Learning device for a model of thermal displacement, estimating device for thermal displacement, processing system and processing method

Abstract

A learning device for a thermal displacement model (2) comprises the following: a training data acquisition unit (21) that acquires training data in which input data and thermal displacement data are temporally linked, wherein the input data includes at least temperature data representing a temperature of a machine tool in time series, and wherein the thermal displacement data represent a magnitude of the thermal displacement of the machine tool in time series; a data set generation unit (22) that generates data sets obtained by extracting a portion of the training data from the training data; and a training unit (23) that, using the data set, generates a thermal displacement model trained to estimate the magnitude of the thermal displacement from the input data relating to the machine tool.The data set generation unit (22) generates the data set such that at least two of the data sets are extracted from a plurality of data sets in time intervals that have different durations from each other.

Description

Area

[0001] The present disclosure relates to a learning device for a thermal displacement model, a thermal displacement estimating device, a machining system and a machining method for correcting a thermal displacement of a machine tool. background

[0002] A machine tool is a machining device that drives a feed rod to change the relative position between a tool and a machining target, called a workpiece, thereby machining the target to a desired shape through chip removal. A typical machine tool, such as a milling machine or a lathe, performs a machining operation by rotating a spindle to which the tool or target is attached. The machine tool includes an actuator to drive the feed rod and the spindle, a servo amplifier to generate the electrical power that drives the actuator, and so on. The machine tool, which includes the actuator, measures the position of the feed rod with a position sensor and performs positioning under a controller that ensures the actual position matches a commanded position.

[0003] When a machine tool processes a workpiece, heat is generated within the machine tool. For example, the actuator and servo amplifier generate heat when they convert electrical power into drive power. Furthermore, during machining processes such as cutting and polishing, heat is generated through material deformation, friction, and other factors. Additionally, friction generated when the feed rod is operated also produces heat. When this heat is transferred to a structure within the machine tool, such as a column or bed, the mechanical structure is subjected to a temperature increase and thermal expansion. Deformation of the structure due to this thermal expansion is called thermal displacement.The structure may not be subject to uniform temperature changes, in which case different positions within the structure are subject to different thermal displacements. This can result in displacements other than simple expansion and contraction, such as buckling, stretching, and twisting of the structure. It is known that such thermal displacements can reduce the positioning accuracy of the machine tool, thus causing an error in the relative position between the tool and the workpiece and consequently a decrease in machining accuracy.

[0004] Patent document 1 discloses a technique that enables the estimation of a thermal displacement magnitude from measurement data by machine learning, using an actual measurement of the thermal displacement magnitude and measurement data obtained during machining by a machine tool. Specifically, a machine learning device disclosed in patent document 1 includes a measurement data acquisition unit, a thermal displacement magnitude acquisition unit, a storage unit, and a training unit for the calculation formula. The measurement data acquisition unit acquires a set of measurement data. The thermal displacement magnitude acquisition unit acquires the actual measurement of the thermal displacement magnitude. The storage unit stores training data.The training data includes the measurement data set as input data, which is associated with the actual measurement of the thermal displacement magnitude as a label. The learning unit for the calculation formula defines a formula to predict the magnitude of the thermal displacement and calculate it based on the measurement data set by performing machine learning using the training data. The storage unit stores the measurement data set as training data for a predetermined period. List of quotations Patent literature

[0005] Patent literature 1: Japanese publication no. 2019-111648 Brief description of the technical problem

[0006] Unfortunately, the conventional technique described above suffers from the problem of potentially reducing the accuracy of estimating the magnitude of thermal displacement occurring in the machine tool if the memory unit stores the training data for a period unsuitable for the thermal displacement occurring in the machine tool. For example, the assumed time it takes for heat from a particular heat source to cause thermal displacement is influenced by positional relationships with other heat sources, the surrounding environment, and so on. If the storage period is too short, the measurement data influencing thermal displacement may therefore not be included in the stored training data at the time of learning, which can reduce the accuracy of estimating the magnitude of thermal displacement.Furthermore, if the storage period is too long, the measurement data included in the training data, but which contribute less to the thermal displacement at the time of learning, will be incorrectly learned, which can reduce the accuracy of estimating the magnitude of the thermal displacement.

[0007] The present disclosure was prepared with regard to the foregoing and one of its objectives is to provide a learning device for a thermal displacement model that is capable of improving the accuracy of estimating a quantity of thermal displacement. Solution to the problem

[0008] To solve the above problem and fulfill the task, a learning device for a thermal displacement model of the present disclosure comprises the following: a learning data acquisition unit for obtaining learning data in which input data and thermal displacement data are temporally linked, wherein the input data includes at least temperature data representing a temperature of a machine tool in time series, and wherein the thermal displacement data represent a quantity of the thermal displacement of the machine tool in time series; a data set generation unit for generating data sets obtained by extracting a portion of the learning data from the learning data;and a training unit for generating, using the dataset, a thermal displacement model that is trained to estimate the magnitude of the thermal displacement from the input data of the machine tool, wherein the dataset generation unit generates the datasets such that at least two of the datasets are extracted from a plurality of datasets in time intervals having different durations from each other. Advantageous effects of the invention

[0009] The present disclosure has the effect of improving the accuracy of estimating the magnitude of thermal displacement. Brief description of the drawings Fig. Figure 1 is a representation illustrating a configuration of a machining system according to a first embodiment. Fig. Figure 2 is a representation illustrating a functional design of a machine tool according to the first embodiment. Fig. Figure 3 is a representation illustrating examples of an installation position of a temperature sensor in the machine tool according to the first embodiment. Fig. Figure 4 is an explanatory illustration of an example of changes in the operating state of the machine tool. Fig. Figure 5 is a representation illustrating a functional design of a learning device for a model of thermal displacement according to the first embodiment. Fig. Figure 6 is an explanatory presentation of a first example of a method for generating data records. Fig. Figure 7 is an explanatory presentation of a second example of the data set generation procedure. Fig. Figure 8 is an explanatory representation of a third example of the data set generation procedure. Fig. Figure 9 is an explanatory presentation of a procedure for learning a model of thermal displacement. Fig. Figure 10 is an explanatory representation of a neural network. Fig. 11 is a flowchart to explain a business operation that is in Fig. 5 illustrated learning device for a model of thermal displacement. Fig. Figure 12 is a representation illustrating a configuration of a machining system according to a second embodiment. Fig. Figure 13 is a representation illustrating a functional design of a machine tool according to the second embodiment. Fig. Figure 14 is a representation illustrating a functional design of an estimating device for thermal displacement according to the second embodiment. Fig. 15 is a flowchart to explain an operation of the in Fig. 12 illustrated processing systems. Fig. Figure 16 is a representation illustrating an example of a hardware design. Fig. Figure 17 is a representation illustrating a functional design of a learning device for a model of thermal displacement according to a variation of the second embodiment. Fig. Figure 18 is a representation illustrating a configuration of a machining system according to a third embodiment. Fig. Figure 19 is a representation illustrating a configuration of a machining system according to a fourth embodiment. Description of embodiments

[0010] A learning device for a thermal displacement model, a thermal displacement estimating device, a machining system and a machining method according to embodiments of the present disclosure are described in more detail below with reference to the drawings. First embodiment.

[0011] Fig. Figure 1 illustrates an embodiment of a machining system 100-1 according to a first embodiment. The machining system 100-1 comprises a machine tool 1 and a learning device for a thermal displacement model 2. The machine tool 1 is a machining device that performs machining operations on a machining target. The learning device for a thermal displacement model 2 learns a thermal displacement model to estimate a thermal displacement magnitude based on input data derived from thermal displacement data, temperature data, and machine state data obtained from the machine tool 1.

[0012] Fig. Figure 2 is a representation illustrating a functional configuration of the machine tool 1 according to the first embodiment. The machine tool 1 includes a feed rod 11, a feed rod motor 12, a spindle 13, a spindle motor 14, a controller 15, a peripheral device 16, a temperature sensor TS, and a displacement sensor DS. The feed rod 11, on which a tool T for machining a workpiece W is mounted, changes the relative position between the tool T and the workpiece W. The feed rod motor 12 is an actuator for driving the feed rod 11. The spindle 13 rotates the tool T or the workpiece W. The spindle motor 14 is an actuator for driving the spindle 13. The controller 15 controls the machine tool 1. The peripheral device 16 is a cooling device for reducing heat generated when the actuator is driven.The temperature sensor TS is provided for measuring the temperature of machine tool 1. The displacement sensor DS measures a relative position between the tool T and the workpiece W.

[0013] The displacement sensor DS is integrated into the machine tool 1 in such a way that it can detect the relative position between the tool T and the workpiece W. The displacement sensor DS measures this relative position and outputs the measured position as thermal displacement data to the learning device for a thermal displacement model 2. The displacement sensor DS, which can be any sensor capable of detecting displacement in at least one axial direction, could be, for example, a triangulation laser sensor or a time-of-flight (TOF) laser sensor. It should be noted that multiple displacement sensors, orthogonal to each other, can be integrated to detect displacement in multiple axial directions.

[0014] The temperature sensor TS is installed in the machine tool 1. The temperature sensor TS detects a temperature and outputs the detected temperature as temperature data to the learning device for a thermal displacement model 2. Although in Fig. As illustrated in Figure 2, where a single temperature sensor TS is shown, the machining system 100-1 ideally incorporates a multitude of temperature sensors TS. The temperature sensor TS is ideally installed at a location expected to experience significant temperature fluctuations and at a location expected to have a significant impact on thermal displacement. Additionally, the temperature sensor TS is ideally also installed outside the machine tool 1 so that it can measure the temperature of the room in which the machine tool 1 is located.

[0015] Fig. Figure 3 is a diagram illustrating examples of the installation position of the temperature sensor TS in the machine tool 1 according to the first embodiment. The diagram shows... Fig. Figure 3 illustrates machine tool 1 as a milling machine comprising a bed 17 to which the feed rod motor 12 for a Y-axis and a table TA are mounted. The workpiece W is placed on the table TA. Furthermore, machine tool 1 also includes a column 18 to which the feed rod motors 12 for the X-axis and Z-axis, the spindle 13, and the spindle motor 14 are mounted. Fig. Figure 3, with its black circles, illustrates an example of the installation position of the temperature sensor TS. As described above, the temperature sensor TS is preferably installed at the location expected to experience significant temperature changes and at the location expected to have a significant influence on thermal displacement. The location expected to experience significant temperature changes includes, for example, drive mechanisms such as the feed rod motors 12 for driving the X, Y, and Z axes and the spindle motor 14 for rotating the spindle 13. The location expected to have a significant influence on thermal displacement includes, for example, structures of the machine tool 1, such as the bed 17, the column 18, and a feed spindle 19.The temperature sensor TS is preferably installed at a vertex of the structure, at a point between a plurality of vertices of the structure, on an inside of the structure, etc.

[0016] It will be revisited Fig. 2. The peripheral device 16, which is the cooling device, is installed to cool the machine tool 1 and prevent thermal displacement. The peripheral device 16 facilitates heat exchange between the structure of the machine tool 1, the workpiece W, the tool T, and the outside air located outside the machine tool 1. For example, the peripheral device 16 includes a coolant device that injects cutting oil to cool the workpiece W, a pump that circulates a refrigerant to cool a heat exchanger tube installed in the machine tool 1, etc. The peripheral device 16 outputs operating information to the controller 15.

[0017] The operating information for peripheral device 16 includes at least one operating status indicating whether peripheral device 16 is operating or not. In addition to the operating status, the operating information for peripheral device 16 may include power consumption, power output of peripheral device 16, etc. The operating status included in the operating information for peripheral device 16 may be represented by a numerical value. For example, operating status "1" represents "a state in which peripheral device 16 is operating" and operating status "0" represents "a state in which peripheral device 16 is not operating".

[0018] A machining program is a series of instructions given to the control 15 so that the feed rod 11, the spindle 13, and the peripheral device 16 of the machine tool 1 can perform the desired operations. Specifically, the machining program can include information representing a position command for the feed rod 11, a speed command for the feed rod 11, a speed command for the spindle 13, a start and stop command for the peripheral device 16, etc.

[0019] The control unit 15 reads the machining program and analyzes the command described in the machining program. Furthermore, the control unit 15 generates a command position and a command speed for the feed rod 11 to enable the feed rod 11 to perform the desired operation, and sends a control input to the feed rod motor 12. The control unit 15 also generates a command speed for the spindle 13 to enable the spindle 13 to perform the desired operation, and sends a control input to the spindle motor 14.

[0020] Furthermore, the controller 15 outputs the machine state data, which represent an operating state of the machine tool 1, to the learning device for a thermal displacement model 2. The machine state data, which represents the operating state of the machine tool 1, specifies, for example, the operating state of the peripheral device 16, which adjusts the temperature of the machine tool 1, and an operating state of the actuator of the machine tool 1. Specifically, the machine state data includes at least one or more types of information from the command position for the feed rod 11, the command speed for the feed rod 11, the command speed for the spindle 13, and the operating information about the peripheral device 16.

[0021] In a case where the position sensor is integrated into the machine tool 1, the controller 15 can perform feedback control such that any difference between the actual position detected by the position sensor and the command position, or between the actual speed and the command speed, becomes zero. In this case, the machine status data can include one or more types of information from the actual position of the feed rod 11, the actual speed of the feed rod 11, and the actual speed of the spindle 13.

[0022] The following is a description of the heat generated inside and outside the machine tool 1 and the heat transfer. A portion of the electrical energy supplied when the controller 15 drives the feed rod motor 12 and the spindle motor 14 is lost as heat. Furthermore, heat loss occurs due to mechanical friction during the operation of the feed rod 11 and the spindle 13. These heat losses increase the temperature of the machine tool 1's structure, but at the same time, the machine tool 1 is cooled by the peripheral device 16. Additionally, machining heat is generated when the workpiece W is machined with the tool T. Finally, if there is a temperature difference between the main body of the machine tool 1 and the surrounding environment, the temperature of the machine tool 1's structure changes over time according to this temperature difference.The heat generation, heat dissipation, and forced cooling that occur inside and outside the machine tool 1, as described above, cause a change in the temperature of the structure and result in thermal expansion and thermal stress within the structure. Such a thermal change of the structure leads to a machining error between the tool T and the workpiece W. In the present disclosure, the machining error between the tool T and the workpiece W caused by the thermal change of the machine tool 1 is referred to as thermal displacement.

[0023] Next, the operating state of the machine tool will be described. Fig. Figure 4 is an explanatory representation of an example of changes in the operating state of machine tool 1. Fig. Figure 4 illustrates changes in the operating state over a period of time in which the machine tool 1 is first switched on, then operated according to the machining program, and finally switched off.

[0024] Fig. Figure 4 illustrates an example of a time-varying transition between the execution state of the machining program and the operating state of each of the components of the machine tool 1, i.e., the spindle 13, the feed rod 11, and the peripheral device 16. The horizontal axis in Fig. 4 represents time, and "t0" to "t9" represent time. Details regarding time t0 to time t9 are described below. The execution status of the machining program indicates whether machine tool 1 is executing a program operation or has been stopped. The term "program operation" means that machine tool 1 is operating according to a command described in the machining program. Machine tool 1 follows this procedure: Machine tool 1 executes the program operation for a specific machining operation. When the program operation is complete, an operator performs an operation, such as exchanging the workpiece W. Afterward, another operation is performed, such as the machine tool executing the program operation for the next machining operation.

[0025] In the Fig. In the illustrated example 4, machine tool 1 is switched on at time t0 and switched off at time t9. The period from time t0 to time t1 represents a standby state before operation is started according to the machining program. The period from time t8 to time t9 represents the period from the completion of operation according to the machining program until the time at which machine tool 1 is switched off.

[0026] Furthermore, as indicated by the execution status of the machining program, program operation according to machining program no. 1 is carried out from time t1 to time t5, and program operation according to machining program no. 2 is carried out from time t6 to time t8. When these machining programs are executed, the spindle speed (the rotational speed of the spindle 13), the feed rate (the speed of the feed rod 11), and the operating state of the peripheral device 16 change over time.

[0027] The “spindle speed” in Fig. 4 indicates the speed of spindle 13. In Fig. 4 indicates that "S1000" means that the spindle 13 rotates at 1000 revolutions per minute. As determined by the spindle speed in Fig. As specified in Figure 4, the spindle motor 14 rotates at a speed of 1000 revolutions per minute from time t1 to time t4, at a speed of 3000 revolutions per minute from time t4 to time t5, and at a speed of 3000 revolutions per minute from time t6 to time t8. When the spindle 13 rotates, heat is generated by the structure of the spindle motor 14 and the spindle 13. The higher the spindle speed, the greater the amount of heat generated by the spindle 13. Since the spindle speed changes according to the instructions of the machining program, the amount of heat generated by the spindle 13 also changes according to the machining program.

[0028] The "feed rate" in Fig. 4 indicates the speed of movement of the feed rod 11. In Fig. 4 indicates that “F100” means that the feed rod 11 is operated at 100 mm per minute. As determined by the feed rate in Fig. As specified in Figure 4, the feed rod motor 12 rotates at a speed that causes the feed rod 11 to operate at 100 mm per minute from time t1 to time t3, at a speed that causes the feed rod 11 to operate at 300 mm per minute from time t3 to time t5, and at a speed that causes the feed rod 11 to operate at 300 mm per minute from time t6 to time t8. When the feed rod 11 is operating, heat is generated by the structure of the feed rod motor 12 and the feed rod 11. The higher the feed rate, the greater the amount of heat generated. Since the feed rate changes according to the instructions of the machining program, the amount of heat generated by the feed rod 11 also changes according to the machining program.

[0029] The “operating status of the peripheral device” in Fig. 4 indicates whether the peripheral device 16 for cooling the machine tool 1 is in operation or not. The peripheral device 16 is not always in operation during machining, but is operated or stopped according to the instructions described in the machining program. In the example for the operating status of the peripheral device in Fig. 4. The peripheral device 16 is in operation from time t2 to time t5, the peripheral device 16 is not in operation from time t5 to time t7, and the peripheral device 16 is in operation from time t7 to time t8. When the peripheral device 16 is operating, heat is transferred from the machine tool 1 to the environment outside the machine tool 1, thus cooling the machine tool 1. Since the operating status of the peripheral device 16 changes according to the instructions of the machining program, the cooling state provided by the peripheral device 16 also changes according to the machining program.

[0030] Since, as with reference to Fig. As described in section 4, the heat generated in the feed rod 11 and the spindle 13 and the cooling condition provided by the peripheral device 16 are not constant, so the magnitude of the thermal displacement of the machine tool 1 also changes over time according to the machining program.

[0031] There is also a period, such as between time t5 and time t6, during which operation according to the machining program is not carried out. Between times t5 and t6, the machine tool 1 may be stopped, for example, due to setup work, such as an operator removing the workpiece W or cleaning the interior of the machine tool 1. When the machine tool 1 is stopped, as described above, the operating states of the spindle 13, the feed rod 11, and the peripheral device 16 differ from those during program operation, so the magnitude of the generated thermal displacement tends to differ from that during program operation.

[0032] As described above, the speeds of the feed rod 11 and the spindle 13, as well as the operating state of the peripheral device 16, affect the occurrence of thermal displacement. In the present embodiment, individual intervals with different operating states of the machine tool 1, such as an interval from time t0 to time t1 and an interval from time t1 to time t2, are defined as an "operating unit". The duration of the operating unit changes depending on the time specified in the instructions described in the machining program and is therefore variable.

[0033] It should be noted that, although Fig. 4 illustrates the example in which the machine tool 1 executes two different types of machining programs, the machining program only needs to be executed at least once, and the type and number of executions of the machining program are not limited to this example.

[0034] Furthermore, after being switched off, machine tool 1 can be switched on again and execute the machining program. In this case, a time interval from the time at which machine tool 1 is to be switched off until the time at which machine tool 1 is to be switched on again is defined as an operating unit.

[0035] Next, a functional design of the learning device for a model of thermal displacement 2 is described. Fig. Figure 5 illustrates the functional design of the learning device for a thermal displacement model 2 according to the first embodiment. Using the data output by the machine tool 1 during operation, the learning device for a thermal displacement model 2 learns the thermal displacement model to estimate the magnitude of the thermal displacement based on the input data, which includes at least the temperature data.

[0036] The learning device for a thermal displacement model 2 includes a learning data acquisition unit 21, a data set generation unit 22, a learning unit 23 and a model storage unit 24.

[0037] The learning data acquisition unit 21 acquires the thermal displacement data, temperature data, and machine state data, all output by machine tool 1, and generates learning data in which the acquired data is time-synchronized. Temperature data and machine state data are examples of information included in the input data. The input data includes at least the temperature data. The learning data is a time series in which the thermal displacement data, temperature data, and machine state data are arranged within a "fixed period." The term "fixed period" represents a sampling interval that is preset in the learning data acquisition unit 21. The learning data acquisition unit 21 receives the data for the operating unit, which consists of at least one or more time intervals, from machine tool 1 and generates the learning data.The learning data acquisition unit 21 outputs the generated learning data to the data set generation unit 22.

[0038] The data set generation unit 22 captures a specific point in time based on the operating state of the machine tool 1, extracts the thermal displacement data, the temperature data, and the machine state data from the training data with respect to the captured point in time, assembles the extracted data into a data set, and outputs a multitude of the data sets to the training unit 23. Hereinafter, the term "specific point in time" captured by the data set generation unit 22 is referred to as an "extraction reference point." The extraction reference point is defined with reference to Fig. 4 described. The data record generation unit 22 records each of the time points t1 to t8. The recorded time point is defined as the time at which the operating state of the machine tool 1 changes. It should be noted that in Fig. 4. The extraction reference point is defined as the point in time at which a change occurs in any of the spindle speed, feed rate, operating status of the peripheral device 16, and the start / end of the machining program. However, the extraction reference point is not limited to these examples. For instance, the time specified for calling and finishing a subprogram, the time specified for changing the tool T, and the like can also be considered as changes in the operating state of the machine tool 1. Furthermore, if a time specification for a change in the operating state is specified in advance in the machining program, the data record generation unit 22 can read this information.It should be noted that the data set generation unit 22 can additionally define a new extraction reference time by temporally interpolating two extraction reference times.

[0039] The data set generation unit 22 generates the data set by extracting a portion of the training data from it with respect to the extraction reference time.

[0040] Fig. Figure 6 is an explanatory presentation of a first example of a data set generation procedure. The first example of the procedure for data set generation unit 22 for generating the data sets from the training data is presented using Fig. 6 described. Fig. Figure 6 illustrates the data sets a1 to an extracted from the training data in a case where an extraction reference time t1 is the starting reference time for data extraction. Input data elements a1 to an are the temperature data and the machine state data, with extraction reference time t1 as the starting time. If an element of i-th input data is an input data element ai, the time interval of input data element ai is extended by a predetermined time extension width T1, following the end of the time interval of an input data element ai-1. The duration of input data element a2 is longer than the duration of input data element a1 by the time extension width T1. Similarly, the duration of input data element a3 is longer than the duration of input data element a2 by the time extension width T1. That is, if the time extension width of input data element a3 is a1, the duration of input data element a3 is a1.a1 T is the time width of input data element no. ai “T+(1i-1)T1”. Input data elements no. a1 to no. an are data elements in time intervals starting at the extraction reference time t1, and each input data element has the corresponding time width “T+(i-1)T1”. Each training data element no. a1 to no. an represents the thermal displacement data at the end time of the corresponding input data element no. a1 to no. an. As in . Fig. As illustrated in Figure 6, the data set generation unit 22 generates the plurality of data sets No. a1 to No. an based on the single extraction reference time t1. Through the processing described above, the data set generation unit 22 can generate the data sets such that the extracted time intervals of the plurality of data sets No. a1 to No. an, generated based on the single extraction reference time t1, have different durations. The data set generation unit 22 performs the processing described above for each of a plurality of extraction reference times included in the training data. It should be noted that, as with data set No. an, the end time can be after "t2", which is the extraction reference time after the extraction reference time t1.

[0041] Fig. Figure 7 is an explanatory presentation of a second example of the data set generation procedure. The second example of the procedure for data set generation unit 22 for generating the data sets from the training data is presented using... Fig. 7 described. Fig. Figure 7 illustrates the data sets b1 to bn extracted from the training data in a case where the extraction reference time t2 is an end reference time for the data extraction. Input data elements b1 to bn are the temperature data and the machine state data, with extraction reference time t2 as the end time. If an element of i-th input data is an input data element bi, the time interval of input data element bi is extended by the predetermined time extension width T1, counting backward from a starting point of the time interval of an input data element bi-1. If the time width of input data element b1 is T, then, as in the first example, the time width of input data element bi is "T+(1i-1)T1". The input data elements b1 to bn are...bn are data elements in time intervals ending at the extraction reference time t2, and each input data element has the corresponding time width "T+(i-1)T1". Each training data element, b1 to bn, represents the thermal displacement data at the end time of the corresponding input data element, b1 to bn. As in . Fig. As illustrated in Figure 7, the data set generation unit 22 generates the plurality of data sets No. b1 to No. bn based on the single extraction reference time t2. Through the processing described above, the data set generation unit 22 can generate the data sets such that the extracted time intervals of the plurality of data sets No. b1 to No. bn, generated based on the single extraction reference time t2, have different durations. The data set generation unit 22 performs the processing described above for each of a plurality of extraction reference times included in the training data. It should be noted that, as with data set No. bn, the starting time can go back to a past time "t1", which is the extraction reference time before the extraction reference time t2.

[0042] Fig. Figure 8 is an explanatory illustration of a third example of the data set generation procedure. In this third example, the data set generation unit 22 generates the data sets in such a way that an aggregate of the data contained in all data sets uniformly encompasses all the obtained data. Specifically, the data set generation unit 22 generates the data sets by dividing a period from the extraction reference time t1 to the extraction reference time t2, which follows extraction reference time t1, into n equal intervals, where n is a natural number. Fig. Figure 8 illustrates data sets c1 to cn, which are extracted from the training data between extraction reference time t1 (the initial reference time for data extraction) and extraction reference time t2 (the end time). Input data element c1 consists of temperature and machine state data, with extraction reference time t1 as the start time and t1 + (t2 - t1) / n as the end time. Input data element c2 consists of temperature and machine state data, with t1 + (t2 - t1) / n as the start time and t1 + 2 × (t2 - t1) / n as the end time. Input data element no. ci consists of the temperature data and the machine state data, with the start time being "t1+(i-1)×(t2-t1) / n" and the end time being "t1+i×(t2-t1) / n". Each training data element no. c1 to no.cn represents the thermal displacement data at the end time of the corresponding input data elements c1 to cn. Through the above processing, the data set generation unit 22 can generate the data sets such that the data elements are evenly distributed in the input data between the extraction reference time t1 and time t2, without overlap. The data set generation unit 22 performs the above processing for each of a plurality of extraction reference times included in the training data. It should be noted that in the third example, the extracted time intervals for data sets c1 to cn, generated between extraction reference times t1 and t2, have uniform durations. However, as in . Fig. As illustrated in Figure 4, the durations between adjacent extraction reference times are not uniform, and thus the duration of the time interval extracted when the data record generation unit 22 generates the data record is determined depending on the duration between the two adjacent extraction reference times. Therefore, by performing the above processing on each of all extraction reference times as a whole, the multitude of data records in extracted time intervals, which have different durations, is generated.

[0043] In the first through third examples of the data set generation procedure described above, each training data element, No. a1 to No. an, No. b1 to No. bn, and No. c1 to No. cn, represents the thermal displacement data for a time interval corresponding to the end time of the respective input data element, No. a1 to No. an, No. b1 to No. bn, and No. c1 to No. cn. However, the duration of the training data element is not limited to this time interval and can be time series data. For example, the training data elements No. a1 to No. an, No. b1 to No. bn, and No. c1 to No. cn can be time series data with the same duration as the input data elements No. a1 to No. an, No. b1 to No. bn, and No. c1 to No. cn, respectively.

[0044] It should be noted that in the first example of the data set generation procedure described above, a time interval No. a1, corresponding to input data element No. a1, is selected "n" times for input data elements No. a1 to No. an and used as input data. In contrast, the number of times an extension interval No. a2 is used as input data is (n-1), and an extension interval No. an is used as input data only once. Similarly, in the second example, a time interval No. b1, corresponding to input data element No. b1, is selected "n" times for input data elements No. b1 to No. bn and used as input data. In contrast, the number of times an extension interval No. b2 is used as input data is (n-1), and an extension interval No. bn is used as input data only once.This means that a multitude of input data elements, generated with an extraction reference time as a temporal reference, includes an interval that has been extracted many times and an interval that has been extracted a fewer number of times. When generating the datasets based on the extraction reference times, the dataset generation unit 22 balances the number of extractions of each of the time intervals and the extension interval across the entire training dataset. Specifically, when all sets of datasets generated with respect to the extraction reference times are combined, the dataset generation unit 22 performs histogram leveling if any variation in the number of times the time intervals are extracted is greater than or equal to a tolerance. A well-known algorithm can be adopted here as the histogram leveling process.Through such processing, the data set generation unit 22 can balance the imbalance in the number of data extractions across the entire training dataset. As a result, the data set generation unit 22 generates the datasets as a whole, without extracting only a specific interval of the training data, so that the multiple datasets thus generated uniformly reflect the operational state contained in the training data. It should be noted that, although the balancing of the number of extractions focusing on the input data has been described here, the number of extractions focusing on the training data can also be balanced.

[0045] As described above, the first through third examples of the data set generation procedure described above can generate the multitude of data sets whose extracted time intervals have different durations, based on the extraction reference time at which the operating state of machine tool 1 changes. Furthermore, in the third example, the data sets are generated in such a way that the aggregate of data contained in all data sets uniformly includes all the acquired data. Consequently, when the learning unit 23, which will be described below, learns the thermal displacement model, all the training data can be subjected to uniform training. This also has the effect of preventing overfitting, which would allow the model to match only some of the data, and improving the generalization performance of the thermal displacement model.

[0046] Additionally, the third example of the data set generation procedure described above describes the example in which the period between the two adjacent extraction reference times is divided into “n” equal intervals, but even if the period between the two adjacent extraction reference times is divided into non-uniform durations so that there is no overlap of the input data elements, an effect similar to that of the third example can be achieved.

[0047] Next, learning unit 23 will be described. Fig. Figure 9 is an explanatory description of a procedure for learning the thermal displacement model. The learning unit 23 includes a function for learning the thermal displacement model to estimate the magnitude of the thermal displacement from the input data using the input data and the training data contained in the dataset generated by the dataset generator 22. Internally, the learning unit 23 includes the thermal displacement model, which outputs the magnitude of the thermal displacement in time series when it receives the input data in time series contained in the dataset. The learning unit 23 optimizes internal parameters of the thermal displacement model to minimize the difference between an estimated value of the thermal displacement magnitude, where the estimated value is the output of the thermal displacement model, and the training data contained in the datasets.Learning Unit 23 thus determines the thermal displacement model that is able to estimate the magnitude of the thermal displacement based on the input data and outputs the thermal displacement model as a trained thermal displacement model.

[0048] The data sets output by the data set generation unit 22 of the learning device for a thermal displacement model 2 have variable lengths. That is, at least two of the multitude of data sets output by the data set generation unit 22 are extracted in time intervals of different durations. Therefore, the learning device 23 uses the thermal displacement model, which is capable of handling the variable-length input data. The learning device 23 learns the magnitude of the thermal displacement, for example, through so-called supervised learning according to a neural network. Supervised learning refers to a process that provides a learning device with the training data, which consists of sets of input and output data, thereby enabling the learning device to learn features in the training data so that it can derive the output from the input.As an example, a description is given in which learning unit 23 learns the model of thermal displacement using a recurrent neural network (RNN).

[0049] The RNN is a neural network that has a structure in which the output of a middle layer is recursively fed into the layer that produced the result. Learning Unit 23, for example, can use a neural network that has a structure in Fig. 10 is illustrated. Fig. Figure 10 is an explanatory illustration of the neural network. The one in Fig. The 10th illustrated neural network is the RNN, where an input layer “n” contains inputs from “x1” to “xn”, a middle layer “m” contains nodes from “h1” to “hm”, and an output layer contains an output of “y1”. An upper diagram in Fig. Figure 10 illustrates a state of the neural network at time t-1 and a lower diagram in Fig. Figure 10 illustrates a state of the neural network at time t. At time t, the middle layer outputs to the output layer a weighted sum of values ​​computed from inputs to the input layer at time t and terms computed from values ​​input by the middle layer at time t-1. If "x t “ represents a vector that combines the inputs “x1” to “xn” of the input layer at time t, and “h t “ represents a vector that combines the nodes “h1” to “hm” of the middle layer at time t, then “h t “can be expressed by the following formula (1). It should be noted that in formula (1), when the vector x tThe vector has one dimension "n" and the vector ht has one dimension "m", and "W" is an n×m-dimensional linear transformation matrix. Additionally, "R" represents an M×M-dimensional linear transformation matrix, "b" represents a distortion vector, and "g" represents an activation function. ht=g(Wxt+Rht−1+b)

[0050] Using backpropagation through time, learning unit 23 adjusts the weighting of the neural network such that the size of the thermal displacement, which is the output, matches the thermal displacement data.

[0051] It should be noted that, although Fig. 10. The example of the neural network illustrates that where the input layer is n-dimensional, the middle layer is m-dimensional, and the output layer is one-dimensional, none of the layers are restricted to the dimension illustrated in this example and can have any dimension. Furthermore, although Fig. 10. The example of the neural network with one middle layer illustrates that it may contain a multitude of middle layers.

[0052] It should be noted that, in order to improve the generalization performance of the thermal displacement model, learning unit 23 can also use a procedure such as “dropout”, which randomly excludes neurons at the time of learning, or “early stopping”, which monitors for errors and terminates learning early.

[0053] It should be noted that the RNNs were described using the Elman network as an example, but the learning can be performed by constructing the thermal displacement model using another well-known RNN such as a Long Short Term Memory (LSTM) or a Gated Recurrent Unit (GRU). Alternatively, the learning unit can perform learning by regression analysis using an autoregressive model.

[0054] Fig. 11 is a flowchart to explain the operation of the in Fig. 5 illustrated learning device for a thermal displacement model 2. The learning device for a thermal displacement model 2 can train the thermal displacement model by executing the in Fig. 11 illustrated processing methods are generated.

[0055] The learning data acquisition unit 21 of the learning device for a thermal displacement model 2 acquires the thermal displacement data, the temperature data, and the machine state data from the machine tool 1 (step S101). The learning data acquisition unit 21 generates the learning data by temporally linking the thermal displacement data, the temperature data, and the acquired machine state data (step S102).

[0056] The data set generation unit 22 generates the multitude of data sets from the training data (step S103). Using the data sets, the training unit 23 learns the thermal displacement model by optimizing the internal parameters of the thermal displacement model such that, in response to receiving an input of the input data contained in the data sets, the thermal displacement model outputs the magnitude of the thermal displacement that matches the training data contained in the data sets (step S104).

[0057] As described above, the data set generation unit 22 according to the first embodiment generates a plurality of data set types having variable lengths, and the thermal displacement model is trained using this plurality of data sets, from which at least two are extracted in time intervals having different durations. Consequently, the thermal displacement model training device 2 can train the thermal displacement model using the data sets generated in time intervals with suitable durations for the thermal displacement generated in the machine tool 1, thus improving the accuracy of estimating the magnitude of the thermal displacement.The learning device for a thermal displacement model 2 can improve the accuracy of estimating the magnitude of the thermal displacement regardless of the operating time of the machine tool 1.

[0058] Machine tool 1 is operated according to a machining program that features variable durations and thus different operating patterns corresponding to the final shapes of various workpiece types W. To generate a thermal displacement model capable of accurately estimating the magnitude of the thermal displacement, the training data sets must be obtained from the machining program and have variable durations.

[0059] In the first embodiment, the data sets used by the learning unit 23 to learn the thermal displacement model are generated based on the times at which the operating state of the machine tool 1, operated according to the machining program, changes. Accordingly, the learning device for a thermal displacement model 2 of the first embodiment can perform the learning of the thermal displacement model using the data sets on a temporal basis as the changes in the operating state of the machine tool 1 during actual machining. The thermal displacement model can therefore learn the input / output relationship that reflects the operating state of the machine tool.

[0060] Furthermore, in a case of using the first example and the second example for the data set generation procedure described above, the data set generation unit 22 can generate the data set by extracting, from the training data, the data which have a time width that is a multiple of the predetermined time extension width T1, based on the time point defined as the extraction reference time point.

[0061] Furthermore, since the thermal displacement model is a time-recursive model, training can be performed to calculate the magnitude of the thermal displacement, which is influenced by information contained not only in the input data entered at the time of calculating the estimated value of the thermal displacement, but also in past input data prior to that time. The learning device for a thermal displacement model 2 can therefore perform training not only with the temperature and machine state data at the time of the thermal displacement calculation, but also taking into account the influence of temporal changes in these data elements. Second embodiment.

[0062] Fig. Figure 12 is a representation illustrating an embodiment of a machining system 100-2 according to a second embodiment. The machining system 100-2 according to the second embodiment includes a machine tool 1A, the learning device 2 for the thermal displacement model, and a thermal displacement estimating device 3. The machining system 100-2 includes the thermal displacement estimating device 3 in addition to the embodiment of the machining system 100-1 according to the first embodiment and includes the machine tool 1A instead of the machine tool 1 of the machining system 100-1.

[0063] The learning device 2 for the thermal displacement model performs similar processing to that of the first embodiment, except that it outputs a trained thermal displacement model to the thermal displacement estimator 3. Therefore, the description of the processing performed by the learning device for a thermal displacement model 2 is omitted here. The thermal displacement estimator 3 receives the trained thermal displacement model generated by the learning device for a thermal displacement model 2 and input data for inference. The input data for inference includes at least temperature data representing the temperature of machine tool 1A in time series, with machine tool 1A being an inference target.The thermal displacement estimator 3 uses the thermal displacement model to generate an estimated value for the magnitude of the thermal displacement of machine tool 1A, which is the inference target, based on the input data. The thermal displacement estimator 3 then outputs the estimated value of the thermal displacement magnitude to machine tool 1A.

[0064] In addition to the functions of machine tool 1, machine tool 1A has a function for outputting temperature data and machine condition data to the thermal displacement estimator 3 and a function for correcting the thermal displacement based on the estimated value of the magnitude of the thermal displacement output by the thermal displacement estimator 3.

[0065] Fig. Figure 13 is a representation illustrating a functional configuration of the machine tool 1A according to the second embodiment. The configuration of the machine tool 1A is similar to that of machine tool 1, except that it includes a control unit 15A instead of the control unit 15 of the machine tool 1. The control unit 15A corrects the thermal displacement by adjusting the control input for at least one of the spindle motor 14, which drives the spindle 13, and the feed rod motor 12, which drives the feed rod 11, based on the estimated value of the magnitude of the thermal displacement output by the thermal displacement estimator 3. The machine tool 1A performs machining operations using the adjusted control input.

[0066] Fig. Figure 14 is a representation illustrating a functional configuration of the thermal displacement estimator 3 according to the second embodiment. The thermal displacement estimator 3 includes an inference data acquisition unit 31, a thermal displacement model acquisition unit 32, and an inference unit 33.

[0067] The inference data acquisition unit 31 acquires the temperature data and the machine state data from the machine tool 1A, which serves as the inference target, and outputs the inference data to the inference unit 33. The acquired temperature data and the machine state data are temporally synchronized in the inference data. The inference data, which includes at least the temperature data, is generated such that it exhibits a data combination similar to that of the input data used by the trained thermal displacement model at the time of learning. Furthermore, the inference data is output sequentially over the same time period as the training data generated by the training device for a thermal displacement model 2.

[0068] The acquisition unit for a thermal displacement model 32 obtains the trained thermal displacement model, used for inference, from the model storage unit 24 of the learning device for a thermal displacement model 2. The acquisition unit for a thermal displacement model 32 outputs the acquired thermal displacement model to the inference unit 33.

[0069] Inference Unit 33 uses the trained thermal displacement model to output the estimated value of the thermal displacement magnitude based on the input data contained in the inference data. Inference Unit 33 outputs the estimated value of the thermal displacement magnitude to machine tool 1A. Inference Unit 33 inputs the machine state data contained in the inference data into the trained thermal displacement model at a time similar to that of the learning process. The machine state data input into the trained thermal displacement model has a data structure similar to that at the time of learning. By inputting the machine state data in this way, Inference Unit 33 can calculate the estimated value of the thermal displacement magnitude at a time similar to that of the input.The inference unit 33 outputs the calculated estimated value of the magnitude of the thermal displacement to the control 15A of the machine tool 1A in the same period as the input cycle of the inference data.

[0070] The control unit 15A of the machine tool 1A receives the estimated value of the thermal displacement output by the inference unit 33 and corrects, for example, the position command for the feed rod 11 to compensate for the thermal displacement indicated by the estimated value. The corrected position command is then used to operate the feed rod motor 12, which is the actuator of the feed rod 11. The correction of the position command is updated for each input cycle of the inference data.

[0071] Fig. Figure 15 is a flowchart to explain the operation of the machining system 100-2, which is located in Fig. Figure 12 illustrates this. The learning device for a thermal displacement model 2 of the machining system 100-2 performs processing to generate the thermal displacement model (step S121). It should be noted that the processing in step S121 is based on the model shown in Fig. The 11 illustrated processes correspond to steps S101 to S104. These processes are carried out before any machining operation that requires thermal displacement correction.

[0072] Machine tool 1A begins program operation for the machining operation for which thermal displacement correction is required (step S122). The machining in and after step S123 is performed during the machining operation for which thermal displacement correction is required.

[0073] The inference data acquisition unit 31 receives the temperature and machine state data output sequentially during machining by machine tool 1A and generates the inference data based on the received data (step S123). The inference data has a data structure similar to that of the input data for the thermal displacement model generated in step S121. The generated inference data is output sequentially to the inference unit 33 at the same time interval as the training data generated by the training device for a thermal displacement model 2.

[0074] The inference unit 33 estimates the magnitude of the thermal displacement based on the inference data using the thermal displacement model (step S124). The estimated value of the thermal displacement magnitude is calculated for a period similar to that of the inference data generated in step S123 and is output to the controller 15A of the machine tool 1A.

[0075] The control unit 15A of machine tool 1A corrects the control input to the actuator using the estimated value of the thermal displacement, thereby correcting the position command and performing the machining operation (step S125). Machine tool 1A determines whether the program operation for the machining operation requiring thermal displacement correction has finished (step S126). If the program operation has not finished (No in step S126), the machining operation returns to step S123, and machining system 100-2 repeats the machining process from step S123 to step S126. If the program operation has finished (Yes in step S126), machining system 100-2 terminates the machining operation.

[0076] As described above, the machining system 100-2 according to the second embodiment includes the thermal displacement estimator 3, which estimates the magnitude of the thermal displacement using the trained thermal displacement model. Furthermore, the machine tool 1A corrects the control input for the actuator of the machine tool 1A based on the estimated value of the thermal displacement magnitude in order to compensate for the thermal displacement indicated by the estimated value.

[0077] Additionally, the inference unit 33 can sequentially calculate the estimated value of the thermal displacement magnitude for successively input inference data and output the estimated value to machine tool 1A. Therefore, regardless of the duration of the machining program executed by machine tool 1A, the thermal displacement magnitude can be estimated. Furthermore, since the estimated value of the thermal displacement magnitude is calculated using a time-recursive thermal displacement model, the thermal displacement magnitude can be estimated not only from the temperature and machine state data at the time of calculation, but also takes into account the influence of temporal changes in this data.The machine tool 1A can therefore accurately correct the thermal displacement regardless of the time elapsed since the start of operation of the machine tool 1A.

[0078] It should be noted that another procedure is described in step S126, which is in Fig. As illustrated in Figure 15, it can be determined whether machine tool 1A is switched off or not, and the machining system 100-2 can return to step S123 and continue processing until machine tool 1A is switched off. In this case, the thermal displacement correction can continue even if the program operation is carried out many times in succession.

[0079] A hardware design of the learning device for a thermal displacement model 2 is described. The learning device for a thermal displacement model 2 of the first and second embodiments is implemented, for example, by a computer system. The learning device for a thermal displacement model 2 can be implemented by a single computer system or by a plurality of computer systems. For example, the learning device for a thermal displacement model 2 can be implemented by a cloud system. In the cloud system, the separation between the hardware of the computer system and devices, such as a server, can be arbitrarily defined for each function. For example, a computer system can have a function as a plurality of devices, or a plurality of computer systems can have a function as a single device.

[0080] An exemplary configuration of the computer system that implements the learning device for a model of thermal displacement 2 is described. Fig. Figure 16 is a diagram illustrating an example of the hardware design. As in Fig. As illustrated in Figure 16, this computer system includes a control unit 101, an input unit 102, a storage unit 103, a display unit 104, a communication unit 105 and an output unit 106, which are connected via a system bus 107.

[0081] In Fig. 16. The control unit 101 is, for example, a central processing unit (CPU) or the like. The control unit 101 executes a learning program for the thermal displacement model, in which each processing operation performed by the learning device for a thermal displacement model 2 of the present embodiment is described. The input unit 102 includes, for example, a keyboard, a mouse, and the like, and is used by a user of the computer system to input various pieces of information. The memory unit 103 includes various types of memory, such as random access memory (RAM) and read-only memory (ROM), as well as a storage device, such as a hard disk, and stores a program to be executed by the aforementioned control unit 101, necessary data acquired during processing, and the like.The memory unit 103 is also used as temporary memory for the program. The display unit 104 includes a liquid crystal display (LCD) or similar device and shows various screens to the user of the computer system. The communication unit 105 is a communication circuit or similar device that performs communication processing. The communication unit 105 can include a variety of communication circuits corresponding to a variety of respective communication systems. The output unit 106 is an output interface that outputs data to an external device, such as a printer or an external storage device.

[0082] It should be noted that Fig. 16 illustrates an example and does not restrict the design of the computer system. For example, the computer system need not include the output unit 106. Furthermore, in a case where the learning device for a model of thermal displacement 2 is implemented by a multitude of computer systems, not all of these computer systems need to include the one shown in Fig. 16 illustrated computer systems. For example, some of the computer systems do not need to include at least one of the display unit 104, the output unit 106, and the input unit 102 shown in Fig. The 16 illustrations are included.

[0083] A description of exemplary operation of the computer system is provided before the thermal displacement model learning program is ready for execution, wherein the thermal displacement model learning program describes the processing of the learning device for a thermal displacement model 2 of the first embodiment and the second embodiment. In the computer system having the foregoing configuration, the thermal displacement model learning program is installed on the storage unit 103, for example, from a compact disc ROM (CD-ROM) or a digital versatile disc ROM (DVD-ROM) inserted into a CD-ROM drive or a DVD-ROM drive, which is not illustrated. During execution, the thermal displacement model learning program read from the storage unit 103 is stored in an area that serves as the main memory of the storage unit 103.In this state, the control unit 101 performs the processing as the learning device for a thermal displacement model 2 of the first embodiment and the second embodiment according to the learning program for the thermal displacement model stored in the storage unit 103.

[0084] It should be noted that the foregoing description provides the program that describes the processing of the learning device for a model of thermal displacement 2 using the CD-ROM or DVD-ROM as a recording medium, but depending on the design of the computer system, the capacity of the provided program and the like, the program can be provided, for example, for use by a transmission medium such as the Internet via the communication unit 105.

[0085] The training program for the thermal displacement model in the first embodiment and the second embodiment causes a computer to perform the following: a step to obtain the training data in which the input data, which includes at least the temperature data representing the temperature of machine tool 1 in time series, is temporally linked with the thermal displacement data indicating the magnitude of the thermal displacement of machine tool 1 in time series; a step to generate the data sets obtained by extracting a portion of the training data from the training data; and a step to generate the trained thermal displacement model using the data sets to estimate the magnitude of the thermal displacement based on the input data of machine tool 1A.In the step of generating the data records, the data records are generated in such a way that at least two of the multitude of data records are extracted in the time intervals, which have different durations.

[0086] The in Fig. 5 illustrated learning data acquisition unit 21 is provided by the in Fig. 16 illustrated communication unit 105 implemented, which in Fig. 5 illustrated model storage unit 24 is part of the in Fig. 16 illustrated storage unit 103 and the in Fig. 5. The illustrated data set generation unit 22 and the learning unit 23 are represented by the in Fig. 16 illustrated control unit 101 implemented.

[0087] As with the learning device for a thermal displacement model 2, the thermal displacement estimating device 3 is also implemented by one or a plurality of computer systems. A program for the thermal displacement estimating device 3 to perform the operation described in the second embodiment is provided by a storage medium, a transmission medium, or the like and installed on the computer system, as with the learning program for the thermal displacement model described above. Consequently, the operation of the thermal displacement estimating device 3 described above is implemented.

[0088] The inference data acquisition unit 31 and the acquisition unit for a thermal displacement model 32, which are in Fig. 14 are illustrated by the in Fig. 16 illustrated communication unit 105 implemented and the in Fig. 14 illustrated inference unit 33 is provided by the in Fig. 16 illustrated control unit 101 implemented.

[0089] It should be noted that the definition of the functions of each in Fig. 5 and Fig. The device illustrated in Figure 14 is an example and not exhaustive. The functions of each device can be defined in any way that enables the machining systems 100-1 and 100-2 to perform the operations described above. For example, as described below as a variation, the thermal displacement model learning device 2 and the thermal displacement estimating device 3 can be integrated, such that the thermal displacement model learning device 2 also incorporates the functions of the thermal displacement estimating device 3.

[0090] Fig. Figure 17 is a representation illustrating a functional configuration of a learning device for a thermal displacement model 2A according to the variation of the second embodiment. The learning device for a thermal displacement model 2A includes the functions of the thermal displacement estimator 3 in addition to the functions of the learning device for a thermal displacement model 2. Since the function of each component is defined with reference to Fig. Since item 14 was described, its description is omitted here. In the Fig. In the illustrated embodiment 17, in a case where the functions of the learning device for a thermal displacement model 2A are implemented by a computer system, the function of the acquisition unit for a thermal displacement model 32 is implemented by exchanging data within the computer system. Third embodiment.

[0091] Fig. Figure 18 illustrates a third embodiment of a machining system 100-3. The machining system 100-3 comprises a plurality of machine tools 1 and a thermal displacement model learning device 2B, which is provided on a server. The thermal displacement model learning device 2B learns a thermal displacement model based on thermal displacement data, temperature data, and machine state data received from the plurality of machine tools 1. The functions of the machine tools 1 are similar to those of the first embodiment, and the functions of the thermal displacement model learning device 2B are similar to those of the first embodiment, except that the thermal displacement model learning is performed for each of the plurality of machine tools 1. Fourth embodiment.

[0092] Fig. Figure 19 illustrates a fourth embodiment of a machining system 100-4. The machining system 100-4 includes a plurality of machine tools 1 and 1A, a learning device for a thermal displacement model 2B provided on a server, and a thermal displacement estimation device 3B.

[0093] The machine tools 1, each of which is a machining setup for a learning objective, are similar to the machine tool 1 of the first embodiment, except that they output feature information to the learning device for a thermal displacement model 2B. The feature information output to the learning device for a thermal displacement model 2B specifies features of the machine tools 1, including at least one model number of the machine tools 1. The machine tool 1A, which is a machining setup serving as an estimation target, is similar to the machine tool 1A of the second embodiment, except that it outputs feature information to the estimation device for thermal displacement 3B. The feature output to the estimation device for thermal displacement 3B specifies a feature of the machine tool 1A, including at least one model number of the machine tool 1A.

[0094] The learning device for a thermal displacement model 2B, which is similar to the learning device for a thermal displacement model 2B of the third embodiment, except for receiving feature information from the machine tools 1, includes a learning data acquisition unit 21B and a model storage unit 24B. The learning data acquisition unit 21B generates learning data based on the data received from the plurality of machine tools 1. The model storage unit 24B stores a plurality of thermal displacement models, each associated with feature information about the machine tools 1 from which the learning data is acquired. The thermal displacement estimator 3B includes a thermal displacement model acquisition unit 32B instead of the thermal displacement model acquisition unit 32 of the thermal displacement estimator 3.The acquisition unit for a thermal displacement model 32B includes a model selection unit 321. Based on the feature information received from the machine tool 1A as the inference target, a model selection unit 321 selects the thermal displacement model to be used by the inference unit 33 from the multitude of thermal displacement models stored in the model storage unit 24B.

[0095] The feature information includes at least the model number and may, in addition to the model number, include information specifying the situations of machine tools 1 and 1A, such as the factory environment in which machine tools 1 and 1A are located, and the conditions of their location. The model selection unit 321 can select the thermal displacement model generated from the data obtained from machine tool 1, in a state similar to that of machine tool 1A, as the inference target, provided that machine tool 1 has the same model number as machine tool 1A as the inference target.Consequently, even if no thermal displacement model exists that was generated from the data of machine tool 1A itself, which is the inference target, an estimated value of the magnitude of the thermal displacement can be obtained using the thermal displacement model generated from the data of machine tool 1, which is used in a similar environment, etc., to that of machine tool 1A, where machine tool 1 has the same model number as machine tool 1A.

[0096] It should be noted that, although the foregoing description is such that the acquisition unit for a thermal displacement model 32B includes the model selection unit 321, the acquisition unit for a thermal displacement model 32B can obtain a plurality of types of thermal displacement models from the model storage unit 24B, and the inference unit 33 can select the estimated value to be output from a plurality of estimated values ​​of a thermal displacement calculated using the plurality of types of thermal displacement models, based on the feature information.

[0097] The embodiments illustrated in the foregoing embodiments each represent an example, so that another known technique can be combined, the embodiments can be combined with each other, or the embodiments can be partially omitted and / or modified without deviating from the scope of the present disclosure.

[0098] For example, the functions of the learning device for a thermal displacement model 2 and the estimating device for thermal displacement 3 can be integrated into the machine tool 1. Furthermore, as in the third and fourth embodiments, where the thermal displacement model is learned using data obtained from the plurality of machine tools 1, the learning device for a thermal displacement model 2B can learn the thermal displacement model, obtain the learning data from the plurality of machine tools 1 used in the same area, or use the learning data collected from the plurality of machine tools 1 operating independently in different areas. In addition, the machine tool 1 from which the learning data is collected can be added or removed during the process.Furthermore, the learning device for a thermal displacement model 2, which has learned the thermal displacement model for one of the machine tools 1, can be applied to another of the machine tools 1, and the thermal displacement model can be relearned and updated for the other machine tool 1. List of reference symbols

[0099] 1, 1A Machine tool; 2, 2A, 2B Learning device for a thermal displacement model; 3, 3B Estimating device for thermal displacement; 11 Feed rod; 12 Feed rod motor; 13 Spindle; 14 Spindle motor; 15, 15A Control; 16 Peripheral device; 17 Bed; 18 Column; 19 Feed spindle; 21, 21B Learning data acquisition unit; 22 Data set generation unit; 23 Learning unit; 24, 24B Model storage unit; 31 Inference data acquisition unit; 32, 32B Thermal displacement model acquisition unit; 33 Inference unit; 100-1, 100-2, 100-3, 100-4 Machining system; 101 Control unit; 102 Input unit; 103 Storage unit; 104 Display unit; 105 Communication unit; 106 Output unit; 321 Model selection unit; DS Displacement sensor; T Tool; TA Table; TS Temperature sensor; W Workpiece. QUOTES INCLUDED IN THE DESCRIPTION

[0000] This list of documents cited by the applicant was automatically generated and is included solely for the reader's convenience. The list is not part of the German patent or utility model application. The DPMA accepts no liability for any errors or omissions. Cited patent literature

[0000] JP 2019-111648

[0005]

Claims

[1] A thermal displacement model learning device comprising: a learning data acquisition unit for acquiring learning data in which input data and thermal displacement data are temporally linked, wherein the input data includes at least temperature data representing a temperature of a machine tool in time series, wherein the thermal displacement data represents a quantity of thermal displacement of the machine tool in time series; a data set generation unit for generating data sets obtained by extracting a part of the learning data from the learning data; and a learning unit for generating, using the data set, a thermal displacement model trained to estimate the magnitude of the thermal displacement from the input data to the machine tool, wherein the data set generation unit generates the data sets such that at least two of the data sets are extracted from a plurality of the data sets at time intervals having different durations from one another. [2] A thermal displacement model learning apparatus according to claim 1, wherein the input data further includes, in addition to the temperature data, machine state data representing an operating state of the machine tool. [3] The thermal displacement model learning apparatus according to claim 1 or 2, wherein the learning unit generates the trained thermal displacement model by optimizing an internal parameter of the thermal displacement model to minimize a difference between an estimated value of the thermal displacement magnitude, the estimated value being an output from the thermal displacement model when the input data included in the data sets are input to the thermal displacement model, and the thermal displacement data corresponding to the input data input to the thermal displacement model. [4] A thermal displacement model learning device according to any one of claims 1 to 3, further comprising: an inference data acquisition unit for acquiring inference data, which is input data for the inference, including at least the temperature data representing the temperature of the machine tool in time series; and an inference unit for outputting, using the thermal displacement model, an estimated value of the thermal displacement magnitude of the machine tool from the inference data input from the inference data acquiring unit. [5] A thermal displacement model learning apparatus according to claim 4, wherein the inference data further includes, in addition to the temperature data, machine state data representing an operating state of the machine tool. [6] A thermal displacement model learning apparatus according to claim 2 or 5, wherein the machine state data indicates an operation status of a peripheral device for adjusting the temperature of the machine tool and an operation status of an actuator of the machine tool. [7] A thermal displacement model learning device according to any one of claims 1 to 3, wherein the learning data acquisition unit acquires the learning data from a plurality of machine tools, and the learning unit generates the thermal displacement model for each of the machine tools. [8] A thermal displacement model learning device according to any one of claims 1 to 3, wherein the learning data acquisition unit acquires the learning data from a plurality of machine tools, and the learning unit generates the thermal displacement model for each of the machine tools and stores the generated thermal displacement model and feature information in association with each other, wherein the feature information is information indicating a feature of the machine tool and includes at least a model number of the machine tool from which the learning data used to generate the thermal displacement model is obtained. [9] A thermal displacement model learning device according to claim 8, further comprising: an inference data acquisition unit for acquiring inference data, which is input data for the inference and includes at least the temperature data representing the temperature of the machine tool in time series, the machine tool being an inference target; and an inference unit for outputting, using one thermal displacement model from a plurality of thermal displacement models, an estimated value of the thermal displacement magnitude of the machine tool that is the inference target from the inference data input from the inference data acquisition unit, wherein the thermal displacement model has the same model number as the machine tool that is the inference target. [10] A thermal displacement model learning device according to claim 9, wherein the feature information further includes, in addition to the model number, information indicating an operating condition of the machine tool, and the inference unit outputs the estimated value of the thermal displacement magnitude of the machine tool that is the inference target using the thermal displacement model generated based on the learning data acquired from the machine tool that has the same model number as the machine tool that is the inference target and that has a similar operating condition to that of the machine tool that is the inference target. [11] Thermal displacement estimation device comprising: a thermal displacement model acquiring unit for acquiring the thermal displacement model trained by the thermal displacement model learning device according to claim 1; an inference data acquisition unit for acquiring estimation data, which is input data for the inference, including at least temperature data representing a temperature of the machine tool in time series; and an inference unit for outputting, using the thermal displacement model acquired by the thermal displacement model acquiring unit, an estimated value of the thermal displacement magnitude from the input data input by the inference data acquiring unit. [12] Processing system comprising: the thermal displacement model learning device according to any one of claims 1 to 10 or the thermal displacement estimating device according to claim 11; and the machine tool, where the machine tool includes: a tool; a spindle for rotating the tool or a machining target; and a feed rod for changing a relative position between the tool and the object to be machined, and the machine tool uses the tool to carry out a machining operation on the object to be machined. [13] A machining method using a machine tool including a tool, a spindle for rotating the tool or a machining target, and a feed rod for changing a relative position between the tool and the machining target, the machine tool performing machining on the machining target with the tool, the machining method comprising: a step of acquiring learning data in which input data and thermal displacement data are temporally linked, wherein the input data includes at least temperature data representing a temperature of the machine tool in time series, wherein the thermal displacement data represents a quantity of thermal displacement of the machine tool in time series; a step of generating data sets obtained by extracting a part of the learning data from the learning data; a step of generating, using the data set, a thermal displacement model trained to estimate the magnitude of the thermal displacement from the input data to the machine tool; a step of estimating, using the trained thermal displacement model, the magnitude of the thermal displacement from input data for the inference including at least the temperature data; a step of correcting, using an estimated value of the magnitude of the thermal displacement, a control input for at least one of an actuator for driving the spindle of the machine tool or an actuator for driving the feed rod; and a step of performing machining using the corrected control input, wherein the step of generating the data set generates the data set such that at least two of the data sets are extracted from a plurality of the data sets at time intervals having different durations from one another.

Images