current, rotation speed, load, etc.) from the machine tool 500 when the control system 10 is running, calculates the virtual temperature (estimated temperature) of the spindle by using the virtual temperature model based on the heat cause data, and outputs it to the temperature data storage unit 202. Since the spindle rotates, it is impossible to install a temperature sensor to actually measure the spindle temperature. Therefore, the virtual temperature calculation unit 201 obtains the virtual temperature of the spindle by using the virtual temperature model.

[0051] Temperature data storage unit 202 stores virtual temperatures (estimated temperatures) and also stores temperature data output from machine tool 500. The temperature data is actual temperature data actually measured using a temperature sensor at a location other than the main spindle.

[0052] The thermal displacement amount calculation unit 203 receives the thermal displacement model from the thermal displacement model creation unit 310 of the machine learning unit 300 after machine learning and stores it. The thermal displacement amount calculation unit 203 reads the temperature data and the virtual temperature from the temperature data storage unit 202 when the control system 10 is running, and calculates the estimated thermal displacement amount based on the temperature data and the thermal displacement model.

[0053] The correction amount calculation unit 204 outputs the estimated amount of thermal displacement to the numerical control unit 100 as the correction amount of thermal displacement.

[0054] like Figure 4 As shown, the machine tool 500 includes: a motor 510 serving as a spindle motor, a spindle 520A serving as a driven body 520 to which a tool 530 is mounted and rotated by the motor 510, a mounting table 540 to which the spindle 520A is mounted, a column 550, a lathe 560, and a reciprocating table 570. The workpiece 600 is processed on the reciprocating table 570 by the tool 530. Figure 4 In the figure, the black circles indicate the places where the temperature sensor was actually measured outside the spindle.

[0055] <Machine Learning Department 300>

[0056] The machine learning unit 300 is composed of a thermal displacement model creation unit 310 and a virtual temperature model creation unit 320 .

[0057] The virtual temperature model creation unit 320 learns, by reinforcement learning, coefficients of a virtual spindle temperature calculation formula that is a virtual temperature model for obtaining a virtual temperature (virtual spindle temperature) of a spindle in the machine tool 500. The spindle is a specific part, and the virtual spindle temperature calculation formula is a virtual temperature calculation formula.

[0058] Furthermore, the thermal displacement model creation unit 310 of the machine learning unit 300 learns coefficients of a thermal displacement estimation amount calculation formula that is a thermal displacement model for obtaining an estimation amount of thermal displacement in the machine tool 500 (thermal displacement estimation amount) through supervised learning.

[0059] Specifically, the thermal displacement model creation unit 310 uses the virtual spindle temperature output from the virtual temperature model creation unit 320 and the measured temperature of the parts other than the spindle, such as the mounting table 540 and the column (pillar) 550. Figure 4 The measured temperature of the black circle is used to estimate the thermal displacement by the thermal displacement estimation formula. In addition, the thermal displacement model preparation unit 310 obtains the error between the estimated thermal displacement and the measured thermal displacement, searches for the coefficient of the thermal displacement estimation formula, and prepares the thermal displacement estimation formula to minimize the error.

[0060] The virtual temperature model generator 320 estimates the spindle temperature using a virtual spindle temperature calculation formula using the heat generation factor data (current, rotation speed, load, etc.) of the machine tool 500. The virtual temperature model generator 320 learns the coefficients of the spindle temperature estimation formula so that the error output from the thermal displacement model generator 310 is minimized.

[0061] In this way, the machine learning unit 300 obtains the coefficients of the optimized spindle temperature estimation formula and the coefficients of the optimized thermal displacement estimation calculation formula in which the error between the thermal displacement estimated by the thermal displacement estimation formula and the actually measured thermal displacement is minimized, and outputs them to the thermal displacement correction unit 200 as a virtual temperature model and a thermal displacement model.

[0062] Below, use Figure 5 The configurations of the thermal displacement model creation unit 310 and the virtual temperature model creation unit 320 will be described.

[0063] First, the virtual temperature model creation unit 320 will be described below.

[0064] The virtual temperature model creation unit 320 includes a heat generation factor data acquisition unit 321 , an error acquisition unit 322 , a storage unit 323 , a learning unit 324 , and a virtual spindle temperature calculation unit 325 .

[0065] The heating factor data acquisition unit 321 acquires heating factor data from the machine tool 500. Here, the heating factor data refers to the drive current of the motor 510, the rotation speed of the spindle of the machine tool 500, the load of the motor 510, the motor temperature, the air temperature, and the like.

[0066] The error acquisition unit 322 acquires the difference between the estimated value of the thermal displacement amount and the actual value of the thermal displacement amount, that is, the error, from the error calculation unit 316 of the thermal displacement model creation unit 310 described later. The error output from the thermal displacement model creation unit 310 is described in the description of the thermal displacement model creation unit 310.

[0067] The storage unit 323 stores the heating cause data acquired by the heating cause data acquisition unit 321 and the error acquired by the error acquisition unit 322 in association with each other.

[0068] The learning unit 324 reads the heating cause data and the error from the storage unit 323, and learns the coefficients A, B, C, and D of the virtual spindle temperature calculation formula based on the read heating cause data so as to minimize the error, thereby setting a virtual spindle temperature calculation formula for calculating the virtual spindle temperature based on the heating cause data.

[0069] The time t is expressed by Mathematical Formula 1 (hereinafter Mathematical Formula 1): i The virtual spindle temperature θ V (t i ).

[0070]

Mathematical formula 1

[0071] θ V (t i )=θ V (t i-1 )+T S ×Q′(t i-1 )-T S ×A×{θ V (t i-1 0-θ r (t i-1 )}

[0072] Mathematical formula 1 T S is the calculation period of the virtual spindle temperature, Q' is the heat generation, A is the coefficient, θ r (t i-1 ) is the previous time t i-1 of atmospheric temperature.

[0073] The calorific value Q' is expressed by Mathematical Formula 2 (hereinafter referred to as Mathematical Formula 2).

[0074]

Mathematical formula 2

[0075]

[0076] In the mathematical formula 2, B represents the coefficient of heat transfer of the motor, C and D represent the coefficient of bearing friction that may cause damage to the machine, and θ M (t i-1) represents the motor temperature, S(t i-1 -t') indicates the spindle speed.

[0077] The initial values ​​of coefficients A, B, C, and D are set in advance, and correction is performed by reinforcement learning to search for coefficients A, B, C, and D. Coefficient A is the second coefficient that determines the amount of heat release, and coefficients B, C, and D are coefficients that determine the amount of heat release.

[0078] like Figure 6 As shown, the spindle temperature θ V (t i-1 ) is affected by the heat transfer of the motor and the bearing friction.

[0079] The virtual spindle temperature calculation unit 325 obtains the virtual spindle temperature from the heat generation factor data using the virtual spindle temperature calculation formulas represented by Mathematical Formula 1 and Mathematical Formula 2 set by the learning unit 324 .

[0080] <Example of reinforcement learning>

[0081] In the following, reinforcement learning is described, but reinforcement learning is known, and its details are described in, for example, Japanese Patent Publication No. 2018-152012, Japanese Patent Publication No. 2019-021024, Japanese Patent Publication No. 2019-021235, etc. Therefore, in the following description, an overview of reinforcement learning is described.

[0082] The basic structure of reinforcement learning is that the agent observes the state of the environment, selects a certain behavior, and the environment changes according to the behavior. As the environment changes, a certain reward is given, and the agent learns to choose better behaviors (decisions). In most cases, the reward is a fragment value based on the change of a part of the environment, and the agent learns to choose behaviors so that the total future rewards are maximized.

[0083] Here, any learning method can be used as reinforcement learning. In the following description, the case of using Q-learning in a certain environmental state S is used as an example. The Q-learning is a method of learning the value Q(S, A) of selecting behavior A.

[0084] The purpose of Q learning is to select the behavior A with the highest value Q(S, A) as the best behavior from the available behaviors A under a certain state S.

[0085] However, at the time when Q learning first begins, the correct value of the value Q(S, A) for the combination of state S and behavior A is completely unknown. Therefore, the agent selects various behaviors A in a certain state S, and for the behavior A at that time, it selects a better behavior based on the reward given, thereby continuing to learn the correct value Q(S, A).

[0086] For example, the virtual temperature model making unit 320 of the machine learning unit 300 performs the above Q learning. Specifically, the virtual temperature model making unit 320 learns the following value Q: the error output from the thermal displacement model making unit 310 and the heating cause data obtained from the machine tool 500 are selected as the state S, and the adjustment of the coefficients A, B, C, and D of the virtual spindle temperature calculation formula related to the state S is selected as the behavior A. In the virtual temperature model making unit 320, a reward is returned each time the behavior A is performed. The reward is determined based on the error. The virtual temperature model making unit 320 searches for the best behavior A by trial and error, for example, so that the total of future rewards is maximized. In this way, the virtual temperature model making unit 320 can select the best behavior A (i.e., the coefficients A, B, C, and D of the virtual spindle temperature calculation formula) for the heating cause data, i.e., the state S.

[0087] The calculation of the reward in the virtual temperature model production unit, for example, sets the absolute value of the error and the square of the absolute value of the error as the evaluation function f. When the evaluation function value f(S') related to the state S' after correction by behavior A is larger than the evaluation function f(S) before correction related to the state information S before correction by behavior information A, it is set to a negative value. When the evaluation function value f(S') is smaller than the evaluation function f(S), the reward value is set to a positive value.

[0088] When the evaluation function f(S') is equal to the evaluation function f(S), the reward value is set to zero.

[0089] Q learning is performed based on the state S, behavior A, the state S' when behavior A is applied to the state S, and the value of the reward calculated as described above, thereby updating the value function.

[0090] Then, based on the value function updated by Q learning, a behavior in which the value function Q(S, A) is maximized is generated as an optimized behavior.

[0091] Next, the thermal displacement model creation unit 310 will be described below.

[0092] The thermal displacement model creation unit 310 includes a displacement acquisition unit 311 , a temperature acquisition unit 312 , a virtual spindle temperature acquisition unit 313 , a storage unit 314 , a learning unit 315 , and an error calculation unit 316 .

[0093] The displacement acquisition unit 311 acquires an actually measured value of the thermal displacement amount of the main shaft of the machine tool 500 detected by, for example, a probe.

[0094] The temperature acquisition unit 312 acquires the actual temperature of the parts other than the main shaft measured by the temperature sensor, for example, the temperature of the mounting table 540 and the column (pillar) 550. Figure 4The black circles are shown as the measured temperatures.

[0095] The virtual spindle temperature acquisition unit 313 acquires the virtual spindle temperature calculated by the virtual spindle temperature calculation unit 325 .

[0096] The storage unit 314 stores the measured value of the thermal displacement acquired by the displacement acquisition unit 311 , the measured temperature of a portion other than the spindle acquired by the temperature acquisition unit 312 , and the virtual spindle temperature acquired by the virtual spindle temperature acquisition unit 313 in association with each other.

[0097] Table 1 shows the measured temperatures θ1 to θ2 at time t from the temperature sensors installed at n locations other than the main axis. n and the virtual spindle temperature θ v of table.

[0098]

Table 1

[0099] Time t <![CDATA[θ1]]> <![CDATA[θ2]]> <![CDATA[θ3]]> … <![CDATA[θ n-1 ]]> <![CDATA[θ n ]]> <![CDATA[θ v ]]> <![CDATA[t1]]> 21 22 31 25 18 30 <![CDATA[t2]]> 22 24 32 26 19 32 …

[0100] The learning unit 315 performs supervised learning based on the measured value of the thermal displacement, the measured temperature of the parts other than the main shaft, and the virtual main shaft temperature, thereby setting a thermal displacement estimation calculation formula that calculates the estimated amount of thermal displacement (thermal displacement estimation amount) based on the measured temperature of the parts other than the main shaft and the virtual main shaft temperature.

[0101] More specifically, the learning unit 315 uses the multivariate regression of the generalized linear model to obtain the actual measured temperatures θ1 to θ2 from the temperature sensors installed at n locations other than the main axis during a predetermined period stored in the storage unit 314. n and the virtual spindle temperature θ v The difference between the calculated estimated thermal displacement amount and the measured value of the thermal displacement amount of the spindle in a predetermined period stored as a label in the storage unit 13 is minimized by, for example, setting an optimized thermal displacement estimated amount calculation formula by the least square method. n and the virtual spindle temperature θ v The estimated thermal displacement is the input data, the measured value of the thermal displacement is the label, and the combination of the label and the input data is the training data.

[0102] Specifically, the learning unit 315 uses the actual measured temperatures θ1 to θn from the temperature sensors provided at n locations other than the spindle and the virtual spindle temperature θ v The estimated value of the thermal displacement is f(θ1, θ2, ..., θ n ,θ V )(n is a natural number), the measured value of thermal displacement is set as Y LWhen setting f(θ1, θ2, ..., θ n ,θ V ) and Y L The thermal shift estimation calculation formula was optimized so that the difference was minimized.

[0103] The error calculation unit 316 outputs the difference between the estimated thermal displacement amount calculated by the thermal displacement estimated amount calculation formula set by the learning unit 315 and the actual measured value of the thermal displacement amount as an error to the error acquisition unit 322 .

[0104] <Example of machine learning method>

[0105] The learning unit 315 uses the training data to implement machine learning, and an example of this method is described in detail in Japanese Patent Application Laid-Open No. 2018-153901.

[0106] A method for setting a calculation formula for estimating thermal displacement is described in, for example, Japanese Patent Application Publication No. 2018-153901.

[0107] For example, as a method of searching for coefficients of the thermal displacement estimation formula and setting the thermal displacement estimation formula, it is possible to infer the thermal displacement estimation formula Y = a1θ1 + a2θ2 + ... + a by machine learning using the least squares method based on the multiple regression of the generalized linear model. n θ n +a V θ V The estimated value of the calculated thermal displacement (the estimated thermal displacement) and the measured value of the thermal displacement Y L The coefficient is set to minimize the square error of . Here, Y is the estimated value of thermal displacement, θ1, θ2, ..., θ n is the measured temperature, θ v is the virtual spindle temperature, a1, a2…a n 、a V are the coefficients determined through multiple regression.

[0108] Specifically, let the estimated thermal displacement value be Y and the label be Y L , the measured temperatures θ1, θ2, …, θ n and the virtual spindle temperature θ v When the input data is set, in the mathematical formula 3 (hereinafter referred to as the mathematical formula 3), the coefficient a is calculated so that the total value of the plurality of training data is the smallest. k 、a v In addition, k is a natural number, n is an arbitrary integer, k≤n. Coefficient a k 、av is the third coefficient.

[0109]

Mathematical formula 3

[0110]

[0111] The method for setting the thermal displacement estimation calculation formula is not limited to the above method, and various methods described in Japanese Patent Publication No. 2018-153901 can be used. For example, instead of the usual multiple regression analysis, a multiple regression analysis taking into account the L2 regularization term can be implemented. In addition, sparse regularization can be implemented. For example, a multiple regression analysis taking into account the L1 regularization term can be implemented.

[0112] In addition, Japanese Patent Application Publication No. 2018-153901 states that as input data for executing the above-mentioned machine learning, a first-order delay element of the measurement data and a time offset element of the measurement data can also be used. Japanese Patent Application Publication No. 2018-153901 states a thermal displacement amount prediction calculation formula using a first-order delay element of the measurement data and a thermal displacement amount prediction formula using a time offset element of the measurement data.

[0113] Furthermore, Japanese Patent Application Publication No. 2018-153901 also describes, as another method, the following content: machine learning involving known neural networks such as single-layer neural networks or multi-layer neural networks can be implemented.

[0114] <Machine Learning Operation>

[0115] Next, the operation of the machine learning unit 300 according to the present embodiment during machine learning will be described. Figure 7 2 is a flowchart showing the operation of the machine learning unit 300 during the machine learning.

[0116] In step S11 , the learning unit 324 of the virtual temperature model creation unit 320 randomly sets coefficients A, B, C, and D of a virtual spindle temperature calculation formula serving as a virtual temperature model at the start of machine learning and during machine learning.

[0117] In step S12, the heat generation factor data acquisition unit 321 acquires the motor temperature θ from the machine tool 500. M , spindle speed S, and heat generation data such as the atmospheric temperature θr. The virtual spindle temperature calculation unit 325 uses the virtual spindle temperature calculation formulas obtained by setting coefficients in the learning unit 324, namely, Mathematical Formula 1 and Mathematical Formula 2, to obtain the virtual spindle temperature.

[0118] In step S13, the learning unit 315 of the thermal displacement model creation unit 310 sets the coefficients of the thermal displacement estimation amount calculation formula so that the estimated value f (θ1, θ2, ..., θ n ,θ V ) and the measured value of thermal displacement Y L The difference (error) is minimal.

[0119] In step S14, the error calculation unit 316 sets the estimated value of the thermal displacement amount to f(θ1, θ2, ..., θ n ,θ V )(n is a natural number), the measured value of thermal displacement is set as Y L When f(θ1, θ2, …, θ n ,θ V ) and Y L of error.

[0120] In step S15, when the error obtained from the error calculation unit 316 has not converged, or when the search of the coefficients A, B, C, and D does not meet the specified number of times, the learning unit 324 of the virtual temperature model preparation unit 320 returns to step S11. On the other hand, when the error converges and the error does not exist or is within a certain range, or when the search of the coefficients A, B, C, and D reaches the specified number of times, the learning is terminated and the virtual temperature model is output to the thermal displacement correction unit 200. The virtual temperature model output at the end of the learning is a virtual temperature model using the coefficients A, B, C, and D of the combination with the smallest error during the search process. The thermal displacement model is output from the learning unit 315 of the thermal displacement model preparation unit 310 to the thermal displacement correction unit 200.

[0121] In addition, the storage unit 314 and the storage unit 323 of the machine learning unit 300 store the virtual spindle temperature calculation formula and the thermal displacement estimation calculation formula. Therefore, when the virtual spindle temperature calculation formula and the thermal displacement estimation calculation formula are requested from the newly installed thermal displacement correction unit 200, the virtual spindle temperature calculation formula and the thermal displacement estimation calculation formula can be sent to the thermal displacement correction unit. In addition, when new training data is obtained, additional machine learning can also be performed.

[0122] <Effects obtained by the first embodiment>

[0123] As described above, in this embodiment, even when a temperature sensor cannot be installed to actually measure the temperature, the temperature can be estimated. Furthermore, the estimated temperature can be used to obtain the estimated amount of thermal displacement of a component such as a spindle of a machine tool.

[0124] (Second embodiment)

[0125] In the first embodiment, a machine learning device is described as follows: a first coefficient of heat generation that determines a virtual spindle temperature calculation formula for obtaining a virtual spindle temperature and a second coefficient that determines a heat release are machine-learned, and a coefficient of a thermal displacement estimation calculation formula that uses the obtained virtual spindle temperature as a parameter to obtain a thermal displacement estimation is machine-learned. In the present embodiment, a machine learning device is described as follows: a first coefficient of heat generation that determines a virtual turret temperature calculation formula for obtaining a virtual turret temperature and a second coefficient that determines a heat release are machine-learned, and a coefficient of a thermal displacement estimation calculation formula that uses the obtained virtual turret temperature as a parameter to obtain a thermal displacement estimation is obtained. The so-called turret is a rotating tool stand (two or more tools are radially mounted).

[0126] Because the spindle rotates, a temperature sensor cannot be installed to actually measure the spindle temperature. On the other hand, a temperature sensor cannot be installed on the turret during actual operation, but a temperature sensor can be installed during data collection. Therefore, when machine learning is performed on the first coefficient that determines the heat generation and the second coefficient that determines the heat generation of the virtual turret temperature calculation formula used to calculate the virtual turret temperature, the actual measured value of the turret temperature can be used.

[0127] In addition, in this embodiment, as described later, Fig.12 As shown in steps S25 to S27 of , the thermal displacement model making unit obtains the error between the measured value of thermal displacement and the estimated value of thermal displacement to search for the coefficient of the thermal displacement estimation calculation formula and make the thermal displacement model, but is not limited to this. For example, the thermal displacement model making unit may also set the coefficient of the thermal displacement estimation calculation formula to make the thermal displacement model. In this case, the thermal displacement model described below is not required. Fig.12 Step S26 and step S27.

[0128] As in the first embodiment, the method of setting the coefficients of the thermal displacement estimation formula can be inferred and set by machine learning using the least squares method based on the multivariate regression of the generalized linear model, using the thermal displacement estimation formula Y=a1θ1+a2θ2+…a n θ n +a V θ VL The estimated value of the thermal displacement (the estimated thermal displacement) and the measured value of the thermal displacement Y L Here, Y is the estimated value of thermal displacement, θ1, θ2, ..., θ n is the measured temperature, θ vL is the virtual turret temperature, a1, a2…a n 、a V is a coefficient determined by multiple regression. In addition, the virtual turret temperature θ vLThis is the virtual turret temperature obtained by a virtual turret temperature calculation formula optimized by learning in a virtual temperature model creation unit 320A described later.

[0129] Figure 8 This is a block diagram showing a configuration of a control system during machine learning according to a second embodiment of the present disclosure. Fig. 9 This is a block diagram showing a configuration of a control system during operation according to the second embodiment of the present disclosure. Fig.10 This is a structural diagram showing a machine tool model including a turret. Fig.11 This is a block diagram showing the configuration of a virtual temperature model creation unit and a thermal displacement model creation unit.

[0130] like Figure 8 as well as Fig. 9 As shown, the control system 10A of the present embodiment inputs the actual temperature from the temperature sensor attached to the turret of the machine tool 500 to the virtual temperature model creation unit 320A. Figure 1 as well as Figure 2 In the control system 10 shown, the error is input from the thermal displacement model creation unit 310 to the virtual temperature model creation unit 320, but in this embodiment, the error is not input from the thermal displacement model creation unit 310A to the virtual temperature model creation unit 320A. The machine learning unit 300A may not include the thermal displacement model creation unit 310A, but only include the virtual temperature model creation unit 320A.

[0131] Compared with the thermal displacement model making unit 310, the thermal displacement model making unit 310A of the machine learning unit 300A has the virtual spindle temperature replaced by the virtual turret temperature, so the virtual spindle temperature acquisition unit 313 is replaced by the virtual turret temperature acquisition unit 313A, and the error calculation unit 316 is not provided. If this aspect is excluded, the structure and operation of the thermal displacement model making unit 310A are the same as the structure and operation of the thermal displacement model making unit 310, so the description is omitted. In addition, the method of performing machine learning on the coefficients of the thermal displacement estimation calculation formula of the thermal displacement model making unit 310A is the same as the method of performing machine learning on the coefficients of the thermal displacement estimation calculation formula of the thermal displacement model making unit 310, so the description is omitted.

[0132] Table 2 shows the measured temperatures θ1 to θ2 at time t from the temperature sensors installed at n locations other than the turret and stored in the storage unit 314. n and the virtual turret temperature θ vL of table.

[0133]

Table 2

[0134]

[0135]

[0136] The virtual temperature model creation unit 320A of the machine learning unit 300A is different from the virtual temperature model creation unit 320 in that the error acquisition unit 322 is replaced by an actual temperature acquisition unit 326 , and the virtual spindle temperature calculation unit 325 is replaced by a virtual turret temperature calculation unit 325A.

[0137] The virtual temperature model creation unit 320A includes a heat generation factor data acquisition unit 321 , an actual temperature acquisition unit 326 , a storage unit 323 , a learning unit 324A, and a virtual turret temperature calculation unit 325A.

[0138] The heating factor data acquisition unit 321 acquires heating factor data from the machine tool 500. Here, the heating factor data refers to the drive current of the motor 510, the rotation speed of the spindle of the machine tool 500, the load of the motor 510, the motor temperature, the air temperature, and the like.

[0139] The actual temperature acquisition unit 326 acquires the actual temperature of the turret measured by the temperature sensor, for example, Fig.10 The measured temperature of the temperature measurement position within the dotted line area of ​​the turret shown in FIG. Fig.10 As shown, there are: a turret 580A, a column (pillar) 550A on which the turret 580A is mounted, a lathe 560A, and a reciprocating table 570A. A motor is installed in the turret 580A. The dotted line area of ​​the turret is a specific part.

[0140] The storage unit 323 stores the heat generation cause data acquired by the heat generation cause data acquisition unit 321 and the actual temperature data of the turret acquired by the actual temperature acquisition unit 326 in association with each other. Table 3 shows the actual temperature θ measured by the temperature sensor installed in the turret at time t. T and the virtual turret temperature θ v of table.

[0141]

Table 3

[0142] Time t <![CDATA[θ1]]> <![CDATA[θ2]]> <![CDATA[θ3]]> … <![CDATA[θ n-1 ]]> <![CDATA[θ n ]]> <![CDATA[θ vL ]]> <![CDATA[t1]]> 21 22 31 25 18 35 <![CDATA[t 12 ]]> 22 24 32 26 19 37 …

[0143] The learning unit 324A obtains the error between the virtual turret temperature data obtained using the heating cause data and the actual temperature data of the turret, performs supervised learning so that the error converges, thereby searching for coefficients of a virtual turret temperature calculation formula for calculating the virtual turret temperature based on the heating cause data, and sets the virtual turret temperature calculation formula. The heating cause data is input data, the actual temperature data of the turret is a label, and a combination of the label and the input data is training data.

[0144] At a certain moment t i The virtual turret temperature θ V (t i) is expressed by Mathematical Formula 4 (hereinafter referred to as Mathematical Formula 4) in which the virtual spindle temperature is replaced by the virtual turret temperature.

[0145]

Mathematical formula 4

[0146] θ V (t i )=θ V (t i-1 )+T S ×Q′(t i-1 )-T S ×E×{θ V (t i-1 )-θ r (t i-1 )}

[0147] Mathematical formula 4 T S is the calculation period of the turret temperature, Q' is the calorific value, E is the coefficient, θ r (t i-1 ) is the previous time t i-1 The calorific value Q' is expressed by Mathematical Formula 5 (hereinafter Mathematical Formula 5). In addition, the calorific value Q' may also increase due to other reasons.

[0148]

Mathematical formula 5

[0149] Q'(t i-1 )=F×{load}+G×{spindle speed}+H×{motor temperature}

[0150] In the mathematical formula 5, F represents the coefficient of the motor load, G represents the coefficient of the spindle speed, and H represents the coefficient of the motor temperature. The coefficient E is the second coefficient that determines the heat release, and the coefficients F, G, and H are coefficients that determine the heat release.

[0151] The coefficients E, F, G, and H are preset, and the learning unit 324A searches for the calculated virtual turret temperature θ V The time series data and the actual temperature of the turret θ T The coefficients E, F, G, and H with the smallest error in the data.

[0152] The method of searching for coefficients of the virtual turret temperature calculation formula and setting the virtual turret temperature calculation formula can use, for example, the method described in Japanese Patent Application Laid-Open No. 2018-153901 as described in the method of setting the thermal displacement estimation calculation formula of the first embodiment.

[0153] Specifically, as a method for searching for coefficients of the virtual turret temperature calculation formula, the virtual turret temperature θ calculated by the virtual turret temperature calculation formula of Formula 4 can be inferred and set by machine learning using the least squares method based on multivariate regression of a generalized linear model.V and the measured temperature θ T The square error (=(θ T -θ V ) 2 )The smallest coefficients E, F, G, H.

[0154] The method of setting the virtual turret temperature calculation formula is not limited to the above method, and various methods described in Japanese Patent Application Laid-Open No. 2018-153901 may be used, as in the method of setting the thermal displacement estimation calculation formula of the first embodiment already described.

[0155] The optimized virtual turret temperature calculation formula having the searched coefficients E, F, G, and H is output to the thermal displacement correction unit 200 as a virtual temperature model.

[0156] The virtual turret temperature calculation unit 325A calculates the virtual turret temperature θ using the virtual turret temperature calculation formula having the searched coefficients E, F, G, and H set by the learning unit 324A. VL The virtual turret temperature acquisition unit 313A of the thermal displacement model creation unit 310A acquires the virtual turret temperature θ from the virtual turret temperature calculation unit 325A. VL .

[0157] <Machine Learning Operation>

[0158] Next, the operation of the machine learning unit 300A according to the present embodiment during machine learning will be described. Fig.12 300A is a flowchart showing the operation of the machine learning unit 300A during the machine learning.

[0159] In step S21 , the learning unit 324A of the virtual temperature model creation unit 320A randomly sets coefficients E, F, G, and H of a virtual turret temperature calculation formula serving as a virtual temperature model at the start of machine learning and during machine learning.

[0160] In step S22, heat factor data acquisition unit 321 acquires heat factor data such as spindle speed, motor 510 load, and motor temperature from machine tool 500A. Learning unit 324A inserts the heat factor data into a virtual turret temperature calculation formula to obtain (estimate) a virtual turret temperature (virtual temperature data).

[0161] In step S23 , the learning unit 324A calculates the error between the virtual turret temperature and the actual temperature acquired from the actual temperature acquisition unit 326 .

[0162] In step S24, when the error does not converge, or when the coefficients E, F, G, and H are searched less than the specified number of times, the learning unit 324A returns to step S21. On the other hand, when the error converges and the error does not exist or is within a certain range, or when the coefficients E, F, G, and H are searched for the specified number of times, the learning is terminated, and the virtual turret temperature calculation unit 325A uses the virtual turret temperature calculation formula obtained by setting the coefficients E, F, G, and H in the learning unit 324A, that is, mathematical formula 4 and mathematical formula 5 to calculate the virtual turret temperature θ VL The learned virtual turret temperature calculation formula is output to the thermal displacement correction unit 200 as a virtual temperature model.

[0163] In step S25, the learning unit 315 of the thermal displacement model creation unit 310A sets coefficients of a thermal displacement estimation amount calculation formula using the virtual turret temperature θ VL and the measured temperatures θ1 to θ n The estimated value f(θ1, θ2, ..., θ n ,θ VL ) and the measured value of thermal displacement Y L The calculation formula with the smallest difference.

[0164] In step S26, the learning unit 315 sets the estimated value of the thermal displacement amount to f(θ1, θ2, ..., θ n ,θ VL )(n is a natural number), the measured value of thermal displacement is set as Y L When f(θ1, θ2, …, θ n ,θ VL ) and Y L of error.

[0165] In step S27, if the error has not converged or the coefficient search has not reached the specified number of times, the learning unit 315 returns to step S25. On the other hand, if the error has converged and the error does not exist or is within a certain range, or the coefficient search has reached the specified number of times, the learning is terminated and the thermal displacement model is output to the thermal displacement correction unit 200.

[0166] In addition, the storage unit 314 and the storage unit 323 of the machine learning unit 300A store the virtual turret temperature calculation formula and the thermal displacement estimation calculation formula. Therefore, when the virtual turret temperature calculation formula and the thermal displacement estimation calculation formula are requested from the newly installed thermal displacement correction unit 200, the virtual turret temperature calculation formula and the thermal displacement estimation calculation formula can be sent to the thermal displacement correction unit. In addition, when new training data is obtained, other machine learning can also be performed.

[0167] <Effects obtained by the second embodiment>

[0168] As described above, in this embodiment, even if a temperature sensor can be installed during data collection, when a temperature sensor cannot be installed during actual operation, the temperature can be estimated. Furthermore, the estimated temperature can be used to obtain the thermal displacement correction amount of the components such as the spindle of the machine tool.

[0169] The functional blocks included in the machine learning units 300 and 300A according to the first and second embodiments have been described above.

[0170] In order to realize these functional blocks, the machine learning unit 300, 300A has a CPU (Central Processing Unit) and other computing devices. In addition, the machine learning unit 300, 300A also has an auxiliary storage device such as a HDD (Hard Disk Drive) that stores various control programs such as application software and OS (Operating System), and a main storage device such as a RAM (Random Access Memory) that stores data temporarily required after the computing device executes the program.

[0171] Furthermore, in the machine learning unit 300, 300A, the operation processing device reads the application software and OS from the auxiliary storage device, expands the read application software and OS in the main storage device, and performs operation processing based on these application software and OS. In addition, according to the operation result, various hardware possessed by each device is controlled. In this way, the functional blocks of this embodiment are realized. In other words, this embodiment can be realized by the cooperation of hardware and software.

[0172] In addition, the thermal displacement correction unit may also include a machine learning unit 300 or a machine learning unit 300A. In this case, arithmetic processing devices such as a CPU (Central Processing Unit), an auxiliary storage device, and a main storage device are shared, and there is no need to set up separate ones for the machine learning unit 300 or the machine learning unit 300A.

[0173] As for the machine learning units 300 and 300A, since the amount of computation accompanying machine learning increases, for example, by using a technology called GPGPU (General-Purpose Computing on Graphics Processing Units) equipped with a GPU (Graphics Processing Units) in a personal computer, high-speed processing can be performed when the GPU is used for computational processing accompanying machine learning. In addition, in order to perform even higher-speed processing, a computer cluster can be constructed using multiple computers equipped with such GPUs, and parallel processing can be performed by the multiple computers included in the computer cluster.

[0174] In the first and second embodiments, the machine tool is described as an example of a machine, but the machine is not limited to the machine tool, and may be a robot, an industrial machine, or the like.

[0175] In the first embodiment, the spindle is cited as an example where a temperature sensor cannot be installed to actually measure the spindle temperature, and in the second embodiment, the turret is cited as an example where a temperature sensor cannot be installed in actual operation but can be installed during data collection. However, the present invention is not limited to these examples, and can also be applied to Fig.13 The ball screw shown is used as the driven body. Fig.13 This is a structural diagram showing a portion of a machine tool including a ball screw.

[0176] Fig.13 This is a block diagram showing a portion of a machine tool including a motor and a ball screw.

[0177] The motor control unit 400 moves the worktable 522 through the motor 510 via the connection mechanism 521 of the driven body 520, thereby processing the workpiece (workpiece) mounted on the worktable 522. The connection mechanism 521 has a coupler 5211 connected to the motor 510, a ball screw 5212 fixed to the coupler 5211, and a nut 5213 screwed with the ball screw 5212. The rotation angle position of the motor 510 is detected by the rotary encoder 511 associated with the motor 510, and the detection speed (actual speed) obtained by the rotation angle position is fed back to the motor control unit 400 (speed FB).

[0178] The screw axis of the ball screw 5212 cannot be installed with a temperature sensor in area R1 of the screw axis to actually measure the temperature due to rotation, so the structure of the first embodiment can be applied. In addition, a temperature sensor can be installed in area R2 of the bearing of the screw axis of the ball screw 5212 during data collection, but a temperature sensor cannot be installed during actual operation, so the structure of the second embodiment can be applied.

[0179] Each structural unit included in the above-mentioned machine learning unit can be implemented by hardware, software or a combination thereof. In addition, the machine learning method performed by the cooperation of each of the structural units included in the above-mentioned machine learning unit can also be implemented by hardware, software or a combination thereof. Here, the so-called implementation by software means that the computer executes the program by reading it.

[0180] Various types of non-transitory computer readable media (non-transitory computer readable medium) can be used to store the program and provide it to the computer. Non-transitory computer readable media include various types of tangible storage medium. Examples of non-transitory computer readable media include: magnetic storage media (e.g., floppy disks, magnetic tapes, hard disk drives), magneto-optical storage media (e.g., optical magnetic disks), CD-ROM (Read Only Memory), CD-R, CD-R / W, semiconductor memories (e.g., mask ROM, PROM (Programmable ROM), EPROM (Erasable PROM), flash ROM, RAM (random access memory)). In addition, the program can be supplied to the computer via various types of transitory computer readable media (transitory computer readable medium).

[0181] The above-described embodiment is a preferred embodiment of the present invention. However, the scope of the present invention is not limited to the above-described embodiment, and the present invention can be implemented in various modified forms without departing from the spirit of the present invention.

[0182] The structure of the control system Figure 1 and Figure 2 ,as well as Figure 8 and Fig. 9 In addition to the structure, the following structure also exists.

[0183] <Modification example in which the machine learning device is provided outside the control system via a network>

[0184] Fig.14 This is a block diagram showing another configuration example of the control system. Fig.14 The control system 10B shown is Figure 1 and Figure 2 The control system 10 shown and Figure 8 and Fig. 9 The control system 10A shown is different in that n (n is a natural number greater than or equal to 2) machine learning devices 300 - 1 to 300 - n , n control devices 700 - 1 to 700 - n , and machine tools 500 - 1 to 500 - n are connected via a network 800 .

[0185] Each of the control devices 700-1 to 700-n has Figure 1 as well as Figure 2 ,or Figure 8 as well as Fig. 9The numerical control unit 100, the thermal displacement correction unit 200, and the motor control unit 400 are shown. Each of the machine learning devices 300-1 to 300-n has Figure 5 The structure of the machine learning unit 300 or the machine learning unit 300A shown in the figure is the same as the structure of the machine learning unit 300.

[0186] Here, the control device 700-1, the machine tool 500-1, and the machine learning device 300-1 are connected in a one-to-one pair so as to be able to communicate. The control devices 700-2 to 700-n, the machine tools 500-2 to 500-n, and the machine learning devices 300-2 to 300-n are also connected in the same manner as the control device 700-1, the machine tool 500-1, and the machine learning device 300-1. Fig.12 In the embodiment, the n groups of control devices 700-1 to 700-n and machine tools 500-1 to 500-n and machine learning devices 300-1 to 300-n are connected via a network 800, but may be directly connected via a connection interface. For example, multiple groups of these control devices 700-1 to 700-n and machine tools 500-1 to 500-n and machine learning devices 300-1 to 300-n may be provided in the same factory, or may be provided in different factories.

[0187] The network 800 is, for example, a LAN (Local Area Network) constructed in a factory, the Internet, a public telephone network, or a combination thereof. The specific communication method in the network 800 is not particularly limited to wired connection or wireless connection.

[0188] <Degree of freedom of system structure>

[0189] In the above-mentioned embodiment, the control devices 700-1 to 700-n and the machine tools 500-1 to 500-n and the control devices 700-1 to 700-n and the machine tools 500-1 to 500-n are respectively one-to-one groups and are connected in a communicative manner, but for example, a machine learning device can also be connected to multiple control devices and machine tools via the network 800 in a communicative manner to implement machine learning for multiple control devices and machine tools.

[0190] In this case, each function of a machine learning device can be implemented as a distributed processing system that is appropriately distributed across multiple servers. In addition, each function of a machine learning device can also be implemented on the cloud using a virtual server function or the like.

[0191] Furthermore, when there are n machine learning devices 300-1 to 400-n corresponding to n control devices 700-1 to 700-n and machine tools 500-1 to 500-n of the same model name, same specification, or same series, the learning results of each machine learning device 300-1 to 300-n can be shared. In this way, a more ideal model can be constructed.

[0192] The machine learning device, control system, and machine learning method according to the present disclosure include the above-mentioned embodiments and can take various embodiments having the following structures.

[0193] (1) A first aspect of the present disclosure is a machine learning device (e.g., machine learning unit 300) comprising: a virtual temperature model creating unit (e.g., virtual temperature model creating unit 320) having a virtual temperature calculation formula, wherein the temperature of a specific part of a machine is estimated by using heat generation cause data and virtual temperature data is obtained by using the virtual temperature calculation formula, wherein the virtual temperature calculation formula includes a first coefficient that determines a heat generation amount and a second coefficient that determines a heat release amount; and

[0194] a thermal displacement model making unit (e.g., thermal displacement model making unit 310) having a thermal displacement estimation calculation formula including a third coefficient, using the virtual temperature data obtained by the virtual temperature calculation formula and actual temperature data obtained from at least one temperature sensor installed at a location other than the specific location, to find an error between the thermal displacement estimated by the thermal displacement estimation calculation formula and the actually measured thermal displacement, and performing machine learning to search for the third coefficient to make an optimized thermal displacement estimation calculation formula so as to minimize the error,

[0195] The virtual temperature model preparation unit performs machine learning to search for the first coefficient and the second coefficient to prepare an optimized virtual temperature calculation formula so that an error obtained by the optimized thermal displacement estimation amount calculation formula using the virtual temperature data estimated by the virtual temperature calculation formula is minimized.

[0196] According to the machine learning device of the first aspect of the present disclosure, even when there is a part where a temperature sensor cannot be installed to actually measure the temperature, the temperature of the part can be estimated, and the estimated thermal displacement amount of the component of the device can be obtained using the estimated temperature.

[0197] (2) Regarding the machine learning device described in (1) above, the first coefficient is a coefficient that determines the amount of heat caused by heat transfer from the electric motor driving the machine and / or bearing friction, and the second coefficient is a coefficient that determines the amount of heat release caused by heat transfer from a specific part of the machine to the surrounding fluid.

[0198] (3) Regarding the machine learning device described in (1) or (2) above, the specific part of the machine is a spindle of a machine tool (for example, spindle 520A).

[0199] (4) Regarding the machine learning device described in (1) or (2) above, the specific part of the machine is the screw shaft of a ball screw (e.g., ball screw 5212) of a machine tool.

[0200] (5) A second aspect of the present disclosure is a machine learning device (e.g., machine learning unit 300A) including a virtual temperature model making unit (e.g., virtual temperature model making unit 320A) having a virtual temperature calculation formula including a first coefficient for determining a heat amount and a second coefficient for determining a heat amount.

[0201] The virtual temperature model making unit uses virtual temperature data of a specific part of the machine obtained through the virtual temperature calculation formula using the heat cause data and actual temperature data obtained from at least one temperature sensor installed at the specific part, calculates the error between the virtual temperature data and the actual temperature data, and performs machine learning to search for the first coefficient and the second coefficient to make an optimized virtual temperature calculation formula so as to minimize the error.

[0202] According to the machine learning device of the second aspect of the present disclosure, even if a temperature sensor can be installed at the time of data collection, when there is a location where the temperature sensor cannot be installed at the time of actual operation, the temperature of the location can be estimated.

[0203] (6) The machine learning device described in (5) above includes a thermal displacement model creation unit having a thermal displacement estimation amount calculation formula including a third coefficient,

[0204] The thermal displacement model creation unit creates an optimized thermal displacement estimation calculation formula by performing machine learning using virtual temperature data obtained using the optimized virtual temperature calculation formula and actual temperature data obtained from at least one temperature sensor installed at a location other than the specific location.

[0205] (7) Regarding the machine learning device described in (5) or (6) above, the specific part of the machine is a turret of a machine tool.

[0206] (8) A third aspect of the present disclosure is a control system (e.g., control system 10, 10A) comprising: a machine learning device (e.g., machine learning unit 300, 300A) as described in any one of (1), (2), (3), (4), (6), and (7);

[0207] a thermal displacement correction device (e.g., thermal displacement correction unit 200) that stores the virtual temperature calculation formula and the thermal displacement estimation calculation formula output from the machine learning device, obtains an estimated value of thermal displacement using the thermal displacement estimation calculation formula using virtual temperature data obtained from the virtual temperature calculation formula and actual temperature data obtained from at least one temperature sensor installed at a location other than the specific location, and obtains a thermal displacement correction amount based on the estimated value of thermal displacement; and

[0208] The numerical control device corrects a control command output to a motor control unit that controls the motor, based on the thermal displacement correction amount.

[0209] According to the control system of the third aspect of the present disclosure, even when a temperature sensor cannot be installed or it is difficult to install a temperature sensor in actual operation, the temperature can be estimated.

[0210] (9) A fourth aspect of the present disclosure is a machine learning method of a machine learning device (e.g., machine learning unit 300), which obtains virtual temperature data by estimating the temperature of a specific part of a machine using heat generation cause data using a virtual temperature calculation formula including a first coefficient that determines a heat generation amount and a second coefficient that determines a heat release amount.

[0211] having a thermal displacement estimation calculation formula including a third coefficient, using the virtual temperature data obtained by the virtual temperature calculation formula and actual temperature data obtained from at least one temperature sensor installed at a location other than the specific location, obtaining an error between the thermal displacement estimated by the thermal displacement estimation calculation formula and the actually measured thermal displacement, performing machine learning to search for the third coefficient to create an optimized thermal displacement estimation calculation formula so as to minimize the error,

[0212] Machine learning for searching the first coefficient and the second coefficient is performed to create an optimized pseudo temperature calculation formula so that an error obtained by the optimized thermal displacement estimation amount calculation formula using the pseudo temperature data estimated by the pseudo temperature calculation formula is minimized.

[0213] According to the machine learning method of the fourth aspect of the present disclosure, even if there is a part where a temperature sensor cannot be installed to actually measure the temperature, the temperature of the part can be estimated. Furthermore, the estimated temperature can be used to obtain the estimated amount of thermal displacement of the component of the device.

[0214] (10) The fifth method of the present disclosure is a machine learning method of a machine learning device (for example, the machine learning unit 300A), which uses virtual temperature data calculated using heat cause data according to a virtual temperature calculation formula and actual temperature data obtained from at least one temperature sensor installed at the specific location, calculates the error between the virtual temperature data and the actual temperature data, and performs machine learning to search for the first coefficient and the second coefficient to create an optimized virtual temperature calculation formula to minimize the error, wherein the virtual temperature calculation formula includes a first coefficient that determines the heat generation and a second coefficient that determines the heat release, and estimates the temperature of a specific location of the machine.

[0215] According to the machine learning method of the fifth aspect of the present disclosure, even if there is a location where a temperature sensor can be installed during data collection but cannot be installed during actual operation, the temperature of the location can be estimated. Furthermore, the estimated temperature can be used to obtain an estimated amount of thermal displacement of a component of the machine.

[0216] Explanation of symbols

[0217] 10, 10A, 10B control system

[0218] 100 Numerical Control Unit

[0219] 200 Thermal displacement correction unit

[0220] 201 Virtual temperature calculation unit

[0221] 202 Temperature data storage unit

[0222] 203 Thermal displacement correction calculation unit

[0223] 300 Machine Learning Department

[0224] 310 Thermal Displacement Model Production Department

[0225] 311 Shift Acquisition Unit

[0226] 312 Temperature acquisition unit

[0227] 313 Virtual spindle temperature acquisition unit

[0228] 314 Storage

[0229] 315 Learning Department

[0230] 316 Error calculation unit

[0231] 320 Virtual Temperature Model Production Department

[0232] 321 Fever cause data acquisition unit

[0233] 322 Error acquisition unit

[0234] 323 Storage

[0235] 324 Learning Department

[0236] 325 Virtual spindle temperature calculation unit

[0237] 400 Motor Control Unit

[0238] 500 Machine Tools

[0239] 510 Electric Motor

[0240] 520 driven body

[0241] 600 Workpiece.

Claims

1. A machine learning device, characterized in that: have: a virtual temperature model making unit having a virtual temperature calculation formula, and using the heat generation cause data to estimate the temperature of a specific part of the machine by the virtual temperature calculation formula and obtain virtual temperature data, wherein the virtual temperature calculation formula includes a first coefficient that determines a heat generation amount and a second coefficient that determines a heat release amount; and a thermal displacement model preparation unit having a thermal displacement estimation calculation formula including a third coefficient, using the virtual temperature data obtained by the virtual temperature calculation formula and actual temperature data obtained from at least one temperature sensor installed at a location other than the specific location, to obtain an error between the thermal displacement estimated by the thermal displacement estimation calculation formula and the actually measured thermal displacement, and performing machine learning to search for the third coefficient to prepare an optimized thermal displacement estimation calculation formula so as to minimize the error, The virtual temperature model preparation unit performs machine learning to search for the first coefficient and the second coefficient to prepare an optimized virtual temperature calculation formula so that an error obtained by the optimized thermal displacement estimation amount calculation formula using the virtual temperature data estimated by the virtual temperature calculation formula is minimized.

2. The machine learning device according to claim 1, characterized in that The first coefficient is a coefficient that determines the amount of heat generated by heat transfer from the motor driving the machine and / or bearing friction, and the second coefficient is a coefficient that determines the amount of heat released by heat transfer from a specific portion of the machine to a surrounding fluid.

3. The machine learning device according to claim 1 or 2, characterized in that: The specific part of the machine is the spindle of the machine tool.

4. The machine learning device according to claim 1 or 2, characterized in that: The specific part of the machine is the screw shaft of a ball screw of a machine tool.

5. A control system, characterized in that: have: The machine learning device according to claim 1 or 2; a thermal displacement correction device that stores a virtual temperature calculation formula and a thermal displacement estimation calculation formula output from the machine learning device, obtains an estimated value of thermal displacement using the thermal displacement estimation calculation formula using virtual temperature data obtained from the virtual temperature calculation formula and actual temperature data obtained from at least one temperature sensor installed at a location other than the specific location, and obtains a thermal displacement correction amount based on the estimated value of thermal displacement; and The numerical control device corrects a control command output to a motor control unit that controls the motor, based on the thermal displacement correction amount.

6. A machine learning method for a machine learning device, characterized in that: The pseudo temperature data is obtained by estimating the temperature of a specific part of the machine using the heat generation cause data using a pseudo temperature calculation formula that includes the first coefficient that determines the heat generation and the second coefficient that determines the heat generation. having a thermal displacement estimation calculation formula including a third coefficient, using the virtual temperature data obtained by the virtual temperature calculation formula and actual temperature data obtained from at least one temperature sensor installed at a location other than the specific location, obtaining an error between the thermal displacement estimated by the thermal displacement estimation calculation formula and the actually measured thermal displacement, performing machine learning to search for the third coefficient to create an optimized thermal displacement estimation calculation formula so as to minimize the error, Machine learning for searching the first coefficient and the second coefficient is performed to create an optimized pseudo temperature calculation formula so that an error obtained by the optimized thermal displacement estimation amount calculation formula using the pseudo temperature data estimated by the pseudo temperature calculation formula is minimized.

Images