JP2022023408A - Machining system and trained model generation device - Google Patents

Abstract

To provide a machining system that predicts machining accuracy using data obtained from a machine tool and a trained model and that determines a correction value based on the predicted machining accuracy, and a trained model generation device that generates the trained model.SOLUTION: The machining system 1 includes a learning processing device 30 and a correction value determining device 40. The learning processing device 30 generates a trained model by machine learning using processing state data PD and inter-core distance data LD acquired from a data collecting device 20 as a training data set. The correction value determining device 40 predicts an inter-core distance L using the machining state data PD acquired from the data collecting device 20 and the generated trained model, calculates a correction value C based on the predicted inter-core distance L, and outputs the correction value C to a machine tool.SELECTED DRAWING: Figure 6

Description

The present invention relates to a machining system and a trained model generator.

Conventionally, for example, in Patent Document 1, machine learning is performed based on sampling data that can be obtained from a grinding machine and values that represent grinding characteristics, and a trained model that can acquire the grinding quality of a workpiece is generated. The technology to be used is disclosed.

Japanese Unexamined Patent Publication No. 2020-44620

By the way, when machining a workpiece, for example, even if the machining conditions and various settings of the machine tool are determined in advance using a learned model, the machining accuracy gradually deteriorates as the number of workpieces to be machined increases. In some cases. Therefore, the operator needs to inspect the machining accuracy of the machined workpiece at a predetermined frequency, and also needs to appropriately change the machining conditions and various settings of the machine tool according to the deterioration of the machining accuracy. .. In this case, it is complicated to inspect the machining accuracy for each machined workpiece using an inspection device, and when the operator changes the machining conditions, the operator's experience (intuition and tips) is used. Since it depends on it, the processing accuracy may vary.

The present invention is a machining system that predicts machining accuracy using data obtained from a machine tool and a trained model, and determines a correction value based on the predicted machining accuracy, and a trained model that generates a trained model. It is intended to provide a generator.

The machining system has a machine tool for machining a workpiece having a meshing structure and an observation device for observing the machining state of the workpiece by the machining machine and collecting machining state data representing the machining state. Machine learning is performed using the collected machining state data as explanatory variables, the machining accuracy data related to the machining accuracy of the meshing structure machined by the machine tool as the objective variable, and the learning data set including the explanatory variables and the objective variables. The trained model storage unit that stores the trained model generated by the operation, the machining accuracy prediction unit that predicts the machining accuracy using the trained model and the machining state data, and the machining accuracy prediction unit predict the machining accuracy. It is provided with a correction value determining unit that calculates and determines a correction value that corrects a command value for operating the machine tool based on machining accuracy, and an output unit that outputs the correction value to the machine tool.

The trained model generator uses the machining state data representing the machining state of the workpiece by the machine tool that processes the workpiece having the meshing structure as an explanatory variable, and the machining accuracy data related to the machining accuracy of the meshing structure machined by the machine tool. It is provided with a trained model generation unit that generates a trained model by performing machine learning using a training data set including explanatory variables and objective variables as objective variables.

According to these, it is possible to generate a trained model showing the correlation between the machining state data observed when the machine tool processes the workpiece and the machining accuracy data representing the machining accuracy of the machined workpiece. can. Then, in the machining system, the trained model and the machining state data are used to predict the machining accuracy of the machined workpiece and correct the command value for operating the machine tool based on the predicted machining accuracy. The value can be determined.

As a result, since the machine tool can predict the machining accuracy of the workpiece while machining the workpiece, it is not necessary to inspect the machining accuracy for each machined workpiece using an inspection device. Therefore, the productivity of the workpiece can be improved. In addition, since the correction value is determined based on the machining accuracy predicted using the generated trained model, it is possible to suppress variations in machining accuracy due to dependence on the operator's experience (intuition and tips). As a result, uniform processing accuracy can be achieved.

It is a figure which shows the structure of a processing system. It is a front view which shows the worm shaft which is an example of a workpiece. The configuration of a screw grinding machine which is an example of a machine tool is shown. It is a figure which shows the structure of the data acquisition apparatus which is an example of an observation apparatus. It is a figure for demonstrating the inter-core distance which is an example of processing accuracy data. It is a figure which shows the functional block composition of the processing system (learned model generator).

(1. Outline of processing system)
The machining system is a system having a machine tool for machining a workpiece having a meshing structure and an observation device for observing the machining state of the workpiece machined by the machine tool. Then, the machining system uses the trained model generated by machine learning and the machining state data representing the machining state observed by the observation device, and the command value (machining condition or machine tool of the machine tool) for machining the workpiece. Machine settings, etc.) are corrected.

Here, the observation device includes various sensors provided in the machine tool. Machine tools for machining workpieces include various cutting devices for cutting (for example, gear machining devices, machining centers, etc.) and various grinding devices for grinding (screw grinders, cylindrical grinders, flat plate grinders, etc.). ) Can be mentioned. Further, the workpiece machined by the machine tool may have a screw structure or a gear structure as a meshing structure, for example, a worm shaft, a rack bar, a spur tooth gear, a tooth gear and the like. In this example, as will be described later, a case where the machine tool is a screw grinding machine for grinding is exemplified.

(2. Outline of the configuration of machining system 1)
An outline of the configuration of the processing system 1 will be described with reference to FIG. The machining system 1 is a thread grinding machine 10 as at least one machine tool, and an observation device that observes the machining state when the thread grinding machine 10 machining a workpiece and collects machining state data representing the machining state. A data collecting device 20 and one arithmetic device (30, 40) are provided. Here, as a workpiece machined by the screw grinding machine 10, in this example, as shown in FIG. 2, a worm shaft W having a screw portion W1 (screw structure) having a meshing structure and mounted on a vehicle is used. It will be illustrated and described.

The data acquisition device 20 includes various sensors provided in the screw grinding machine 10 and the inspection device 2 (see FIG. 4). The data collection device 20 functions as a database that can updately store the processing state data acquired from various sensors.

The arithmetic unit (30, 40) applies machine learning using the processing state data accumulated in the data collecting device 20, so that the screw grinding machine 10 processes the threaded portion W1 of the worm shaft W. A correction value for correcting a command value (machining conditions, machine settings, etc.) of the board 10 is determined. In FIG. 1, the arithmetic unit (30, 40) is composed of a learning processing device 30 as a trained model generation device and a correction value determining device 40, and the learning processing device 30 and the correction value determining device 40 are independent of each other. Although shown as a configuration, it can also be a single device. Further, a part or all of the arithmetic unit (30, 40) can be incorporated into the screw grinding machine 10.

In this example, the case where the learning processing device 30 and the correction value determining device 40 have independent configurations will be taken as an example. Further, the learning processing device 30 has a so-called server function, and when there are a plurality of screw grinding machines 10, they are communicably connected to a correction value determining device 40 provided in each screw grinding machine 10. There is. The correction value determining device 40 is provided one-to-one with each screw grinding machine 10 and is communicably connected to the corresponding screw grinding machine 10. That is, the correction value determining device 40 functions as a so-called edge computer, and can realize high-speed arithmetic processing.

(3. Configuration of screw grinding machine 10)
The configuration of the screw grinding machine 10 will be described with reference to FIG. The screw grinding machine 10 is a machine tool for grinding the threaded portion W1 of the worm shaft W, which is a workpiece. In this example, the screw grinding machine 10 exemplifies a table traverse type grinding machine. However, it is also possible to apply the grindstone base traverse type to the screw grinding machine 10.

The screw grinding machine 10 mainly includes a bed 11, a movable table 12, a headstock 13, a tailstock 14, a moving table 15, a rotary table 16, a grindstone table 17, a grindstone 18, a coolant device 19, and a control device S. Be prepared.

The bed 11 is fixed on the installation surface. The movable table 12 is installed on the bed 11 so as to be horizontally movable and relatively movable in the Z-axis direction with respect to the bed 11. The movable table 12 reciprocates in the Z-axis direction by being driven by a motor 12a provided on the bed 11. In the following description, the direction orthogonal to the Z-axis direction on the horizontal plane is defined as the X-axis direction, and the direction orthogonal to the Z-axis direction and the X-axis direction is defined as the Y-axis direction.

The headstock 13 is provided on the upper surface of the movable table 12 on the front side in the X-axis direction (lower side in FIG. 3) and one end side in the Z-axis direction (left side in FIG. 3). The headstock 13 rotatably supports the worm shaft W around the Z axis. The worm shaft W is rotated by driving a motor 13a provided on the headstock 13. The tailstock 14 is located on the upper surface of the movable table 12 at a position facing the headstock 13 in the Z-axis direction, that is, the front side in the X-axis direction (lower side in FIG. 3) and the other end side in the Z-axis direction. It is provided (on the right side of FIG. 3). That is, the headstock 13 and the tailstock 14 rotatably support both ends of the worm shaft W.

The moving table 15 is installed on the back side of the bed 11 in the X-axis direction (upper side in FIG. 3) so as to be horizontally movable and relatively movable in the X-axis direction with respect to the bed 11. The moving table 15 reciprocates in the X-axis direction by being driven by a motor 15a provided on the bed 11. The rotary table 16 is rotatably provided around the X axis on the upper surface of the mobile table 15. The rotary table 16 rotates about the X axis by driving the motor 16a provided on the mobile table 15.

The grindstone table 17 is provided on the upper surface of the rotary table 16. The grindstone wheel 18 is rotatably supported by the grindstone stand 17. The grindstone 18 is rotated by driving a motor 17a provided on the grindstone base 17. The coolant device 19 supplies the coolant to the grinding point at the threaded portion W1 of the worm shaft W by the grindstone 18. The coolant device 19 cools the recovered coolant to a predetermined temperature and supplies it to the grinding point again.

The control device S controls each drive device based on an NC program generated based on operation command data such as the shape of the threaded portion W1 of the worm shaft W, the machining conditions, the shape of the grindstone 18, and the coolant supply timing information. Control. That is, the control device S inputs the operation command data, generates an NC program based on the operation command data, and controls the motors 12a, 13a, 15a, 16a, 17a, the coolant device 19, and the like based on the NC program. As a result, the threaded portion W1 of the worm shaft W is ground. Here, the actual operation command data includes, for example, a command cutting speed, a command position, a command rotation speed of the grindstone 18, a command rotation speed of the worm shaft W, coolant supply information, and the like.

Further, although not shown in FIG. 3, the screw grinding machine 10 includes various sensors 21, 22, 23 (shown in FIG. 4) described later that constitute the data acquisition device 20. For example, the screw grinding machine 10 is supplied to the grinding point by a sensor that detects actual operation data of each motor or the like, data indicating an operating state, a vibration state, etc. of the structural members constituting the screw grinding machine 10, particularly a coolant device 19. It is equipped with a temperature sensor or the like that detects the temperature of the coolant. The details of the various sensors 21 to 23 will be described later.

Next, the operation of the screw grinding machine 10 will be briefly described. When the screw grinding machine 10 grinds the threaded portion W1 of the worm shaft W, first, the worm shaft W is supported so as to be horizontal between the headstock 13 and the tailstock 14. Then, the control device S drives each of the motors 12a, 13a, 15a, 16a, 17a. As a result, the threaded portion W1 of the worm shaft W is ground by using the grindstone 18 that rotates while relatively moving the grindstone 18 and the worm shaft W.

Here, when the grindstone 18 grinds the threaded portion W1, the grindstone 18 is tilted by an amount corresponding to the angle of the thread groove of the threaded portion W1. That is, the control device S rotates the rotary table 16 around the X axis and adjusts the grindstone 18 so that it has a predetermined inclination angle (specifically, the angle of the thread groove). Then, the screw grinding machine 10 moves the grindstone 18 in the X-axis direction, presses the grindstone 18 against the thread groove in the screw portion W1 of the worm shaft W, and grinds the screw portion W1.

The data collecting device 20 is an observation device that acquires sampling data acquired with the operation of the screw grinding machine 10. The sampling data collected by the data acquisition device 20 is used as learning data for machine learning. Here, the sampling data constitutes a time-series data group in a predetermined sampling cycle for each worm shaft W.

As shown in FIG. 4, the data collection device 20 includes various sensors 21, 22, 23, and a storage unit 24 for accumulating and storing various collected data. The sensor 21 is provided on the screw grinding machine 10 and collects actual operation data representing the operating state of the drive device controlled by the control device S, for example, the motors 12a, 13a, 15a, 16a, 17a and the like. Examples of the actual operation data include the drive current and the actual position of each of the motors 12a, 13a, 15a, 16a, 17a and the like.

The sensor 22 is provided on the screw grinding machine 10 and collects the first actual measurement data when the grindstone 18 is grinding the screw portion W1 of the worm shaft W. Examples of the first actual measurement data include vibration and deformation amount of the grindstone stand 17 which is a structural member. The sensor 23 is provided on the screw grinding machine 10 and collects the second actual measurement data when the screw portion W1 of the worm shaft W is being ground by the grindstone 18. The sensor 23 is also provided in the inspection device 2 and collects the second actual measurement data when the screw portion W1 ground by the screw grinding machine 10 is inspected. The second actual measurement data includes, for example, the dimension of the threaded portion W1 of the worm shaft W in the screw grinding machine 10, the grinding point temperature (including the coolant temperature), and the core spacing corresponding to the machining accuracy detected by the inspection by the inspection device 2. The distance etc. can be mentioned.

Here, each of the various sensors 21, 22, and 23 acquires actual operation data, first actual measurement data, and second actual measurement data for a predetermined period for each worm shaft W. The predetermined period is, for example, from the initial stage of grinding to the final stage of grinding, from the initial stage of rough grinding to the final stage of rough grinding, and the like. Since the grinding is unstable when it is in an unsteady state, it is possible to collect data only in the steady state.

As shown in FIG. 1, the learning processing device 30 as a learned model generation device includes a processor 31, a storage device 32, an interface 33, and the like. Further, the learning processing device 30 has a server function, and is connected to a data collecting device 20 (or a screw grinding machine 10 of another factory separated from the screw grinding machine 10) connected to each of the plurality of screw grinding machines 10. It is communicably connected to (including the data collecting device 20).

The learning processing device 30 outputs the processing state data PD (specifically, at least one of the actual operation data, the first actual measurement data, and the second actual measurement data) output from the data acquisition device 20 and the inspection device 2. Machine learning is performed based on the inter-core distance data LD representing the inter-core distance L as the machining accuracy data related to the processed machining accuracy. Then, the learning processing device 30 generates a trained model for predicting the machining accuracy of the threaded portion W1 of the worm shaft W.

In this example, of the second actual measurement data acquired by the data acquisition device 20, the case where the second actual measurement data regarding the coolant temperature is used as the processing state data PD is exemplified. This is based on the knowledge obtained from repeated experiments and the like, and in the thread grinding machine 10 of this example, the machining state data PD regarding the coolant temperature and the inter-core distance data LD (inter-core distance L) which is the machining accuracy data. Based on the high correlation with.

Therefore, other sampling data as the machining state data PD, for example, the current value supplied to the motor 13a of the spindle base 13 and the motor 17a of the grindstone base 17 collected as the actual operation data, or the first actual measurement data is collected. It is not hindered to use the vibration amplitude and the like of the headstock 13 and the tailstock 14 (that is, the movable table 12). In particular, when a machine tool other than the screw grinding machine 10 is used, it goes without saying that it is possible to use other sampling data collected by the data collecting device 20 as the machining state data PD.

Here, the inter-core distance L as the processing accuracy inspected by the inspection device 2 will be described. As shown in FIG. 5, the core-to-core distance L is the worm shaft W in a state where the worm shaft W ground by the screw grinding machine 10 and the master worm gear Wm as a reference member having a gear having a meshed structure are meshed with each other. Represents the distance between the axis of the master worm gear and the axis of the master worm gear Wm. Therefore, in the inspection device 2, the inter-core distance L is measured (inspected) with the worm shaft W ground by the screw grinding machine 10 actually meshed with the master worm gear Wm, and the inter-core distance L is calculated. When the specified value is L0 or less, it is judged as a good product because it has good processing accuracy. The inter-core distance L is detected by the sensor 23 of the data acquisition device 20 described above.

Each correction value determining device 40 includes a processor 41, a storage device 42, an interface 43, and the like. Further, the correction value determining device 40 is communicably connected to the learning processing device 30 as a server and the data collecting device 20 connected to the screw grinding machine 10.

The correction value determining device 40 is arranged corresponding to each screw grinding machine 10 and functions as an edge computer. The correction value determining device 40 uses the trained model generated by the learning processing device 30 and the machining state data PD acquired from the data collecting device 20 during the grinding of the threaded portion W1 of the worm shaft W for screw grinding. The correction value C for correcting and correcting the operation command data (NC program) for operating the board 10 is determined. Then, the correction value determining device 40 outputs the determined correction value C to the control device S of the screw grinding machine 10.

The processing system 1 further includes a display device 50. The display device 50 is arranged corresponding to the screw grinding machine 10. However, the processing system 1 may be configured not to include the display device 50.

(4. Functional block configuration of machining system 1)
The functional block of the machining system 1 will be described with reference to FIG. As described above, the processing system 1 includes an inspection device 2, a screw grinding machine 10, a data acquisition device 20, a learning processing device 30, a correction value determination device 40, and a display device 50.

The learning processing device 30 is a processing state data PD collected by the data collection device 20, a processing state data PD relating to the temperature of the coolant in this example, and a core distance L representing the core distance L output from the inspection device 2. A trained model for determining the correction value C is generated based on the data LD. The learning processing device 30 as a learned model generation device includes a learning data set acquisition unit 61, a learning data set storage unit 62, and a model generation unit 63.

The learning data set acquisition unit 61 acquires a learning data set for performing machine learning. The learning data set acquisition unit 61 includes a processing state data acquisition unit 61a and an intercore distance data acquisition unit 61b. The machining state data acquisition unit 61a acquires, for example, the machining state data PD regarding the temperature of the coolant acquired when one worm shaft W is ground from the data acquisition device 20 as a learning data set. Here, the machining state data PD of this example includes temperature data TD0 representing the temperature of the coolant stored in the storage unit 24 of the data acquisition device 20 during the previous cutting process in the grinding process of the worm shaft W. The temperature data TD1 representing the temperature of the coolant acquired at the time of the current cutting process is acquired as the machining state data PD.

The core-to-core distance data acquisition unit 61b is a core-to-core distance data LD representing, for example, the core-to-core distance L between one ground worm shaft W and the master worm gear Wm from the inspection device 2 via the data acquisition device 20. As a training data set.

The learning data set storage unit 62 stores the processing state data PD acquired by the learning data set acquisition unit 61 and the intercore distance data LD in association with each other as a learning data set. That is, the learning data set storage unit 62 was measured by engaging the master worm gear Wm with the machined state data PD having the temperature data TD0 and the temperature data TD1 and the threaded portion W1 of the worm shaft W machined this time. It is stored in association with the inter-core distance data LD, which is the inter-core distance L.

The model generation unit 63 performs machine learning using the learning data set stored in the learning data set storage unit 62. Specifically, the model generation unit 63 uses the processing state data PD stored in association with the learning data set storage unit 62 as an explanatory variable, and aims at the inter-core distance data LD (that is, the inter-core distance L). Perform machine learning as a variable. In this case, the model generation unit 63 includes temperature data TD0 (temperature of the coolant collected at the time of the previous grinding) and temperature data TD1 (temperature of the coolant collected at the time of the current grinding) included in the processing state data PD. The difference between the two, that is, the amount of change (TD1-TD0) is calculated (extracted) as the feature amount, and the calculated (extracted) amount of change (feature amount) is used as the explanatory variable. Then, the model generation unit 63 generates a trained model showing the correlation between the machining state data PD and the inter-core distance data LD.

The correction value determining device 40 estimates the inter-core distance L using the machining state data PD when the thread grinding machine 10 grinds the threaded portion W1 of the worm shaft W, and based on the estimated inter-core distance L, The machining conditions and machine settings of the screw grinding machine 10, that is, the correction value C for correcting the command value of the operation command data (NC program) are determined. The correction value determination device 40 includes a model storage unit 71, a processing state data acquisition unit 72, a core-to-core distance prediction unit 73 as a processing accuracy prediction unit, a correction value calculation unit 74 as a correction value determination unit, and an output unit 75.

The model storage unit 71 stores the trained model generated by the model generation unit 63. The processing state data acquisition unit 72 acquires the processing state data PD (temperature data TD0 and temperature data TD1) from the data collection device 20. Here, the processing state data acquisition unit 72 performs the same processing as the processing state data acquisition unit 61a of the learning data set acquisition unit 61.

In this example, the processing state data acquisition unit 72 will be described as a separate element from the processing state data acquisition unit 61a of the learning data set acquisition unit 61. However, the processing state data acquisition unit 61a of the learning data set acquisition unit 61 can also be used together with the processing state data acquisition unit 72. That is, the function of the element 61a in the learning processing device 30 is also used as a part of the function of the correction value determining device 40.

The inter-core distance prediction unit 73 as the processing accuracy prediction unit acquires the trained model stored in the model storage unit 71. Further, the inter-core distance prediction unit 73 acquires the processing state data PD (temperature data TD0 and temperature data TD1) acquired by the processing state data acquisition unit 72. As a result, the inter-core distance prediction unit 73 is based on the trained model and the processing state data PD, that is, the amount of change (feature amount) which is the difference (= TD1-TD0) between the temperature data TD1 and the temperature data TD0. Then, the inter-core distance L is predicted.

The correction value calculation unit 74 calculates a correction value C for correcting the command value of the operation command data (NC program) for operating the screw grinding machine 10 based on the core distance L predicted by the core distance prediction unit 73. do. For example, when the inter-core distance L predicted by the inter-core distance predicting unit 73 is larger than the specified value L0, the amount of grinding of the threaded portion W1 of the worm shaft W by the grindstone 18 of the thread grinding machine 10 is small. Therefore, the correction value calculation unit 74 may, for example, bring the grindstone 18 (more specifically, the grindstone stand 17) closer to the worm shaft W by correcting the command value for further moving the grindstone 18 (more specifically, the grindstone stand 17) in the X-axis direction, or the grindstone wheel. The correction value C is calculated so as to increase the rotation speed of 18 or decrease the amount of coolant, that is, the temperature of the grinding point.

In this example, when the grinding amount is small, the correction value calculation unit 74 sets a command value in the operation command data (NC program) that specifies, for example, the rotation position of the motor 15a for moving the grindstone 18 (grindstone base 17). The correction value C for increasing (or decreasing in some cases) is calculated. Alternatively, when the grinding amount is small, the correction value calculation unit 74 increases (in some cases, decreases) the command value for designating the rotation speed of the motor 17a for rotating the grindstone 18 in the operation command data (NC program). The correction value C for this is calculated. Further, when the grinding amount is small, the correction value calculation unit 74 sets the correction value C for increasing (or decreasing in some cases) the command value for specifying the amount of coolant in the operation command data (NC program). calculate.

The output unit 75 outputs the correction value C calculated (determined) by the correction value calculation unit 74 to the control device S of the screw grinding machine 10. As a result, the control device S automatically corrects the NC program by inputting the correction value C output from the correction value determination device 40. Then, the control device S grinds the threaded portion W1 of the worm shaft W by controlling each of the motors 12a, 13a, 15a, 16a, 17a, the coolant device 19, and the like based on the modified NC program.

Further, the output unit 75 outputs the correction value C calculated (determined) by the correction value calculation unit 74 to the display device 50. Thereby, for example, the operator of the screw grinding machine 10 can operate the control device S according to the correction value C displayed on the display device 50 to correct the operation command data (NC program). Therefore, the control device S can grind the threaded portion W1 of the worm shaft W by controlling the motors 12a, 13a, 15a, 16a, 17a, the coolant device 19, and the like based on the modified NC program. can.

As can be understood from the above explanation, according to the machining system 1, it is observed when the thread grinding machine 10 which is a machine tool grinds the threaded portion W1 (meshing structure) of the worm shaft W as a workpiece. It is possible to generate a trained model showing the correlation between the machining state data PD and the inter-core distance data LD, which is the machining accuracy data representing the machining accuracy of the machined screw portion W1 (worm shaft W). Then, using the trained model and the machining state data PD, the inter-core distance L, which is the machining accuracy of the machined thread portion W1 (worm shaft W), is predicted, and the screw is screwed based on the predicted inter-core distance L. The correction value C for correcting the command value for operating the grinding machine 10 can be determined.

As a result, the thread grinding machine 10 can predict the machining accuracy of the threaded portion W1 (worm shaft W) while machining the threaded portion W1 (worm shaft W). Therefore, an inspection device is used for each machined worm shaft W. It is not necessary to inspect the processing accuracy using 2. Therefore, the productivity (machining efficiency) of the worm shaft W can be improved. Further, since the correction value C is determined based on the inter-core distance L predicted using the generated trained model, the variation of the inter-core distance L due to the dependence on the operator's experience (intuition and tips) can be obtained. It can be suppressed, and as a result, the inter-core distance L, that is, uniform processing accuracy can be achieved.

(5. Others)
In this example described above, the trained model generated by the model generation unit 63 of the learning processing device 30 is stored in the model storage unit 71 of the correction value determination device 40. By the way, the trained model is generated by performing machine learning using the learning data set stored in the learning data set storage unit 62. Therefore, as the number of training data sets increases, in other words, as learning by machine learning progresses, it is possible to generate a trained model with higher accuracy. Further, in the screw grinding machine 10 which is a machine tool, as the number of machining of the worm shaft W increases, for example, the grindstone 18 is worn, and the moving mechanism for moving the movable table 12 and the moving table 15 is mechanically deteriorated. Alternatively, the bearings of the headstock 13, the tailstock 14, and the grindstone 17 are mechanically deteriorated. Therefore, the model generation unit 63 can sequentially update the trained model.

Further, in the above-mentioned example, the model generation unit 63 and the inter-core distance prediction unit 73 use the difference between the temperature data TD0 and the temperature data TD1 included in the machining state data PD, that is, the amount of change as the feature amount. However, the feature amount is not limited to the change amount exemplified in this example. For example, the maximum value, the minimum value, the average value, and the statistical calculation value (dispersion value, standard deviation value, etc.) of the processing state data PD for a predetermined period can be used as the feature amount.

Alternatively, for example, when the processing state data PD contains a plurality of types of sampling data, the relationship between the plurality of types of sampling data for a predetermined period is expressed by an approximate relational expression (linear relationship or non-linear relationship) in the approximate relational expression. It is also possible to use a differential value, an extreme value, or the like as a feature amount. When the approximate relational expression is expressed as a non-linear relation, a more accurate feature amount can be obtained, so that the accuracy of the generated trained model and, by extension, the prediction accuracy of the inter-core distance L are improved, and as a result, more. An appropriate correction value C can be determined.

Further, in the above-mentioned example, the machining system 1 has a screw grinding machine 10 which is a grinding machine as a machine tool. As described above, the machine tool is not limited to the grinding machine, and a gear processing device (skiving processing device) can also be used. When the machining system 1 has a gear machining device (skiving machining device) as a machine tool, for example, a workpiece having a meshing structure (a gear having a gear structure) and a machining tool having a meshed structure as a reference member (skiving). The distance between the cores with the cutter) can be predicted. Then, based on the predicted inter-core distance, it is possible to determine a correction value for correcting and correcting the operation command data (NC program) for operating the gear processing device. Therefore, even in this case, the same effect as that of this example described above can be obtained.

Further, in the above-mentioned example, the model generation unit 63 and the inter-core distance prediction unit 73 obtain the feature amount based on the processing state data PD. However, the model generation unit 63 and the inter-core distance prediction unit 73 can generate a trained model or generate an inter-core distance L by directly using the machining state data PD without obtaining the feature amount, if necessary. It is also possible to make predictions. In this case as well, the same effect as that of this example described above can be expected.

Further, in the above-mentioned example, the coolant which is the second actual measurement data acquired by the sensor 23 of the data acquisition device 20 as the processing state data PD (that is, the feature amount) used by the learning processing device 30 and the correction value determining device 40. The temperature data related to the temperature is used. However, the feature amount is not limited to the temperature data related to the coolant temperature, for example, the drive current (grindstone shaft current) supplied to the motor 17a of the grindstone stand 17 that can be collected by the sensor 23, the outside temperature, and the diameter of the grindstone wheel 18. Etc. can also be used. In this case, as the processing state data PD (feature amount), for example, the average value, the maximum value, the skewness, etc. calculated by performing statistical calculation on each value obtained by the sensor 23, or the measured value and the average value. The difference value with and the like can be used.

1 ... Machining system, 2 ... Inspection equipment, 10 ... Screw grinder (machine tool), 11 ... Bed, 12 ... Movable table, 13 ... Headstock, 14 ... Trampling table, 15 ... Moving table, 16 ... Rotating table , 17 ... Grinding platform, 18 ... Grinding wheel, 19 ... Coolant device, 20 ... Data collection device (observation device), 21-23 ... Sensor, 24 ... Storage unit, 30 ... Learning processing device (learned model generation device), 31 ... Processor, 32 ... Storage device, 33 ... Interface, 40 ... Correction value determination device, 41 ... Processor, 42 ... Storage device, 43 ... Interface, 50 ... Display device, 61 ... Learning data set acquisition unit, 61a ... Processing State data acquisition unit, 61b ... Core distance data acquisition unit, 62 ... Learning data set storage unit, 63 ... Model generation unit, 71 ... Model storage unit, 72 ... Processing state data acquisition unit, 73 ... Core distance prediction unit (Machining accuracy prediction unit), 74 ... Correction value calculation unit (correction value determination unit), 75 ... Output unit, PD ... Machining state data, LD ... Core-to-core distance data (machining accuracy data), L ... Core-to-core distance, W ... Worm shaft (machine tool), W1 ... Threaded part (meshing structure), Wm ... Master worm gear (reference member), S ... Control device

Claims (15)

A machine tool that processes a workpiece with a meshing structure,
It has an observation device that observes the machining state of the workpiece by the machine tool and collects machining state data representing the machining state.
The machining state data collected by the observation device is used as an explanatory variable, the machining accuracy data related to the machining accuracy of the meshing structure machined by the machine tool is used as an objective variable, and the learning including the explanatory variable and the objective variable is used. A trained model storage unit that stores the trained model generated by performing machine learning using the data set for
A machining accuracy prediction unit that predicts the machining accuracy using the trained model and the machining state data,
A correction value determination unit that calculates and determines a correction value for correcting a command value for operating the machine tool based on the processing accuracy predicted by the processing accuracy prediction unit.
An output unit that outputs the correction value to the machine tool,
Equipped with a processing system. The processing state data is
Claimed data including actual operation data representing an observable operating state of the machine tool and actual measurement data of the machine tool and the work piece that can be observed by the machine tool processing the meshing structure. Item 1. The processing system according to Item 1. The processing accuracy data is
The processing system according to claim 1 or 2, which is data measured in a state where a reference member having a meshed structure that meshes with the meshing structure is meshed with the meshing structure. The processing accuracy data is
It is the core-to-core distance data representing the inter-core distance between the axis of the workpiece and the axis of the reference member in a state where the meshing structure of the workpiece and the meshed structure of the reference member are in mesh with each other. The processing system according to claim 3. The machining system according to claim 3, wherein the meshing structure is a screw structure or a gear structure machined by the machine tool. The machining state data collected by the observation device is used as the explanatory variable, the machining accuracy data relating to the machining accuracy of the meshing structure machined by the machine tool is used as the objective variable, and the explanatory variable and the objective variable are used. The processing system according to any one of claims 1-5, comprising a trained model generation unit that generates the trained model by performing machine learning using the including training data set. The trained model generation unit
The processing system according to claim 6, wherein the feature amount extracted from the processing state data is used as the explanatory variable. The feature amount is
The machining state data collected by the observation device in connection with the machining of the meshing structure by the machine tool this time, and the machining state data collected by the observation device in the previous machining of the meshing structure by the machine tool. The processing system according to claim 7, which is a change amount representing the difference between. The processing system according to any one of claims 1-8, wherein the processing by the machine tool is cutting or grinding. The processing system according to any one of claims 1-9, wherein the workpiece is a worm shaft mounted on a vehicle. The machining state data representing the machining state of the workpiece by the machine tool for machining the workpiece having the meshing structure is used as an explanatory variable, and the machining accuracy data relating to the machining accuracy of the meshing structure machined by the machine tool is used as the objective variable. A trained model generation device including a trained model generation unit that generates a trained model by performing machine learning using the training data set including the explanatory variables and the objective variable. A learning data set acquisition unit that acquires the processing state data and the processing accuracy data as a learning data set, and
The trained model generation device according to claim 11, further comprising a learning data set storage unit that stores the learning data set acquired by the learning data set acquisition unit. The trained model generation unit
The trained model generation device according to claim 11 or 12, wherein the feature amount extracted from the processing state data is used as the explanatory variable. The feature amount is
13. The amount of change according to claim 13, which is a change amount representing the difference between the machining state data in the current machining of the meshing structure by the machine tool and the machining state data in the previous machining of the meshing structure by the machine tool. Trained model generator. The trained model generator according to any one of claims 11-14, wherein the meshing structure is a screw structure or a gear structure machined by the machine tool.

Images