JP7792233B2 - Surface condition estimation system - Google Patents

Description

The present invention relates to a surface condition estimation system.

Grinding is performed, for example, by bringing a rotating grinding wheel into contact with a rotating workpiece. When grinding a workpiece using a rotating tool, if chatter vibration occurs, the machining accuracy of the ground surface may decrease or excessive load may be applied to the grinding wheel. For this reason, Patent Document 1 discloses technology for generating a trained model that can acquire the grinding quality of a workpiece.

Japanese Patent Application Laid-Open No. 2020-23040

When generating a trained model through machine learning, the settings of the explanatory variables and target variables in the training dataset used for machine learning are important. Workpieces that are machined by grinding include those formed into a cylindrical shape centered on the center of rotation, and those with a grinding surface on the periphery, such as cams and eccentric parts, that have non-circular parts that are not circular relative to the center of rotation.

A grinding wheel is equipped with many abrasive grains, each with a different protruding amount and angle. Therefore, when grinding the peripheral surface of a workpiece such as a cylindrical or non-circular part, the abrasive grains of the grinding wheel come into contact with the workpiece differently in both the circumferential and axial directions.

As a result, when chatter vibration occurs on the ground surface, the unevenness of the ground surface around the workpiece's periphery varies depending on the position on the workpiece. Furthermore, when setting explanatory variables and objective variables in machine learning, it may not be possible to obtain highly accurate estimation results unless consideration is given to the fact that the unevenness varies depending on the position on the workpiece.

The present invention was made in consideration of these issues, and aims to provide a surface condition estimation system that generates a highly accurate trained model and evaluates the surface condition of a workpiece during grinding processing.

One aspect of the present invention is a grinding machine that grinds a peripheral surface of a workpiece with a grinding wheel;
an observation device that observes vibration data generated in association with grinding of the peripheral surface of the workpiece by the grinding wheel;
an evaluation device that evaluates the surface condition of the ground surface of the peripheral surface of the workpiece based on the vibration data observed by the observation device;
A surface condition estimation system comprising:
The evaluation device
a trained model storage unit that stores a trained model generated by performing machine learning using the vibration data observed by the observation device and peripheral surface irregularity height data for each of three or more axial positions on the ground surface of the peripheral surface of the workpiece as a training data set, feature quantities extracted from the vibration data as explanatory variables, and statistical values of the peripheral surface irregularity height data at three or more positions as objective variables;
an estimation unit that estimates the statistical value of the grinding surface using the trained model stored in the trained model storage unit and the vibration data observed by the observation device;
an evaluation unit that evaluates a surface condition of the ground surface based on the statistical value of the ground surface estimated by the estimation unit;
Equipped with
The feature quantity is in a surface condition estimation system, and is data extracted from vibration acceleration data or vibration displacement data as the vibration data, from a frequency band determined based on the number of waviness in the peripheral surface unevenness height data and the rotational speed of the workpiece.

The surface condition estimation system targets cases where the peripheral surface of a workpiece is ground using a grinding wheel. The estimation calculation device that makes up this surface condition estimation system uses a trained model generated through machine learning to evaluate the surface condition of the ground peripheral surface of the workpiece.

In particular, when generating a trained model, the estimation calculation device uses the statistical values of the peripheral surface unevenness height data at each of three or more axial positions on the ground surface of the peripheral surface of the workpiece as the objective variable. In other words, rather than using peripheral surface unevenness height data at only one axial position, it uses peripheral surface unevenness height data at each of three or more axial positions.

When grinding the peripheral surface of a workpiece, the condition of the circumferential unevenness varies depending on the axial position, with some axial positions having a high circumferential unevenness height and others having a low circumferential unevenness height. Therefore, by using peripheral unevenness height data for three or more axial positions as described above, the effects of variation due to axial position can be reduced.

As described above, the estimation calculation device uses the statistical value of the circumferential surface unevenness height data at three or more axial positions as the dependent variable. In other words, the dependent variable is a value that takes into account the circumferential surface unevenness height data at three or more axial positions. Therefore, the dependent variable is a value that takes into account the variation due to axial position, rather than relying solely on the state of the circumferential surface unevenness height data at one specific axial position. In other words, the generated trained model is a model that represents the relationship between the feature values of the vibration data as explanatory variables and the statistical values of the circumferential surface unevenness height data at three or more axial positions as dependent variables.

The estimation unit that constitutes the estimation calculation device then estimates the statistical values of the ground surface using the trained model generated as described above and the vibration data observed by the observation device. In other words, the estimated statistical values of the ground surface correspond to the statistical values of the peripheral surface irregularity height data at three or more axial positions. Therefore, the estimated statistical values can be values that accurately represent the surface condition of the ground surface of the peripheral surface of the workpiece.

Furthermore, the evaluation unit that constitutes the estimation calculation device evaluates the surface condition of the ground surface based on the estimated statistical values of the ground surface. Therefore, the surface condition of the ground surface can be evaluated with high accuracy.

As described above, the above aspect provides a surface condition estimation system that generates a highly accurate trained model and evaluates the surface condition of a workpiece during grinding.

FIG. 1 is a diagram for explaining the configuration of a surface state estimation system. FIG. 2 is a diagram for explaining the configuration of the grinding machine of FIG. 1 . FIG. 2 is a diagram for explaining the observation device of FIG. 1 . FIG. 2 is a diagram for explaining the configuration of the measurement device of FIG. 1 . FIG. 5 is a diagram for explaining measurement positions by the measurement device of FIG. 4 . (a) shows the distance from the center of rotation of the workpiece relative to the angle of the workpiece at a non-circular portion, which is the ground surface of the workpiece. The dashed line shows the target shape (ideal shape) of the non-circular portion, and the solid line shows the actual shape when chatter occurs. (b) shows error data (surface irregularity height data) measured by a measuring device. FIG. 2 is a block diagram showing the configuration of an evaluation device.

(Embodiment)
1. Overview of the Configuration of the Surface Condition Estimation System 1 An overview of the configuration of the surface condition estimation system 1 of this embodiment will be described below with reference to Fig. 1. As shown in Fig. 1, the surface condition estimation system 1 includes a grinding machine 2, an observation device 3, at least one evaluation device 4, and a measurement device 5. The surface condition estimation system 1 evaluates the surface condition of the ground surface of the workpiece W ground by the grinding machine 2.

In this embodiment, the surface condition estimation system 1 is configured to include multiple grinding machines 2, but it can also be configured to include only one grinding machine 2. The grinding machine 2 rotates the workpiece W and rotates the grinding wheel, bringing the grinding wheel T into contact with the circumferential surface of the workpiece W, thereby grinding the circumferential surface of the workpiece W. In particular, this embodiment illustrates a case in which the workpiece W, which is the object to be ground by the grinding machine 2, has a non-circular portion W1 that is not circular with respect to the center of rotation. The non-circular portion W1 is, for example, a cam portion of a camshaft or a crank pin of a crankshaft. The grinding machine 2 grinds the circumferential surface of the non-circular portion W1. However, the workpiece W can also be a cylindrical shaft, and the grinding machine 2 may grind the cylindrical portion of the workpiece W. The workpiece W can also be the cylindrical portion of a camshaft (cam journal portion) or the cylindrical portion of a crankshaft (crank journal portion).

The grinding machine 2 includes a grinding machine body 11, which is the structural part that performs the grinding process on the workpiece W described above, a control device 12 for controlling the drive device that constitutes the grinding machine body 11, and an operation panel 13. The operation panel 13 functions as both an input device and a display device.

The observation device 3 is provided on the grinding machine 2 and observes vibration data Sv generated as the grinding wheel T grinds the non-circular portion W1 of the workpiece W. More specifically, the observation device 3 is provided on a structure that constitutes the grinding machine 2 and observes the vibrations of that structure. The observation device 3 obtains, for example, vibration acceleration data or vibration displacement data as the vibration data Sv obtained through observation.

The evaluation device 4 evaluates the surface condition of the ground surface of the workpiece W based on the vibration data Sv observed by the observation device 3. In particular, the evaluation device 4 evaluates the presence or absence of chatter during grinding and the degree of chatter as the surface condition of the workpiece W, as an example.

In the evaluation device 4 of this embodiment, an example is given in which the learning calculation device 40 that executes the learning phase and the estimation calculation device 50 that executes the estimation phase are independent devices. The learning calculation device 40 may also have server functionality and is communicatively connected to the estimation calculation device 50 provided on each of the multiple grinding machines 2. Furthermore, the estimation calculation device 50 is provided one-to-one with each of the multiple grinding machines 2, functions as an edge computer, and enables high-speed calculation processing.

However, the evaluation device 4 can also be configured as a single calculation device that executes the learning phase and the estimation phase. In this case, the learning calculation device 40 and the estimation calculation device 50 are integrated. The evaluation device 4 can also be an embedded system in the grinding machine 2, or an embedded system in a production line consisting of multiple grinding machines 2.

The learning calculation device 40 acquires the vibration data Sv observed by the observation device 3 and the peripheral surface irregularity height data Sd1, Sd2, and Sd3 obtained by the measurement device 5 as a training data set. Then, in the learning phase, the learning calculation device 40 generates a trained model TM by performing machine learning using the acquired training data set.

The estimation calculation device 50 stores the learned model TM generated by the learning calculation device 40. Then, in the estimation phase, the estimation calculation device 50 evaluates the surface condition of the ground surface of the workpiece W using the learned model TM and the vibration data Sv observed by the observation device 3.

The measuring device 5 is an external measuring device separate from the grinding machine 2. The measuring device 5 measures the surface shape of the non-circular portion W1 of the ground workpiece W. In particular, the measuring device 5 measures the peripheral surface irregularity height of the non-circular portion W1 and obtains peripheral surface irregularity height data Sd1, Sd2, and Sd3. The measuring device 5 outputs the peripheral surface irregularity height data Sd1, Sd2, and Sd3 obtained by the measurement to the learning calculation device 40 of the evaluation device 4.

2. Configuration of Grinding Machine 2 The configuration of the grinding machine 2 will be described with reference to Fig. 2. Examples of the grinding machine 2 include a cam grinding machine and a cylindrical grinding machine. In addition, in this embodiment, the workpiece W is a camshaft having a cam portion as the non-circular portion W1.

The grinding machine 2 comprises a grinding machine main body 11, a control device 12 that controls the grinding machine main body 11, and an operation panel 13 that constitutes an input device and display device. In this embodiment, the grinding machine main body 11 is a table traverse type grinding machine as an example. However, a wheelhead traverse type grinding machine can also be used as the grinding machine main body 11.

The grinding machine body 11 includes a bed 21, a grinding wheel head 22, a workpiece table 23, a headstock 24, a tailstock 25, and a rest device 26. While this embodiment exemplifies a configuration in which the grinding machine body 11 includes a rest device 26, it is also possible to configure the grinding machine body 11 without the rest device 26.

In the grinding machine body 11, the wheel head 22, workpiece table 23, headstock 24, and tailstock 25 are arranged on the bed 21. A wheel head guide surface 21a extending in the X-axis direction is provided on the bed 21. The wheel head 22 is supported on the wheel head guide surface 21a so that it can move in the X-axis direction. The wheel head 22 moves in the X-axis direction by driving an X-axis motor 21b provided on the bed 21.

The grinding wheel head 22 supports the grinding wheel T so that it can rotate around an axis parallel to the Z axis. The grinding wheel T is driven to rotate by a grinding wheel rotation motor 22a provided on the grinding wheel head 22. The grinding wheel T moves toward or away from the workpiece W as the grinding wheel head 22 moves in the X axis direction. In this embodiment, the width of the grinding wheel T is set larger than the width of one non-circular portion W1 of the workpiece W to be ground. Therefore, one non-circular portion W1 is ground by plunge grinding in one location.

Also, a table guide surface 21c extending in the Z-axis direction is provided on the bed 21 at a position spaced apart in the X-axis direction from the grinding wheel head guide surface 21a. A workpiece table 23 is supported on the table guide surface 21c so that it can move in the Z-axis direction. The workpiece table 23 moves in the Z-axis direction by being driven by a Z-axis motor 21d provided on the bed 21.

A headstock 24 and a tailstock 25 are arranged on the workpiece table 23 so that they face each other in the Z-axis direction. The headstock 24 and the tailstock 25 rotatably support both ends of the workpiece W. The headstock 24 is also provided with a spindle motor 24a, which rotates the workpiece W when driven.

The rest device 26 is positioned on the bed 21 so that the workpiece W is sandwiched between it and the grinding wheel T. In other words, the rest device 26 is positioned on the opposite side of the grinding wheel head 22 in the X-axis direction, with the workpiece table 23 as the reference. The rest device 26 supports, for example, the back side of the workpiece W relative to the position being ground by the grinding wheel T, and also supports the lower side in the direction of gravity. In this embodiment, the rest device 26 supports the back side and lower surface of the grinding point of the non-circular portion W1 to be ground. In other words, the rest device 26 has the function of suppressing bending and deformation of the workpiece W during grinding. However, the rest device 26 may also support the back side and lower surface at the axial center of the workpiece W.

3. Description of Observation Device 3 The observation device 3 will be described with reference to Figures 2 and 3. As described above, the observation device 3 is provided on the grinding machine 2, and observes vibration data Sv generated in association with grinding of the non-circular portion W1 of the workpiece W by the grinding wheel T. The observation device 3 acquires the vibration data Sv consisting of time-series data.

The purpose of the observation device 3 is to observe vibrations occurring in the workpiece W, but because the workpiece W is rotating, it is not easy to detect the movement of the workpiece W. Therefore, rather than directly observing the workpiece W, the observation device 3 observes the vibrations of the components to which the vibrations of the workpiece W are transmitted.

In particular, it is preferable that the observation device 3 be less affected by noise vibrations other than those of the workpiece W. Therefore, it is preferable that the observation device 3 observes the vibration data Sv of the tailstock 25 or rest device 26, among the structures that make up the grinding machine 2.

The tailstock 25 and rest device 26 are units that make up the grinding machine 2 and do not have an actuator that drives them during grinding. In other words, the tailstock 25 and rest device 26 are not units that are driven by an actuator during grinding. Units that have actuators, such as the wheel head 22 and headstock 24, may generate vibrations due to the actuator being driven during grinding. However, because the tailstock 25 and rest device 26 do not have an actuator that drives them during grinding, noise vibrations other than those of the workpiece W are small during grinding, allowing the desired vibration data Sv to be observed with high accuracy.

The components that make up the tailstock 25 and rest device 26 support the workpiece W during grinding and are examples of units that do not move during grinding. In other words, by having the observation device 3 observe the vibration data Sv of these units, the observation device 3 can observe the desired vibration data Sv with high accuracy, with little noise vibration other than that of the workpiece W. The units that the observation device 3 observes the vibration data Sv from can also be units other than the tailstock 25 and rest device 26.

The observation device 3 can observe the vibration data Sv as acceleration data of the vibration of the tailstock 25 or rest device 26, which is the object being observed, or displacement data of the vibration of the object being observed.

When observing vibration acceleration data as the vibration data Sv, the observation device 3 can be placed directly on the tailstock 25, as shown in Figures 2 and 3. In this case, the observation device 3 observes the vibration acceleration data of the tailstock 25 as the vibration data Sv. Of course, the observation device 3 can also be placed directly on the rest device 26, and observe the vibration acceleration data of the rest device 26 as the vibration data Sv.

Furthermore, when observing vibration displacement data as vibration data Sv, the observation device 3 can be placed on the bed 21 to observe the vibration displacement data of the tailstock 25 or rest device 26, which is the object being observed, as vibration data Sv.

4. Description of Measuring Device 5 The configuration of the measuring device 5 will be described with reference to Figures 4 to 6. As shown in Figure 4, the measuring device 5 includes a measuring probe 31 that contacts the outer peripheral surface of the cam portion, which is the non-circular portion W1 of the workpiece W, and a finger 32 that supports the measuring probe 31. The measuring probe 31 is provided so as to contact the outer peripheral surface of the cam portion, which is the non-circular portion W1 of the workpiece W. The finger 32 is provided so as to be movable in the vertical direction. Furthermore, the measuring device 5 is supported by an axial movement device 33 that moves along the axial direction of the workpiece W (a direction parallel to the central axis C).

The measuring device 5 is positioned in the axial direction of the workpiece W at a position radially opposite the cam portion, which is the non-circular portion W1 of the workpiece W, and while rotating the workpiece W, controls the height so that the measuring probe 31 remains in contact with the outer peripheral surface of the cam portion, which is the non-circular portion W1. In this way, the measuring device 5 measures the height of the peripheral irregularities on the outer peripheral surface of the cam portion, which is the non-circular portion W1 of the workpiece W.

As shown in Figure 5, the measurement positions taken by the measuring device 5 are three axial positions: measurement position P1, which is the axial center position of the cam portion, which is the non-circular portion W1; and measurement positions P2 and P3, which are the axial end positions of the cam portion (above and below measurement position P1 in Figure 5). In other words, the measuring device 5 acquires peripheral surface irregularity height data Sd1 at measurement position P1, peripheral surface irregularity height data Sd2 at measurement position P2, and peripheral surface irregularity height data Sd3 at measurement position P3. However, the measurement positions taken by the measuring device 5 are not limited to three axial positions of the cam portion; three or more axial positions may be used. However, it is preferable that the measurement positions include the axial center position and both axial end positions.

The peripheral surface irregularity height data Sd1, Sd2, and Sd3 will be explained with reference to Figures 6(a) and 6(b). Figure 6(a) shows the distance from the central axis C of the cam portion, which is the non-circular portion W1 of the workpiece W, relative to the rotation angle of the workpiece W. The target shape of the cam portion, which is the non-circular portion W1, is shown by the dashed line in Figure 6(a). In other words, the base circle portion of the cam portion shows a constant value, and the lift portion of the cam portion shows a value according to the lift amount.

When chatter occurs during grinding, the surface of the non-circular portion W1, which is the ground surface, is formed with a periodic uneven shape. Therefore, as shown by the solid line in Figure 6(a), the actual shape of the non-circular portion W1 of the workpiece W is a shape in which periodic unevenness is added to the target shape (ideal shape).

The measuring device 5 can measure error data relative to the target shape of the non-circular portion W1. Therefore, the measuring device 5 measures error data from the target shape, using the target shape of the cam portion, which is the non-circular portion W1 shown by the dashed line in Figure 6(a), as a reference. The measurement data measured by the measuring device 5 becomes the peripheral surface unevenness height data Sd1, Sd2, Sd3 as shown in Figure 6(b). Note that Figure 6(b) shows the peripheral surface unevenness height data Sd1 at measurement position P1, and the peripheral surface unevenness height data Sd2, Sd3 at the other measurement positions P2, P3 have different amplitudes and phases, but have similar waviness waveforms.

Here, it is known that the number of undulation peaks indicated by the peripheral surface irregularity height data Sd1, Sd2, and Sd3 is a value determined by the rotation period of the grinding wheel T, the rotation period of the workpiece W, the outer diameter of the grinding wheel T, and the outer diameter of the non-circular portion W1 of the workpiece W. However, the amplitudes and phases of the peripheral surface irregularity height data Sd1, Sd2, and Sd3 are different.

5. Functional Block Configuration of Evaluation Device 4 The functional block configuration of the evaluation device 4 that constitutes the surface condition estimation system 1 will be described with reference to FIG. 7. As described above, the surface condition estimation system 1 includes the evaluation device 4. As shown in FIG. 7, the evaluation device 4 includes a learning calculation device 40 that executes the learning phase, and an estimation calculation device 50 that executes the estimation phase. Both the learning calculation device 40 and the estimation calculation device 50 include a processor that executes calculation processing, a storage device that stores data, and an interface that performs input/output with external devices.

The learning calculation device 40 generates a learned model TM for estimating an index value for the surface unevenness, including the waviness, of the non-circular portion W1, based on the vibration data Sv observed by the observation device 3 and the peripheral unevenness height data Sd1, Sd2, and Sd3 of the non-circular portion W1 measured by the measurement device 5. The learning calculation device 40 includes a training dataset acquisition unit 41, a training dataset storage unit 42, and a model generation unit 43.

The training dataset acquisition unit 41 acquires a training dataset for machine learning. The training dataset acquisition unit 41 includes a vibration data acquisition unit 41a, a feature extraction unit 41b, a height data acquisition unit 41c, and a statistical value calculation unit 41d.

The vibration data acquisition unit 41a acquires vibration data Sv observed by the observation device 3. The vibration data Sv is time-series data obtained during grinding of the non-circular portion W1.

The feature extraction unit 41b extracts feature A from the vibration acceleration data or displacement data as the vibration data Sv acquired by the vibration data acquisition unit 41a. For example, the feature extraction unit 41b sets data extracted from a specific frequency band in the vibration acceleration data or vibration displacement data as feature A. More specifically, the feature extraction unit 41b sets the amplitude of the acceleration data or displacement data in a specific frequency band as feature A. Feature A may be the maximum amplitude or the average amplitude. The feature extraction unit 41b may also extract multiple feature A from the acceleration data or displacement data.

Feature A is data (such as maximum amplitude or average amplitude) extracted from the vibration displacement data or vibration acceleration data, from a frequency band determined based on the number of waviness in the peripheral surface unevenness height data Sd1, Sd2, Sd3 and the rotational speed of the workpiece W. The frequency band to be extracted may be set, for example, using an integer multiple of the rotational frequency component of the grinding wheel T. The rotational frequency component of the grinding wheel T is a frequency component consisting of the rotational frequency of the grinding wheel T and its harmonics. Feature A extracted by the feature extraction unit 41b then becomes one of the training data sets.

The height data acquisition unit 41c acquires multiple pieces of peripheral surface irregularity height data Sd1, Sd2, and Sd3 measured by the measurement device 5 at measurement positions P1, P2, and P3, respectively. The multiple pieces of peripheral surface irregularity height data Sd1, Sd2, and Sd3 are error data relative to the target shape of the non-circular portion W1, as shown in Figure 6(b).

The statistical value calculation unit 41d calculates a statistical value SD for the multiple peripheral surface irregularity height data Sd1, Sd2, and Sd3 acquired by the height data acquisition unit 41c. The statistical value SD may be, for example, the maximum value, average value, first quartile, third quartile, variance, or standard deviation. The calculated statistical value SD becomes one of the training data sets. For example, the statistical value SD is the maximum value of the three peripheral surface irregularity height data Sd1, Sd2, and Sd3.

The training dataset storage unit 42 associates the feature A of the vibration data Sv acquired by the training dataset acquisition unit 41 with the statistical values SD for the multiple peripheral surface irregularity height data Sd1, Sd2, and Sd3, and stores them as a training dataset.

The model generation unit 43 performs machine learning using the training dataset stored in the training dataset storage unit 42. Specifically, the model generation unit 43 uses the feature amount A of the vibration data Sv as an explanatory variable and the statistical value SD for the multiple peripheral surface unevenness height data Sd1, Sd2, Sd3 as a target variable, and performs machine learning to generate a trained model TM that represents the correlation between the feature amount A and the statistical value SD.

If the feature value A of the vibration data Sv is the maximum amplitude or average amplitude, then if the degree of chatter that occurs during grinding is small, then the feature value A will be small. Also, if the statistical value SD is the maximum value, then if the degree of chatter that occurs during grinding is small, then the statistical value SD will also be small. Conversely, if the degree of chatter that occurs during grinding is large, then both the feature value A and the statistical value SD will be large.

In the above case, the feature A of the vibration data Sv and the statistical value SD for the multiple peripheral surface irregularity height data Sd1, Sd2, and Sd3 have a monotonically increasing relationship. For example, the trend between the feature A and the statistical value SD can be expressed by, for example, a linear approximation formula (first-order approximation formula) or a multidimensional approximation formula.

The model generation unit 43 therefore defines the relationship between the feature amount A and the statistical value SD as a trained model TM. Machine learning is performed to generate the trained model TM. That is, the model generation unit 43 performs machine learning using the feature amount A of the vibration data Sv obtained when grinding a large number of workpieces W and the statistical values SD for the multiple peripheral surface unevenness height data Sd1, Sd2, and Sd3. The model generation unit 43 then performs machine learning using the multiple feature amounts A and the multiple statistical values SD to generate a trained model TM that represents the relationship between the feature amount A and the statistical values SD.

The estimation calculation device 50 evaluates the surface condition of the ground surface of the non-circular portion W1 of the workpiece W using the learned model TM generated by the learning calculation device 40 and the vibration data Sv observed by the observation device 3 during estimation.

As shown in FIG. 7, the estimation calculation device 50 includes a model storage unit 51, an evaluation data acquisition unit 52, an estimation unit 53, an evaluation unit 54, and an output unit 55. The model storage unit 51 (trained model storage unit) stores the trained model TM generated by the model generation unit 43.

The evaluation data acquisition unit 52 acquires vibration data Sv observed by the observation device 3 during grinding of the workpiece W to be estimated. The evaluation data acquisition unit 52 includes a vibration data acquisition unit 52a and a feature extraction unit 52b. Here, the vibration data acquisition unit 52a and the feature extraction unit 52b of the evaluation data acquisition unit 52 perform the same processing as the vibration data acquisition unit 41a and the feature extraction unit 41b of the training dataset acquisition unit 41.

In this embodiment, the evaluation data acquisition unit 52 is described as a separate element from the vibration data acquisition unit 41a and feature extraction unit 41b of the training data set acquisition unit 41. However, the vibration data acquisition unit 41a and feature extraction unit 41b of the training data set acquisition unit 41 can also serve as the vibration data acquisition unit 52a and feature extraction unit 52b of the evaluation data acquisition unit 52. In other words, the functions of elements 41a and 41b in the learning calculation device 40 are shared with some of the functions of the estimation calculation device 50.

The estimation unit 53 acquires the trained model TM stored in the model storage unit 51. Furthermore, the estimation unit 53 acquires the feature quantity A of the vibration data Sv acquired by the evaluation data acquisition unit 52. Here, as described above, the trained model TM defines the relationship between the feature quantity A of the vibration data Sv and the statistical value SD for the multiple pieces of peripheral surface irregularity height data Sd1, Sd2, and Sd3.

Therefore, the estimation unit 53 estimates an estimated statistical value SS corresponding to the statistical value SD for the multiple peripheral surface irregularity height data Sd1, Sd2, and Sd3 based on the trained model TM and the feature A of the vibration data Sv.

The evaluation unit 54 evaluates the surface condition of the ground surface of the non-circular portion W1 based on the estimated statistical value SS of the ground surface of the non-circular portion W1 estimated by the estimation unit 53. In particular, the evaluation unit 54 evaluates the presence or absence of chatter or the degree of chatter on the ground surface of the non-circular portion W1 based on the estimated statistical value SS.

For example, the evaluation unit 54 determines whether the estimated statistical value SS is equal to or greater than a preset reference value. If the estimated statistical value SS is equal to or greater than the reference value, the evaluation unit 54 determines that chatter is occurring during grinding of the non-circular portion W1 of the workpiece W with the grinding wheel T. Alternatively, the evaluation unit 54 determines that the degree of chatter is large.

On the other hand, if the estimated statistical value SS is less than the reference value, the evaluation unit 54 determines that no chatter is occurring during grinding of the non-circular portion W1 of the workpiece W with the grinding wheel T. Alternatively, if the estimated statistical value SS is less than the reference value, the evaluation unit 54 determines that the degree of chatter is small. The evaluation unit 54 can also store multiple step-by-step reference values and evaluate the degree of chatter in steps by determining which step the estimated statistical value belongs to.

The output unit 55, for example, digitizes the evaluation result by the evaluation unit 54 and outputs the digitized evaluation result R to the operation panel 13. If the evaluation unit 54 determines that chatter has occurred or that the degree of chatter is large, the output unit 55 increases the value of the evaluation result R and outputs it to the operation panel 13. On the other hand, if the evaluation unit 54 determines that chatter has not occurred or that the degree of chatter is small, the output unit 55 decreases the value of the evaluation result R and outputs it to the operation panel 13.

The operation panel 13 displays the evaluation result R output by the evaluation unit 54 on a display screen that functions as a display device. By checking the display screen of the operation panel 13, the worker can determine whether chatter has occurred on the surface of the workpiece W that has been ground. Furthermore, the worker can determine the extent of chatter that has occurred on the surface of the workpiece W.

In particular, the estimation calculation device 50 functions as an edge computer for the grinding machine 2. Therefore, immediately after grinding the workpiece W on the grinding machine 2, the estimation calculation device 50 can perform high-speed calculations. For example, before grinding the next workpiece W begins, the estimation calculation device 50 can output the evaluation result R to the operation panel 13 and display the evaluation result R on the operation panel 13. In other words, the operator can check the evaluation result R for the workpiece W that has just been ground before starting grinding the next workpiece W.

Furthermore, the output unit 55 may be configured to output the evaluation result R by the evaluation unit 54 to the control device 12. For example, if the evaluation result R indicates that the degree of chatter is large, the control device 12 may control the grinding machine main body 11 to pause rather than start grinding the next workpiece W. In this case, the control device 12 may display an indication of a pause on the display screen of the operation panel 13.

6. Effects The surface condition estimation system 1 of the above embodiment is intended for a case where the peripheral surface of a workpiece W is ground using a grinding wheel T. The estimation calculation device 50 constituting the surface condition estimation system 1 evaluates the surface condition of the ground peripheral surface of the workpiece W using a trained model TM generated by machine learning.

In particular, when generating the trained model TM, the estimation calculation device 50 uses the statistical value SD for the peripheral surface unevenness height data Sd1, Sd2, Sd3 at three or more axial positions on the ground peripheral surface of the workpiece W as the objective variable. In other words, rather than using only the peripheral surface unevenness height data at one axial position, the estimation calculation device 50 uses the peripheral surface unevenness height data Sd1, Sd2, Sd3 at three or more axial positions.

In particular, when grinding the peripheral surface of the workpiece W, the state of circumferential irregularities varies depending on the axial position, with some axial positions having a large circumferential irregularity height and others having a small circumferential irregularity height. Therefore, by using the peripheral irregularity height data Sd1, Sd2, and Sd3 for three or more axial positions as described above, the effects of variation due to axial position can be suppressed.

As described above, the estimation calculation device 50 uses the statistical value SD of the circumferential surface irregularity height data Sd1, Sd2, and Sd3 at three or more axial positions as the objective variable. In other words, the objective variable is a value that takes into account the circumferential surface irregularity height data Sd1, Sd2, and Sd3 at three or more axial positions. Therefore, the objective variable is a value that takes into account variation due to axial position, rather than relying solely on the state of the circumferential surface irregularity height data at one specific axial position. In other words, the generated trained model TM is a model that represents the relationship between the feature A of the vibration data Sv as the explanatory variable and the statistical value SD of the circumferential surface irregularity height data Sd1, Sd2, and Sd3 at three or more axial positions as the objective variable.

The estimation unit 53 constituting the estimation calculation device 50 then uses the trained model TM generated as described above and the vibration data Sv observed by the observation device 3 to estimate the statistical value (estimated statistical value SS) of the ground peripheral surface of the workpiece W. In other words, the estimated statistical value SS corresponds to the statistical value SD of the peripheral surface irregularity height data Sd1, Sd2, Sd3 at three or more axial positions. Therefore, the estimated statistical value SS can be a value that accurately represents the surface condition of the ground peripheral surface of the workpiece W.

Furthermore, the evaluation unit 54 constituting the estimation calculation device 50 evaluates the surface condition of the ground peripheral surface of the workpiece W based on the estimated statistical value SS. Therefore, the surface condition of the ground peripheral surface of the workpiece W can be evaluated with high accuracy.

In addition, the surface condition estimation system 1 is configured so that the three or more axial positions (measurement positions P1, P2, P3) in the peripheral surface irregularity height data Sd1, Sd2, Sd3 include the axial center position and both axial end positions of the peripheral surface of the workpiece W. By including these three positions, it is possible to obtain a statistical value SD that takes variability into account.

Furthermore, feature A of the vibration data Sv is data extracted from the vibration acceleration data or vibration displacement data as the vibration data Sv, from a frequency band determined based on the number of waviness in the peripheral surface unevenness height data Sd1, Sd2, Sd3 and the rotational speed of the workpiece W. Chatter occurring on the peripheral surface of the workpiece W can be predicted in advance using the above elements (number of waviness and rotational speed of the workpiece W). Therefore, the desired feature A can be extracted by limiting the extraction of feature A to a frequency band specified using the above elements.

The observation device 3 is also configured to observe vibration data of the tailstock 25 or rest device 26. This makes it possible to obtain vibration data Sv related to chatter occurring on the peripheral surface of the workpiece W with high accuracy. As a result, the evaluation device 4 can evaluate the presence or absence of chatter and the extent of chatter with high accuracy. The observation device 3 may also observe vibration data of a unit that supports the workpiece W and is stationary during grinding. For example, the components that make up the tailstock 25 or rest device 26 can be given as examples of such units.

As described above, the grinding machine 2 can be used to grind a non-circular portion W1 of a workpiece W, which has a non-circular portion W1 that is non-circular with respect to the center of rotation, using a grinding wheel T. In this case, the evaluation device 4 evaluates the surface condition of the ground surface of the non-circular portion W1 of the workpiece W based on the vibration data Sv observed by the observation device 3.

In particular, when grinding the non-circular portion W1 of the workpiece W, the height of the position where the grinding wheel T comes into contact with the non-circular portion W1 of the workpiece W changes. Furthermore, when grinding the non-circular portion W1 of the workpiece W, the grinding is sometimes performed while changing the rotational speed of the workpiece W. As a result, the peripheral speed of the grinding position on the workpiece W varies depending on the circumferential position of the workpiece W. For example, one cause of chatter on the ground surface is the change in peripheral speed.

Therefore, when grinding the non-circular portion W1 of the workpiece W, the setting of the explanatory variables and objective variables of the trained model TM is more important than when grinding a cylindrical portion. When grinding the non-circular portion W1 of the workpiece W, as described above, the estimation calculation device 50 generates the trained model TM using the statistical values SD for the peripheral surface irregularity height data Sd1, Sd2, and Sd3 at three or more axial positions on the ground surface of the non-circular portion W1 of the workpiece W as objective variables. In other words, the generated trained model TM is a model that represents the relationship between the feature A of the vibration data Sv as the explanatory variable and the statistical values SD for the peripheral surface irregularity height data Sd1, Sd2, and Sd3 at three or more axial positions as objective variables. Therefore, the surface condition of the ground surface of the non-circular portion W1 of the workpiece W can be evaluated with high accuracy.

(others)
In the above-described embodiment, the objective variable in the machine learning is the statistical value SD of the peripheral surface irregularity height data Sd1, Sd2, Sd3 that represent the outer peripheral surface shape of the cam portion measured by the measurement device 5. Alternatively, or in addition, the objective variable can be, for example, a current value (e.g., a built-in motor current value) generated by displacing the finger 32 or the measurement probe 31 of the measurement device 5, which can be used as the peripheral surface irregularity height data Sd1, Sd2, Sd3, and the statistical value of the peripheral surface irregularity height data Sd1, Sd2, Sd3.

1 Surface condition estimation system 2 Grinding machine 3 Observation device 4 Evaluation device 51 Model storage unit (trained model storage unit)
53 Estimation unit 54 Evaluation unit W Workpiece W1 Non-circular part of workpiece T Grinding wheel Sv Vibration data A Feature amount of vibration data Sd1, Sd2, Sd3 Circumferential surface irregularity height data SD Statistical value SS Estimated statistical value TM Trained model

Claims (13)

a grinding machine that grinds the peripheral surface of a workpiece with a grinding wheel;
an observation device that observes vibration data generated in association with grinding of the peripheral surface of the workpiece by the grinding wheel;
an evaluation device that evaluates the surface condition of the ground surface of the peripheral surface of the workpiece based on the vibration data observed by the observation device;
A surface condition estimation system comprising:
The evaluation device
a trained model storage unit that stores a trained model generated by performing machine learning using the vibration data observed by the observation device and peripheral surface irregularity height data for each of three or more axial positions on the ground surface of the peripheral surface of the workpiece as a training data set, feature quantities extracted from the vibration data as explanatory variables, and statistical values of the peripheral surface irregularity height data at three or more positions as objective variables;
an estimation unit that estimates the statistical value of the grinding surface using the trained model stored in the trained model storage unit and the vibration data observed by the observation device;
an evaluation unit that evaluates a surface condition of the ground surface based on the statistical value of the ground surface estimated by the estimation unit;
Equipped with
the feature amount is data extracted from a frequency band of vibration acceleration data or vibration displacement data as the vibration data, the frequency band being determined based on the number of waviness in the peripheral surface unevenness height data and the rotational speed of the workpiece. 2. The surface condition estimation system according to claim 1, wherein the observation device observes vibration data of a member that supports the workpiece during grinding and to which vibrations of the workpiece are transmitted, in order to observe vibrations generated in the workpiece. The surface condition estimation system of claim 1 or 2, wherein the statistical value is one of the maximum value, average value, first quartile, third quartile, variance, and standard deviation of the peripheral surface irregularity height data at three or more locations. A surface condition estimation system according to any one of claims 1 to 3, wherein the peripheral surface irregularity height data is error data relative to a target shape of the peripheral surface of the workpiece. The surface condition estimation system according to any one of claims 1 to 4, wherein the trained model is an approximate expression that represents a correlation between amplitude in vibration acceleration data or displacement data as the vibration data and the statistical value for the peripheral surface irregularity height data at three or more locations. The surface condition estimation system according to any one of claims 1 to 5 , wherein the three or more axial positions in the peripheral surface irregularity height data include an axial center position and axial end positions of the peripheral surface of the workpiece. 7. The surface condition estimation system according to claim 1 , wherein the observation device observes vibration data of a unit that supports the workpiece and is stationary during grinding. The surface condition estimation system according to any one of claims 1 to 7 , wherein the observation device observes vibration data of a tailstock or a rest device. The surface condition estimating system according to any one of claims 1 to 8 , wherein the vibration data is time-series data. The surface condition estimating system according to any one of claims 1 to 9 , wherein the evaluation unit evaluates the presence or absence of chatter that occurs during grinding of the workpiece by the grinding wheel, or the degree of chatter that occurs. The surface condition estimation system according to any one of claims 1 to 10, wherein the evaluation device further comprises a trained model generation unit that generates the trained model by performing machine learning using the feature amounts extracted from the vibration data as the explanatory variables and the statistical values of the peripheral surface unevenness height data at three or more locations as the objective variables. the grinding machine uses the grinding wheel to grind the non-circular portion of the workpiece, the non-circular portion having a non-circular shape with respect to a rotation center;
The surface condition estimation system according to any one of claims 1 to 11 , wherein the evaluation device evaluates the surface condition of the ground surface of the non-circular portion of the workpiece based on the vibration data observed by the observation device. The explanatory variables are only feature quantities extracted from the vibration data,
The surface condition estimating system according to claim 1 , wherein the estimating unit estimates the statistical value using only the trained model and the vibration data.