JP7294448B2 - Grinding system, correction amount estimation device, computer program and grinding method - Google Patents

Description

The present invention relates to a grinding system, a correction amount estimation device, a computer program, and a grinding method.

In Patent Document 1, polishing condition data indicating polishing processing conditions is observed as a state variable representing the current state of the environment, and based on the state variable, wear of the polishing tool with respect to the polishing processing conditions is measured. Abrasive tool wear amount predictor is described that includes a machine learning device that learns or predicts using a learning model that models the amount.

JP 2019-139755 A

Grinding of a workpiece by a grinding machine is accomplished by repeatedly pressing and grinding the grinding tool against the workpiece until the desired machining result is obtained. As machining progresses, not only the workpiece but also the grinding tool wears out at the same time, so in order to ensure machining accuracy, the workpiece after machining is measured to determine whether or not remachining is necessary and the conditions. is necessary, and it is difficult to improve productivity.

The invention described in Patent Document 1 uses a machine learning device to predict the wear of the abrasive tool, and estimates the wear of the abrasive tool based on various processing conditions. It is difficult to accurately estimate the amount of tool wear according to the invention because it strongly depends on the state of the workpiece.

The present invention has been made in view of such circumstances, and an object of the present invention is to acquire a correction amount in the pressing direction that reflects the state of a workpiece in a grinding system.

The invention disclosed in the present application for solving the above problems has various aspects, and the outlines of typical aspects are as follows.

A grinding system according to one aspect of the present invention includes a reaction force related information acquisition unit that acquires reaction force related information related to the pressing reaction force of a grinding tool during grinding, and a difference calculation unit for calculating the difference between the acquired two pieces of the reaction force-related information; a correction amount estimation unit for estimating the correction amount of the pressing direction of the grinding tool based on the difference; a target position; a machining control unit for controlling the position of the grinding tool during machining based on the quantity.

Moreover, in the grinding system according to another aspect of the present invention, the correction amount estimator may obtain the correction amount by inputting the difference into a machine learning model.

Further, in a grinding system according to another aspect of the present invention, a cumulative correction amount calculation unit that calculates a cumulative correction amount, which is a cumulative value of the correction amounts for the grinding wheel of the grinding tool, is provided, and the machining control unit may control the position of the grinding tool during machining based on the target position and the cumulative correction amount when machining different workpieces.

Further, in the grinding system according to another aspect of the present invention, the machining target portion of the workpiece may be divided, and the correction amount estimator may estimate the correction amount for each division.

Further, in the grinding system according to another aspect of the present invention, the grinding tool may be supported by an electric motor control support mechanism, and the reaction force-related information may be external force torque of the electric motor in the pressing direction of the grinding tool. .

Further, in the grinding system according to another aspect of the present invention, the estimated number of iterative machining is estimated by inputting the reaction force-related information into a second machine learning model and estimating the estimated number of iterative machining at the same location of the workpiece. may have a part.

Further, the correction amount estimating device according to one aspect of the present invention provides two pieces of reaction force-related information relating to the pressing reaction force of the grinding tool during the grinding process, which are acquired in different repetition times for the repetitive machining at the same location of the workpiece. and a correction amount estimation unit for estimating the correction amount of the pressing direction of the grinding tool based on the difference.

In addition, a computer program according to an aspect of the present invention provides a computer with two sets of reaction force-related information relating to the pressing reaction force of a grinding tool during grinding, which are obtained in different repetition times for repeated processing at the same location on a workpiece. It functions as a correction amount estimating device having a difference calculating section for calculating the difference and a correction amount estimating section for estimating the correction amount of the pressing direction of the grinding tool based on the difference.

Further, a grinding method according to one aspect of the present invention acquires reaction force-related information about the pressing reaction force of the grinding tool during grinding, and obtains two pieces of information obtained in different repetition times for repeated machining at the same place of the workpiece. A difference in the reaction force-related information is calculated, a correction amount of the pressing direction of the grinding tool is estimated based on the difference, and the position of the grinding tool during machining is controlled based on the target position and the correction amount. do.

1 is a schematic diagram showing the overall configuration of an example grinding system according to a preferred embodiment of the present invention; FIG. It is a block diagram showing the functional configuration of the grinding system. It is a figure which shows the torque command value which is the reaction force related information obtained in the n-th grinding process and the (n+1)-th grinding process, and the example of the difference. It is a figure which shows the torque command value which is the reaction force related information obtained in the n-th grinding process and the (n+1)-th grinding process, and the example of the difference. FIG. 2 is a block diagram showing the functional configuration of a machine learning model learning device using a grinding system; FIG. 2 is a diagram showing an outline of a configuration for estimating the estimated number of iterations in a grinding system; It is a figure which shows the outline|summary of a system configuration. It is a figure which shows the grinding|polishing object workpiece|work to verify. It is a figure which shows the robot configuration of a verification system. It is a table|surface which shows the spec of a disc grinder and a whetstone. It is a figure which shows the outline|summary of polishing operation. FIG. 10 is a diagram showing the appearance of a weld bead after polishing every two times; FIG. 10 is a diagram showing an example of data of disturbance torque _τd in the 2nd, 4th, 6th, 8th, and 10th polishing operations; FIG. 10 is a table showing results of estimation using evaluation data from an NN model generated by learning; FIG. FIG. 10 is a table showing estimation results when polishing operation of the same grindstone is continuously performed for 11th to 15th works, and 10 polishing operations are performed. FIG. FIG. 2 is a block diagram showing the functional configuration of a grinding system including a repetition number estimator;

FIG. 1 is a schematic diagram showing the overall configuration of an example grinding system 100 according to a preferred embodiment of the present invention.

The grinding system 100 includes a grinding tool 1 and a support mechanism 3 that movably supports the grinding tool 1 relative to a workpiece 2 and enables grinding of the workpiece 2 . FIG. 1 also shows a support table 4 that supports the workpiece 2 being processed and a controller 5 that controls the support mechanism 3 .

Here, the grinding tool 1 is a disk grinder in this embodiment, and has a grinder body 6 and a grinding wheel 7 attached to the tip thereof. A general multi-axis industrial robot is used as the support mechanism 3, and the grinding tool 1 is attached as its end effector. Controller 5 is a robot controller.

In addition, since the workpiece 2 here is a member obtained by bead welding a steel plate, and the weld bead 8 rises in a ridged shape, the grinding system 100 removes the unnecessary rise of the weld bead 8 from the surface by grinding. It is exemplified that the purpose of the processing is to remove it so that it becomes smooth.

It should be noted that the grinding system 100 shown in FIG. 1 is merely an example, and the gist of the present invention is not affected even if the specific configuration of each part is replaced with another configuration. For example, although the grinding tool 1 is shown in FIG. 1 as using a commonly available disc grinder, the grinding tool 1 may be designed and manufactured specifically for the grinding system 100 . Further, although the support mechanism 3 is shown here as using a vertical articulated robot, other than this, various mechanisms such as a SCARA robot, an orthogonal robot, and a gantry mechanism may be used to support the grinding tool 1. Not only the side but also the support base 4 that supports the workpiece 2 may be movable.

Further, the controller 5 may be not only a robot controller but also any device for automatically controlling the support mechanism 3, such as a servo controller, a PLC (Programmable Logic Controller), or a PC. may be a combination of a plurality of devices. The workpiece 2 is not limited as long as it is an object that requires grinding. It may require various grinding processes such as external grinding and cutting.

FIG. 2 is a block diagram showing the functional configuration of the grinding system 100. As shown in FIG. The controller 5 is provided with a machining control unit 50 , which controls the position of the support mechanism 3 by program control, thereby moving the grinding tool 1 supported by the support mechanism 3 to the workpiece 2 as desired. to move along the trajectory of In the present embodiment, the support mechanism 3 is an electric motor control support mechanism driven by an electric motor, so the machining control unit 50 includes a motor controller for driving the electric motor. The motor is servo-controlled.

The support mechanism 3 is provided with a reaction force-related information acquisition unit 30, which acquires reaction force-related information relating to the pressing reaction force of the grinding tool 1 during grinding. Here, the reaction force-related information means information that can indicate the pressing reaction force of the grinding tool 1 directly or by some conversion, and may be the pressing reaction force itself or other information. good too. The pressing reaction force itself can be obtained, for example, by attaching the grinding tool 1 to the support mechanism 3 via a load cell and directly measuring it. As other information, in the grinding system 100 according to the present embodiment, the reaction force torque acting on the electric motor of the support mechanism 3 in the pressing direction of the grinding tool 1, that is, the torque command value is used as the reaction force related information. there is Employing motor control information, particularly current command values and torque command values, as the reaction force-related information is economical because the support mechanism 3 does not require a special additional configuration for acquiring the reaction force-related information. . Henceforth, in this embodiment, the torque command value of the electric motor of the support mechanism 3 shall be used as reaction force related information.

In addition to the processing control unit 50 , the controller 5 is provided with a correction unit 51 , and outputs the cumulative correction amount to the processing control unit 50 in this embodiment. Here, before the detailed description of the configuration of the correction unit 51, the meaning of the correction by the correction unit 51 will be explained.

In the grinding process using the grinding tool 1, wear of the grinding wheel 7 is inevitable as the process progresses. Since the wear of the grinding wheel 7 means the retreat of the grinding wheel 7, when the grinding tool 1 is subjected to the position control by the support mechanism 3, the grinding wheel 7 wears and the grinding wheel 7 moves backward. As a result, the workpiece 2 cannot be machined into a desired shape. Therefore, in order to perform accurate grinding, the amount of wear of the grinding wheel 7 must be estimated by some method, and the position control by the support mechanism 3 must be corrected accordingly.

Measuring the amount of wear of the grinding wheel 7 with high accuracy each time processing complicates the grinding system 100 such as the installation and adjustment of the measuring device, increases the cost, and makes maintenance difficult. On the other hand, even if an attempt is made to estimate the amount of wear of the grinding wheel 7, the amount of wear of the grinding wheel 7 depends greatly on the amount of grinding of the workpiece 2 in the grinding process. It is difficult to estimate this only by the conditions on the grinding system 100 side.

Therefore, in the grinding system 100 according to the present embodiment, the correction unit 51 is provided with the difference calculation unit 52, and the difference between the two pieces of reaction force-related information obtained in different repetition times for the repetitive machining at the same location of the workpiece 2 is calculated. calculate.

Generally, in grinding, it is rare that the desired shape of the workpiece 2 is obtained by grinding once, so the same place is ground a plurality of times along the same locus until the desired shape is obtained.

FIG. 3 shows, in the grinding system 100, in repetitive machining at the same location of the workpiece 2, the torque command value, which is reaction force-related information obtained in the n-th grinding process and the (n+1)-th grinding process, and its FIG. 5 is a diagram showing an example of difference; In the upper graph of FIG. 3, the n-th torque command value is indicated by a solid line, and the (n+1)-th torque command value is indicated by a broken line. A solid line indicates the difference obtained by subtracting the torque command value.

The torque command value reflects the pressing reaction force when the grinding tool 1 is pressed against the workpiece 2 during grinding. Therefore, the waveform of the torque command value reflects the shape of the portion of the working portion 2 that requires machining. Then, as the grinding process progresses, the portion that needs to be processed is cut and its size becomes smaller, so the torque command value in the later iteration becomes smaller than the torque command value in the previous iteration. There is a tendency.

The difference between the torque command values obtained in different iterations reflects the difference between the shape of the portion requiring machining in the previous iteration and the shape of the portion requiring machining in the later iteration. It is nothing but a reflection of the amount of change in the shape of the part requiring machining between the previous iteration and the subsequent iteration. In the example shown in FIG. 3, the difference shown in the lower graph is associated with the amount of change in the shape of the portion requiring machining caused by the n-th machining, that is, the amount of grinding by the grinding tool 1 .

Therefore, in the grinding system 100 according to the present embodiment, the correction amount in the pressing direction of the grinding system 100 is further estimated by the correction amount estimation section 53 provided in the correction section 51 based on this difference.

Any algorithm for estimating the correction amount may be used as long as the correction amount can be reasonably estimated based on the difference. Further, this correction amount may be any correction amount necessary for highly accurate grinding, including displacement of the machining point of the grinding tool 1 due to wear of the grinding wheel 7 . In the present embodiment, the movement distance of the machining point due to wear of the grinding wheel 7 (hereinafter simply referred to as "abrasion amount") is used as the correction amount. A correction amount may be an amount including a mechanical change amount such as deflection compensation of No. 3.

Then, as already mentioned, it is thought that there is a correlation between the difference shown in the lower graph of FIG. 3 and the necessary correction amount. It's not always easy to do. Therefore, the correction amount estimating unit of the grinding system 100 according to the present embodiment uses a machine learning model in which the relationship between this difference and the amount of wear is learned in advance, and inputs the difference to the machine learning model. , the correction amount is obtained as an output.

Although the architecture of this machine learning model is not particularly limited, it is preferably based on RNN (Recurrent Neural Network) because the difference, which is the input value, is time-series data. Learning of the machine learning model will be described later.

The correction amount obtained in this manner indicates the amount of wear of the grinding wheel 7 caused by the grinding process between two different iterations of obtaining the torque command value. Since the wear of the grinding wheel 7 accumulates each time processing is performed, the final correction to the position control of the support mechanism 3 is based on the accumulated value of this correction amount.

Therefore, the cumulative correction amount calculation unit 54 provided in the correction unit 51 calculates and holds the cumulative correction amount by accumulating each time the correction amount is estimated by the correction amount estimation unit 53, and the cumulative correction amount is calculated and stored. The amount is output to the processing control section 50 .

Since this cumulative correction amount is an amount corresponding to the individual wear amount of the grinding wheel 7 of the grinding tool 1, it is held as the amount for the grinding wheel 7 in use, and when the grinding wheel 7 is replaced, the Value is reset.

The machining control unit 10 receives the cumulative correction amount obtained as described above from the correction unit 51, and adjusts the grinding tool based on not only the pre-programmed target position but also the cumulative correction amount during the grinding process. 1 position. More specifically, the cumulative correction amount is added to the coordinates of the target position of the grinding tool 1 in the pressing direction. Therefore, it can be said that the machining control unit 50 controls the position of the grinding tool 1 during machining based on the target position and the correction amount estimated by the correction amount estimation unit 53 .

Also, since the cumulative correction amount is the amount for the grinding wheel 7 in use, it is natural that even if a workpiece 2 has been completely ground and replaced with a different workpiece 2, As long as the grinding wheel 7 is not replaced, it will continue to be used. Therefore, when machining a different workpiece 2, the machining control unit 50 controls the position of the grinding tool 1 during machining based on the target position and the cumulative correction amount.

In addition, in the controller 5 of the grinding system 100 described above, when the correction unit 51 is configured as an independent device, this can be considered as a correction amount estimation device. In that case, the correction amount estimation device is an independent device having the difference calculation unit 52, the correction amount estimation unit 53, and the cumulative correction amount calculation unit 54 described above.

The correction amount estimation device may be designed as a dedicated device, or may be implemented using a general computer. FIG. 5 is a diagram showing the configuration of a general computer 11 that can be used as a correction amount estimation device. The computer 11 includes a CPU (Central Processing Unit) 11a which is a processor, a RAM (Random Access Memory) 11b which is a memory, an external storage device 11c, a GC (Graphics Controller) 11d, input devices 11e and 11f (Input/Output) 306. They are connected by a data bus 11g so that electric signals can be exchanged with each other. It should be noted that the hardware configuration of the computer 11 shown here is merely an example, and may have a configuration other than this.

The external storage device 11c is a device capable of statically recording information, such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive). A signal from the GC 11d is output to a monitor 11h, such as a CRT (Cathode Ray Tube) or a so-called flat panel display, on which a user visually recognizes an image, and displayed as an image. The input device 11e is one or a plurality of devices such as a keyboard, mouse, touch panel, etc. for the user to input information, and the I/O 11f is one or a plurality of interfaces for the computer 11 to exchange information with external devices. is. The I/O 11f may include various ports for wired connection and a controller for wireless connection.

A computer program for causing the computer 11 to function as the machine learning data generation device 1 and the machine learning device 2 is stored in the external storage device 11c, read out to the RAM 11b as necessary, and executed by the CPU 11a. That is, the RAM 11b stores codes for realizing various functions shown as the correction unit 51 in FIG. 2 by being executed by the CPU 11a. Even if such a computer program is provided by being recorded on an appropriate computer-readable information recording medium such as an appropriate optical disk, magneto-optical disk, flash memory, etc., it can be accessed via an external information communication line such as the Internet via the I/O 11f. may be provided.

FIG. 5 shows the n-th time of repetitive machining at the same location of the workpiece 2 when the length of the object to be ground is long or the shape of the part to be machined changes greatly during machining. It is a figure which shows the example of the torque command value which is the reaction force related information obtained in the grinding process, the (n+1)th grinding process, and the difference.

In this case, the torque command value obtained in the same repetition is divided into a plurality of sections, and in the example shown in FIG. The difference between the torque command value and the (n+1)th torque command value is also divided into similar three sections. Then, the correction amount estimating section 53 may receive the difference for each section, and obtain the correction amount for each section as a result.

The final cumulative correction amount is the cumulative value of the correction amount for each segment thus obtained. In FIG. 4, each segment is made with respect to time, but since the grinding process is performed by the grinding tool 1 moving over the workpiece 2 with the passage of time, this segment is a portion of the workpiece 2 to be machined. It is nothing but something that distinguishes between In this way, by properly dividing the workpiece 2 into sections to be processed and estimating the correction amount for each section, the correction amount can be obtained with high accuracy even when the processing conditions change greatly during the grinding process. be done.

For example, if the machining conditions are significantly different, such as the size of the weld bead 8 is significantly different between the front half and the rear half of the part to be machined of the workpiece 2, rather than estimating the correction amount collectively, This is because it is possible to estimate the correction amount more accurately by estimating the correction amount separately for the front half and the rear half, which are considered to have approximately the same processing conditions. The parts to be machined of the workpiece 2 may be divided at regular intervals, or may be divided at arbitrary intervals according to the properties of the workpiece 2 .

As described above, the learning of the machine learning model used in the correction amount estimation unit 53 is any method that can learn the machine learning model so that the difference is input and the correction amount is output. is not particularly limited. Therefore, an example of such a learning method is described below.

FIG. 6 is a block diagram showing the functional configuration of the machine learning model learning device 101 using a grinding machine. In addition, in the same figure, the same code|symbol is attached|subjected to the structure equivalent to the grinding system 100 shown in FIG. 2, and the overlapping description shall be abbreviate|omitted.

The machine learning device 101, similar to the grinding system 100, has the grinding tool 1 attached thereto, the support mechanism 3 having the reaction force related information acquisition unit 30, and the controller 5 having the processing control unit 50 for controlling the support mechanism 3. In addition, a learning device 9 and a cumulative correction amount measuring device 10 are provided.

In this example, the cumulative correction amount measuring device 10 is a measuring device that measures the amount of wear of the grinding wheel 7 of the grinding tool 1 attached to the support mechanism 3 . The amount of wear can be measured by measuring the surface position of the grinding wheel 7 with an arbitrary sensor, such as a non-contact laser sensor or an arbitrary contact sensor, or by driving the support mechanism 3 to move the grinding tool 1 in advance. This may be done by pressing against a set reference surface and detecting the coordinates of the support mechanism 3 at the time when the grinding wheel 7 and the reference surface come into contact with each other. That is, the cumulative correction amount measuring device 10 may be an independent device, or may be configured using one or more existing devices of the machine learning device 101 .

Since what is measured at this time is the amount of wear of the grinding wheel 7, that is, the cumulative value of wear due to the grinding process so far, in this example, the amount of cumulative correction is actually measured.

Therefore, when performing repetitive machining on an arbitrary workpiece 2, the reaction force-related information acquisition unit 30 acquires the actual measurement value of the reaction force-related information during machining at an arbitrary repetition time, and similarly, at the arbitrary repetition time , it is possible to obtain an actual measurement value of the cumulative correction amount.

The learning device 9 has a difference calculator 90 , a machine learning model 91 and a correction amount calculator 92 . The difference calculation unit 90 receives the reaction force-related information acquired by the reaction force-related information acquisition unit 30, here the torque command value, thereby calculating the difference between the two pieces of reaction force-related information acquired in different iterations. can do.

Further, the correction amount calculation unit 92 can actually calculate the correction amount caused by the processing between the iterations from the difference between the two actual measurement values of the cumulative correction amount obtained in different iterations.

Therefore, the learning device 9 learns the machine learning model 91 by repeatedly learning the machine learning model 91 using the difference obtained based on the actual measurement as input data and the correction amount obtained based on the actual measurement as correct data. can be made

The hardware that implements the learning device 9 is not particularly limited. A learner 9 may be implemented as part of the controller 5, or a general computer 11 shown in FIG. 4 may be used as the learner 9, for example. Further, in the example of the machine learning device 101 shown in FIG. 6, the learning device 9 is connected to the direct instruction mechanism 3 and the cumulative correction amount measuring device 10, and each time the support mechanism 3 is driven to perform the grinding process, machine learning is performed. Although the configuration in which the model 91 is learned has been shown, in addition to this, a large number of measured values of the reaction force related information and the cumulative correction amount are acquired in advance and accumulated, and later, the independently configured learning device 9 is provided. You may make it learn collectively using it.

Further, the grinding system 100 may have a configuration for estimating the estimated number of iterations, which is an estimated value of the number of iterations required to complete the machining when performing the iterative machining at the same location on the workpiece 2 . The configuration for estimating the estimated number of iterative machining operations in the grinding system 100 will be described below.

<System overview>
FIG. 7 shows an overview of the system configuration. This system uses the control torque data collected while the robot is performing polishing work to generate a model that estimates the state of the work target. A disk grinder is attached to the tip of the robot to perform grinding work. A damper mechanism is provided between the disc grinder and the hand to absorb external force to some extent even if they come into contact with each other. The robot controller (hereafter referred to as RC) collects work data using the MotoPlus (Note) app and saves it in a USB memory. Using the saved data, learning is performed with machine learning software in the PC. A neural network (hereinafter referred to as NN) consisting of four layers including an input layer and an output layer is used for the learning model. The learned NN model is downloaded to RC via USB memory. The RC inputs control torque data during work to this NN and estimates the work state based on the output result.
(Note) MotoPlus: A function to develop application software that runs inside the RC.

<Method for estimating working conditions>
In a general sensorless hand force estimation method, the hand force F is calculated using the Jacobian matrix J, where _τd is the disturbance torque, which is the difference between the control torque estimated from the motion trajectory and the actual control torque. , can be calculated by Equation 1.

手先力推定

hand force estimation

Since the state S of the contact work including the tool is a system in which F is input, if the system function is G, Equation 2 is obtained.

手先力を入力とした作業状態推定

Work state estimation with hand force as input

Here, it is replaced with a system function with the disturbance torque τ _d as an input, as shown in Equation 3.

外乱トルクを入力とした作業状態推定

Work state estimation with disturbance torque as input

<Technical Evaluation>
We will evaluate work state estimation using this technology. The work state to be estimated shall be the achieved state of polishing.

<Work for work>
A work to be polished to be verified is shown in FIG. Welding work is performed on an iron plate with a thickness of 4 mm, and a weld bead is arbitrarily attached, and polishing of this weld bead is the object. The weld bead has a length of about 155 mm and a height of about 2.5 mm.

<Verification system configuration>
Fig. 9 shows the robot configuration of the verification system. A 6-axis vertical multi-joint robot MOTOMAN-GP7 (hereinafter referred to as GP7) is used as the robot, and YRC1000 is used as the RC. An electric disk grinder G10SH5 manufactured by HiKOKI is attached to the tip of GP7. By attaching a whetstone to the disk grinder, the rotating surface can be pressed against the surface of the disk for grinding. A resinoid flexible wheel manufactured by HiKOKI is used as the grindstone. The welding bead to be polished is fixed on a table installed in front of the robot, and the polishing work is performed. Tables 1 and 2 in FIG. 10 show the specifications of the disk grinder and the grindstone, respectively.

<Polishing target>
FIG. 11 outlines the polishing operation. Grinding is performed by pulling the hand toward the front of the robot. This operation has been taught in advance, and is an operation using only three axes (second, third, and fifth axes) among the axes of the GP7.

We will actually polish and collect data. Ten polishing operations are performed on one weld bead. It is empirically known that when a new whetstone is used, polishing is approximately completed in ten polishing operations. Also, in order to obtain data using a new grindstone, the grindstone is replaced after polishing of 10 workpieces is completed. Fig. 12 shows the appearance of the weld bead after polishing every two times. The weld bead becomes flatter each time it is polished, and the polishing is achieved after 10 times of polishing.

<Evaluation results>
A total of 500 data were prepared by polishing 50 workpieces 10 times. Of these, a total of 50 data obtained by polishing the fifth workpiece after replacing with a new grindstone were used as evaluation data, and learning was performed using the remaining 450 data. FIG. 13 shows an example of data of the disturbance torque _τd in the 2nd, 4th, 6th, 8th and 10th polishing operations.

Table 3 in FIG. 14 shows the result of estimation using the evaluation data from the NN model generated by learning. This is a table called a mixing matrix that shows the number of estimated labels for each correct label (in this case, the number of polishings). The number of correctly estimated numbers for each label is indicated by the diagonal gray frame. As a result, 49 data out of 50 data can be estimated with the label of the correct answer or in the range of ±1 times, and it can be said that the estimation is performed with high accuracy.

The reason for the failure in the inference result is assumed to be as follows. The number of times of polishing was used as the label for estimation in this study, but in reality, the grinding wheel deteriorates with each polishing, and even with the same number of times of polishing, there was a difference in the quality of polishing, and the number of times of polishing was misidentified as higher or lower. Table 4 in Fig. 15 shows the estimated results when the polishing work with the same whetstone is continuously performed for 11 to 15 workpieces and the polishing operation is performed 10 times. It can be seen that it is estimated that the polishing achievement status of In other words, it can be determined that sufficient polishing has not been achieved due to deterioration of the grindstone.

That is, it can be said that the polishing state is estimated each time the polishing operation is performed once, and the polishing operation is stopped when the polishing operation is estimated to be the 10th polishing operation.

With the configuration described above, the estimated iteration number estimating unit 55 for estimating the estimated number of iterations can be configured. FIG. 16 is a block diagram showing the functional configuration of the grinding system 100 including the repetition number estimator 55. As shown in FIG. Since each configuration of the grinding system 100 according to this example is basically the same as that shown in FIG. 2, the same reference numerals are given to the same configurations, and redundant explanations will be omitted. The only difference from that shown in FIG.

In this example, the estimated repetition number estimating unit 55 estimates the estimated number of iterations of machining at the same location on the workpiece 2, so the obtained estimated number of iterations of machining is used to estimate the remaining time required for the current machining. Also, it can be used to determine the timing for replacing the grinding wheel 7 .

Claims (9)

a reaction force-related information acquisition unit that acquires reaction force-related information about the pressing reaction force of the grinding tool during grinding;
a difference calculation unit that acquires the difference between two pieces of the reaction force-related information acquired in different iterations for repetitive machining at the same location of the workpiece;
a correction amount estimating unit that estimates a correction amount of the pressing direction of the grinding tool based on the difference;
a machining control unit that controls the position of the grinding tool during machining based on the target position and the correction amount;
Grinding system with 2. The grinding system according to claim 1, wherein said correction amount estimator inputs said difference to a machine learning model to obtain said correction amount. a cumulative correction amount calculation unit that calculates a cumulative correction amount, which is a cumulative value of the correction amounts for the grinding wheel of the grinding tool;
3. The grinding system according to claim 1, wherein the machining control unit controls the position of the grinding tool during machining based on the target position and the cumulative correction amount when machining different workpieces. 4. The grinding system according to any one of claims 1 to 3, wherein the workpiece is divided into portions to be machined, and the correction amount estimator estimates the correction amount for each division. The grinding system according to any one of claims 1 to 4, wherein the grinding tool is supported by an electric motor control support mechanism, and the reaction force-related information is an external force torque of the electric motor in the pressing direction of the grinding tool. 6. The method according to any one of claims 1 to 5, further comprising an estimated number of repeated machining times estimating unit that inputs the reaction force-related information to a second machine learning model and estimates an estimated number of repeated machining times at the same location of the workpiece. grinding system. a difference calculation unit that acquires the difference between the reaction force-related information regarding the pressing reaction force of the grinding tool during the two grinding operations, which is acquired in different repetition times for the repetitive machining at the same location of the workpiece;
a correction amount estimating unit that estimates a correction amount of the pressing direction of the grinding tool based on the difference;
A correction amount estimating device having the computer,
a difference calculation unit that acquires the difference between the reaction force-related information regarding the pressing reaction force of the grinding tool during the two grinding operations, which is acquired in different repetition times for the repetitive machining at the same location of the workpiece;
a correction amount estimating unit that estimates a correction amount of the pressing direction of the grinding tool based on the difference;
A computer program for functioning as a correction amount estimation device having Acquire reaction force-related information about the pressing reaction force of the grinding tool during grinding,
Obtaining the difference between the two pieces of reaction force-related information obtained in different iterative times for repetitive machining at the same location of the workpiece,
estimating a correction amount of the pressing direction of the grinding tool based on the difference,
controlling the position of the grinding tool during machining based on the target position and the correction amount;
grinding method.

Images