JP2021148457A - Chatter prediction system - Google Patents

Abstract

To provide a chatter prediction system that can accurately predict a chatter vibration in polish processing.SOLUTION: A chatter prediction system 1 includes a learning processor 20 and a prediction operation device 30. The learning processor 20 acquires state data detected by a first detector 13 and extracts the feature amount of the acquired state data. The learning processor 20 acquires chatter-related data output from a second detector 14 (a chatter evaluation device 70) and generates a learned model by a mechanical learning using a training dataset of the feature amount and the chatter-related data. The prediction operation device 30 predicts an evaluation value for evaluating a chatter vibration and outputs the evaluation value by using the state data detected by the first detector 13 and the generated learned model.SELECTED DRAWING: Figure 4

Description

The present invention relates to a chatter prediction system.

In general, when a workpiece is ground, vibration caused by the operation of the grinding machine or wear of the grindstone is generated, so that chatter vibration (hereinafter, also simply referred to as “chatter”) may occur. be. If chatter occurs in the grinding process, it may cause deterioration of the surface texture of the workpiece, and it is extremely important to know whether or not chatter occurs. For this reason, the techniques disclosed in Patent Document 1 have been conventionally known.

Patent Document 1 describes the presence or absence of chatter vibration by comparing the magnitude of vibration in a certain frequency band and the degree of change in vibration within a certain period of time with a discrimination reference value for the vibration of a grinding machine during grinding. The technique for determining the frequency is disclosed. As a result, in particular, chatter vibration at the initial stage of generation is detected.

Japanese Unexamined Patent Publication No. 2000-23338

In the technique disclosed in Patent Document 1, the magnitude of chatter vibration is determined based on the analysis result of vibration data detected during grinding and a preset discrimination reference value. The magnitude of the chatter vibration generated when the grindstone grinds the workpiece changes depending on the condition of the ground surface of the workpiece with which the grindstone comes into contact. In other words, the state of the ground surface of the workpiece affects the operating state of the grindstone, so that the magnitude of chatter vibration changes. Therefore, in order to accurately predict the magnitude of chatter vibration, it is necessary to take into account the operating state of the grindstone when the grindstone grinds the workpiece.

An object of the present invention is to provide a chatter prediction system capable of accurately predicting chatter vibration in grinding.

The chatter prediction system detects and detects state data related to the grinding device in which the grindstone grinds the workpiece through multiple grinding processes controlled by the control device and the operating state of the grindstone that can be observed in the grinding process by the grinding device. The first detector that outputs the state data, the second detector that detects the chatter-related data related to the magnitude of the chatter vibration generated in the grinding process by the grinding device, and the second detector that outputs the chatter-related data. It is generated by performing machine learning using the training data set including the explanatory variables and the objective variables, with the feature quantity of the state data acquired from the device as the explanatory variable and the grinding-related data acquired from the second detector as the objective variable. Evaluation value prediction that predicts and outputs the evaluation value for evaluating the chatter vibration generated in the grinding process by the grinding device using the trained model storage unit that stores the trained model and the trained model and the state data. It has a part and.

According to this, evaluation for evaluating chatter vibration using a trained model showing the correlation between the feature amount of the state data regarding the operating state of the grindstone in grinding and the chatter-related data regarding the magnitude of chatter vibration. The value can be predicted and output. Therefore, the chatter vibration generated in the grinding process can be accurately predicted based on the predicted evaluation value.

It is a figure which shows the structure of the chatter prediction system. It is a figure which shows an example of a grinding apparatus. It is a figure which shows the functional block composition of an example of the 2nd detector. It is a figure which shows the functional block composition of the chatter prediction system. It is a figure which shows an example of the state data. It is a figure for demonstrating the feature amount extracted from the state data of FIG. It is a figure which shows the functional block composition of the chatter prediction system of the first alternative example. It is a figure which shows the functional block structure of the model update part of FIG.

(1. Grinding equipment to which the chatter prediction system is applied)
The chatter prediction system evaluates the state of chatter in a grinding device that grinds a workpiece through a plurality of grinding steps. As the grinding machine, a grinding machine having various configurations such as a cylindrical grinding machine, a cam grinding machine, and a surface grinding machine can be applied. The grinding process by the grinding device includes a rough grinding process, a fine grinding process, a fine grinding process, a spark-out process, and the like.

(2. Outline of the configuration of chatter prediction system 1)
The outline of the configuration of the chatter prediction system 1 will be described with reference to FIG. The chatter prediction system 1 includes at least one grinding device 10 and one arithmetic unit (20, 30). The grinding device 10 may be targeted at one unit, or may be targeted at a plurality of units as shown in FIG. In this example, the chatter prediction system 1 includes a case where a plurality of grinding devices 10 are provided as an example.

The grinding device 10 includes at least a first detector 13 that detects observable state data during grinding of the workpiece W. The arithmetic unit (20, 30) predicts the state of occurrence of chatter in the grinding process by the grinding device 10, that is, the chatter evaluation, by applying machine learning using the state data detected by the first detector 13. do.

In FIG. 1, the arithmetic unit (20, 30) is composed of the learning processing device 20 and the prediction arithmetic unit 30, and the learning processing unit 20 and the prediction arithmetic unit 30 are shown as independent configurations, but one device. It can also be. Further, a part or all of the arithmetic units (20, 30) can be incorporated into the grinding apparatus 10.

In this example, the case where the learning processing device 20 and the prediction calculation device 30 have independent configurations will be taken as an example. Further, the learning processing device 20 has a so-called server function, and is communicably connected to a plurality of grinding devices 10. On the other hand, the prediction calculation device 30 is provided on each grinding device 10 on a one-to-one basis and is communicably connected to each grinding device 10. That is, the plurality of prediction arithmetic units 30 function as so-called edge computers, and can realize high-speed arithmetic processing.

(3. Details of the configuration of chatter prediction system 1)
The configuration of the chatter prediction system 1 will be described in more detail with reference to FIG. The chatter prediction system 1 includes a plurality of grinding devices 10, one learning processing device 20 that functions as a part of an arithmetic unit, and a plurality of prediction calculation devices 30 that function as another part of the arithmetic unit.

Each of the grinding devices 10 includes a grinding machine 11 that grinds a workpiece W using a grindstone T, a control device 12 that controls the grinding machine 11, a first detector 13, a second detector 14, and so on. It mainly includes an interface 15. The control device 12 includes a CNC device, a PLC device, and the like, and controls a drive device and the like in the grinding machine 11. The interface 15 is a device that enables communication with the outside with the grinding machine 11, the control device 12, the first detector 13 and the second detector 14. The outside includes a learning processing device 20 and an inspection device 2 for inspecting the workpiece W ground by the grinding device 10.

The first detector 13 detects state data related to the observable operating state of the grinding machine 11 (working load, drive load of the drive device, etc.) on the grinding machine 11 during the grinding process of the workpiece W. The first detector 13 is, for example, a displacement sensor that detects relative position (or distance) data between the grindstone T and the workpiece W, a current sensor that detects drive current data of a motor as a drive device, and a current sensor during machining. A temperature sensor that detects the temperature data of the grindstone T and the workpiece W, a vibration sensor (acceleration sensor) that detects the vibration data of the constituent members of the grinding machine 11, a microphone that detects the sound data during processing, and the like. That is, the state data detected by the first detector 13 is, for example, position data, drive current data, temperature data, vibration data (acceleration data), sound data, and the like.

The second detector 14 detects chatter-related data related to the magnitude of observable chatter vibration in the grinding machine 11 during or after machining the workpiece W. The second detector 14 is, for example, a roughness meter that detects the surface roughness of the machined surface of the workpiece W, an acceleration sensor or a displacement sensor that detects acceleration data or displacement data caused by the surface roughness, and the like. That is, the chatter-related data detected by the second detector 14 is, for example, surface roughness (roughness data), acceleration data (vibration data), displacement data, etc., which are related to the magnitude of chatter vibration. In this example, the second detector 14 is provided integrally with the grinding device 10, but can also be provided separately from the grinding device 10.

Then, the second detector 14 calculates and outputs the chattering amount (hereinafter, referred to as “chattering evaluation value”) as an evaluation value, in particular, by statistical calculation. Here, the second detector 14 uses each state data (drive current data, vibration data, etc.) detected by the first detector 13 or detection data by an acceleration sensor or displacement sensor provided separately. As chatter-related data, it is also possible to output the chatter amount indicating the magnitude of the chatter vibration that occurs.

The inspection device 2 can detect and output the same data as the chatter-related data detected by the second detector 14. That is, the inspection device 2 can detect and output the amount of chatter that represents the magnitude of the chatter vibration that occurs in the grinding process.

The learning processing device 20 includes a processor 21, a storage device 22, an interface 23, and the like. Further, the learning processing device 20 has a server function, and is communicably connected to a plurality of grinding devices 10 (including, for example, grinding devices 10 installed in other factories separated from each other).

The learning processing device 20 performs machine learning based on the state data detected by the first detector 13, the second detector 14 and / and the chatter-related data output from the inspection device 2. Then, the learning processing device 20 generates a trained model for predicting the chatter evaluation value in the grinding process.

In particular, the learning processing device 20 has an amplitude of the grinding wheel rotation frequency extracted by FFT (Fast Fourier Transform) of the drive current data and the vibration data among the state data output from the first detector 13 in time series. Is the feature quantity. Then, the learning processing device 20 generates a trained model using the extracted feature amount and the chatter-related data (specifically, the chatter evaluation value).

Further, the learning processing device 20 divides the state data output in time series for each grinding process, extracts the feature amount, and generates a learned model. For example, the learning processing device 20 extracts a plurality of feature quantities for each grinding process and generates one trained model using the plurality of feature quantities for each grinding process.

Each prediction arithmetic unit 30 includes a processor 31, a storage device 32, an interface 33, and the like. Further, the prediction calculation device 30 is communicably connected to the learning processing device 20 as a server and the corresponding grinding device 10.

The prediction calculation device 30 is arranged at a position close to each grinding device 10, and functions as a so-called edge computer. The prediction arithmetic unit 30 uses the trained model generated by the learning processing unit 20 to grind based on the feature amount extracted from the state data detected by the first detector 13 during the machining of the workpiece W. Predict the chatter evaluation value in processing.

The chatter prediction system 1 further includes a common display device 40 and a plurality of individual display devices 50. However, the chatter prediction system 1 may be configured not to include the common display device 40, or may be configured not to include the individual display device 50. The common display device 40 is arranged corresponding to the learning processing device 20. Further, the individual display device 50 is arranged corresponding to each of the grinding devices 10.

(4. Example of grinding machine 11)
As the grinding machine 11 of this example, as shown in FIG. 2, a grindstone base traverse type cylindrical grinding machine 60 is taken as an example. As the grinding machine 11, a table traverse type can also be used. The workpiece W to be ground by the cylindrical grinding machine 60 has a peripheral surface around the axis.

The cylindrical grinding machine 60 is a machine for grinding a workpiece W. The cylindrical grinding machine 60 mainly includes a bed 61, a headstock 62, a tailstock 63, a traverse base 64, a grindstone base 65, a grindstone 66 (grindstone T), a sizing device 67, a grindstone correction device 68, and a coolant. The device 69 is provided.

The bed 61 is fixed on the installation surface. The headstock 62 is provided on the upper surface of the bed 61 on the front side in the X-axis direction (lower side in FIG. 2) and one end side in the Z-axis direction (left side in FIG. 2). The headstock 62 rotatably supports the workpiece W around the Z axis. The workpiece W is rotated by driving a motor 62a provided on the headstock 62. The tailstock 63 is located on the upper surface of the bed 61 at a position facing the headstock 62 in the Z-axis direction, that is, the front side in the X-axis direction (lower side in FIG. 2) and the other end side in the Z-axis direction (the lower side in FIG. 2). It is provided on the right side of FIG. 2). As a result, the geographic structure W is rotatably supported at both ends by the headstock 62 and the tailstock 63.

The traverse base 64 is provided on the upper surface of the bed 61 so as to be movable in the Z-axis direction. The traverse base 64 is moved by driving a motor 64a provided on the bed 61. The grindstone base 65 is provided on the upper surface of the traverse base 64 so as to be movable in the X-axis direction. The grindstone base 65 is moved by driving a motor 65a provided on the traverse base 64. The grindstone wheel 66 is rotatably supported by the grindstone base 65. The grindstone wheel 66 is rotated by driving a motor 66a provided on the grindstone base 65. The grindstone 66 is configured by fixing a plurality of abrasive grains with a bond material.

The sizing device 67 measures the dimension (diameter) of the workpiece W. The sizing device 67 is provided on the upper surface of the bed 61 so as to be movable in the Z-axis direction. The position of the sizing device 67 in the Z-axis direction is controlled by the feed mechanism 67a provided on the bed 61. The sizing device 67 includes a stylus 67b, which is a contact portion that comes into contact with the ground surface of the workpiece W. Further, in the sizing device 67, an acceleration sensor 67c is provided on an arm that supports the stylus 67b.

The acceleration sensor 67c outputs acceleration data representing the acceleration detected when the stylus 67b is in contact with the ground surface of the rotating workpiece W. Instead of using the acceleration sensor 67c, a displacement sensor that outputs displacement data representing the displacement (corresponding to the surface roughness) detected when the stylus 67b is in contact with the ground surface of the rotating workpiece W. It is also possible to use.

The grindstone correction device 68 corrects the shape of the grindstone 66. The grindstone correction device 68 is a device for performing the growing of the grindstone 66. The grindstone correction device 68 may be a device for dressing the grindstone 66 in addition to or in place of the trueing. Here, the trueing is a reshaping work, such as a work of forming the grindstone 66 according to the shape of the workpiece W when the grindstone 66 is worn by grinding, a work of removing the runout of the grindstone 66 due to uneven wear, and the like. Is.

Dressing is a work of sharpening (sharpening), adjusting the amount of protrusion of abrasive grains, and creating a cutting edge of abrasive grains. Dressing is the work of correcting fouling, clogging, spillage, etc., and is usually performed after truing.

The coolant device 69 supplies the coolant to the grinding point of the workpiece W by the grindstone 66. The coolant device 69 cools the recovered coolant to a predetermined temperature and supplies it to the grinding point again.

The control device 12 provided in the cylindrical grinding machine 60 is based on an NC program generated based on operation command data such as the shape of the workpiece W, the machining conditions, the shape of the grindstone 66, and the coolant supply timing information. Control the drive unit. That is, the control device 12 inputs the operation command data and generates an NC program based on the operation command data.

Then, the control device 12 grinds the workpiece W by controlling the motors 62a, 64a, 65a, 66a, the coolant device 69, and the like based on the NC program. In particular, the control device 12 grinds the workpiece W until it has a finished shape based on the diameter of the workpiece W measured by the sizing device 67. Further, the control device 12 corrects the grindstone 66 (trueing and dressing) by controlling the motors 64a, 65a, 66a, the grindstone repair device 68, and the like at the timing of correcting the grindstone 66. ..

Here, the control device 12 outputs the state data output from the first detector 13 in time series (continuously) to the learning processing device 20. Further, the control device 12 outputs the current grinding process, that is, the rough grinding process, the fine grinding process, the fine grinding process, and the spark-out process, which are output in time series based on the NC program generated according to the operation command data. The process information that identifies is linked to the state data.

(5. Example of the second detector 14)
In this example, the chatter evaluation device 70 in which the second detector 14 outputs the chatter evaluation value is taken as an example. The second detector 14 can also be configured by using a displacement sensor or an acceleration sensor to output displacement data or acceleration data (vibration data).

As shown in FIG. 3, the chatter evaluation device 70 mainly includes an actual data acquisition unit 71, a displacement conversion unit 72, a chatter evaluation value calculation unit 73, and an output unit 74. The actual data acquisition unit 71 acquires the acceleration data detected by the acceleration sensor 67c provided in the sizing device 67 in time series. Here, the actual data acquisition unit 71 acquires acceleration data detected at a plurality of different axial positions in the axial direction of the workpiece W.

In this case, the sizing device 67 is in a state where the contact position between the stylus 67b and the workpiece W is fixed in the axial direction, that is, the contact position moves around the outer circumference of the workpiece W in a circular shape. Get the acceleration data for the minute. After acquiring the acceleration data for one round of the outer circumference of the work W, the sizing device 67 moves to the next axial position in the axial direction, and this axial position, that is, the next contact position is again the outer circumference of the work W. Is moved in a circle to acquire acceleration data. By repeating this operation of the sizing device 67, the actual data acquisition unit 71 sequentially acquires acceleration data for one outer circumference detected at each of the plurality of axial positions (plural contact positions).

The displacement conversion unit 72 FFT (Fast Fourier Transform) the acceleration data acquired from the acceleration sensor 67c in time series, and obtains data on acceleration having a rotation frequency component (specific frequency component) corresponding to the rotation speed of the grindstone wheel 66. Extract. Then, the displacement conversion unit 72 reverse-FFTs the data regarding the acceleration having the extracted specific frequency component. As a result, the displacement conversion unit 72 converts the displacement of the stylus 67b of the sizing device 67 having a specific frequency component, that is, the displacement data (roughness data) relating to the unevenness (surface roughness) on the ground surface of the workpiece W. Convert. The specific frequency component is a rotation speed of the grindstone 66 and a frequency component that is an integral multiple of the rotation speed.

Here, the cylindrical grinding machine 60 grinds the workpiece W while rotating the grindstone 66. Therefore, the surface shape of the grindstone 66 is transferred to the grindstone 66 at each rotation cycle, that is, at each rotation speed, and appears on the ground surface of the workpiece W. Specifically, when there are large protruding abrasive grains on the surface of the grindstone 66, a recess is formed on the ground surface of the workpiece W in which the portion in contact with the abrasive grains is largely scraped off. In this case, the recesses formed in the workpiece W are formed at equal intervals in the rotation direction, and the spacing between the recesses in the circumferential direction of the workpiece W coincides with the rotation cycle (for each rotation number) of the grindstone 66. Therefore, by extracting the data related to the acceleration having the specific frequency component, the chatter vibration generated due to the surface state of the grindstone 66 and the unbalanced state of the grindstone 66 can be extracted.

The chatter evaluation value calculation unit 73 calculates the chatter evaluation value of the work W based on the roughness data in the circumferential direction at each axial position of the work W. The amount of chatter in the circumferential direction at each axial position of the workpiece W is calculated. Here, as a method for evaluating the amount of chatter in the circumferential direction, for example, it can be quantitatively quantified by calculating the difference between the maximum value and the minimum value of the roughness data.

Then, the chatter evaluation value calculation unit 73 calculates, for example, the average value and the variation of the chatter amount in the circumferential direction at each axial position of the workpiece W by statistical calculation. As a result, the output unit 74 calculates the average value and variation calculated by the chatter evaluation value calculation unit 73 with respect to the chatter amount, that is, the basic statistics as the chatter evaluation value in the circumferential direction of the entire workpiece W. Here, the chatter evaluation value calculation unit 73 calculates the chatter evaluation value of the final machined surface (grinded surface) after spark-out. In the chatter evaluation device 70, the chatter evaluation value calculation unit 73 calculates the process information (rough grinding step, fine grinding step, etc.) in the chatter evaluation value in the same manner as the control device 12 associates the process information with the state data. It is also possible to link the fine grinding process and the spark-out process).

The output unit 74 outputs the chatter evaluation value calculated by the chatter evaluation value calculation unit 73. Here, in addition to outputting the chatter evaluation value, the output unit 74 can output the additional information added by the chatter evaluation value calculation unit 73. The additional information includes information for determining the disposal of the work W and selecting the workmanship of the work W. For example, the chatter evaluation value calculation unit 73 can add additional information when the average value or variation of the chatter amount in the circumferential direction is large and the chatter evaluation value is larger than a preset threshold value.

Further, the additional information can include information for determining the necessity of growing the grindstone 66. When the surface of the grindstone 66 is deteriorated by the grinding process, the amount of chatter in the circumferential direction increases in the entire axial region of the workpiece W.

For this reason, the chatter evaluation value calculation unit 73 determines, for example, when the average value of the amount of chatter in the circumferential direction at each axial position of the workpiece W increases or when the variation in the amount of chatter in the circumferential direction increases, the truing Add additional information for determining the necessity. The magnitude of the variation in the amount of chatter is determined based on, for example, the difference between the maximum and minimum values of the amount of chatter in the circumferential direction at each axial position of the workpiece W, and the standard deviation. be able to.

(6. Functional block configuration of chatter prediction system 1)
The functional block of the chatter prediction system 1 will be described with reference to FIG. The chatter prediction system 1 includes a control device 12, a first detector 13, a second detector 14 (chatter evaluation device 70), a learning processing device 20, a prediction calculation device 30, and display devices 40 and 50.

As described above, the first detector 13 detects the observable state data in the cylindrical grinding machine 60 during the machining of the workpiece W. State data (driving current data and vibration data) is detected for each grinding process, that is, for each rough grinding process, fine grinding process, fine grinding process, and spark out, from the start to the end of machining on one workpiece W. Is output. The state data includes, for example, drive load data in the motor 66a that rotationally drives the grindstone 66. The drive load data corresponds to the drive current data of the motor 66a. The state data may preferably include vibration data among position data, temperature data, vibration data, processing sound data, and the like.

As described above, the second detector 14, that is, the chatter evaluation device 70, detects (outputs) the chatter evaluation value which is observable chatter-related data in the cylindrical grinding machine 60 during the machining of the workpiece W. The chatter evaluation value is data calculated by statistically calculating the chatter amount in each circumferential direction at a plurality of positions in the axial direction of one workpiece W. The chatter evaluation value can be detected for each grinding process, but it is desirable to calculate (evaluate) it on the final machined surface after spark-out.

Further, the control device 12 of the cylindrical grinding machine 60 includes a counter. The counter counts the number of workpieces W ground by the cylindrical grinding machine 60. In this example, the counter counts the number of times the workpiece W has been machined since the grindstone repairing device 68 slewed or dressed the grindstone 66. In addition to the control device 12, the counter may be provided by the first detector 13 itself or by the inspection device 2.

The learning processing device 20 has been trained to predict chatter vibration based on the state data detected by the first detector 13 and the chatter evaluation value output from the chatter evaluation device 70 (second detector 14). Generate a model. The learning processing device 20 includes a training data set acquisition unit 81, a training data set storage unit 82, and a model generation unit 83.

The training data set acquisition unit 81 acquires a training data set for performing machine learning. The training data set acquisition unit 81 includes a state data acquisition unit 81a, a state data division unit 81b, a feature amount extraction unit 81c, and a chatter-related data acquisition unit 81d.

The state data acquisition unit 81a acquires the state data detected by the first detector 13 during the machining of the workpiece W, particularly the drive current data which is the drive load data, in chronological order via the control device 12. do. The state data acquisition unit 81a acquires state data (driving current data) detected by each of the first detectors 13 provided on the plurality of cylindrical grinding machines 60 in chronological order.

The state data dividing unit 81b divides the state data acquired by the state data acquisition unit 81a from the control device 12 in time series (continuously) for each grinding process. That is, the state data dividing unit 81b obtains the state data continuously acquired by the state data acquisition unit 81a based on the process information associated with the state data, for example, in the rough grinding process, the fine grinding process, and the fine grinding. Divide into each process and spark-out process.

The feature amount extraction unit 81c displays the operating state of the cylindrical grinding machine 60 with respect to the state data for each grinding process (rough grinding process, fine grinding process, fine grinding process, spark-out process, etc.) divided by the state data dividing unit 81b. Extract the feature amount to be reflected. In this example, the feature amount extraction unit 81c performs frequency analysis (specifically, FFT (Fast Fourier Transform)) on the drive current data which is the state data having the frequency characteristic. As a result, the feature amount extraction unit 81c extracts the grindstone rotation frequency corresponding to the rotation speed of the grindstone wheel 66 described above, that is, the amplitude of the specific frequency, as the feature amount from the plurality of frequency components of the drive current data.

In this case, as for the amplitude (feature amount) of the grindstone rotation frequency, the larger the amplitude, the larger the drive current required for the grinding process. It can reflect the worsening operating condition. On the other hand, as for the amplitude (feature amount) of the grindstone rotation frequency, the smaller the amplitude, the smaller the drive current required for grinding. For example, the grindstone 66 of the cylindrical grinding machine 60 appropriately grinds the surface of the workpiece W. The operating state can be reflected.

Specifically, as shown in FIG. 5, the feature amount extraction unit 81c performs FFT (Fast Fourier Transform) as a frequency analysis on the drive current data divided by the state data division unit 81b. As a result, as shown in FIG. 6, the amplitudes of each of the plurality of frequency components included in the drive current data can be obtained. Then, the feature amount extraction unit 81c extracts the amplitude A of the grindstone rotation frequency F, which is a specific frequency, as the feature amount from the amplitudes of the plurality of frequency components shown in FIG.

The chatter-related data acquisition unit 81d acquires the chatter evaluation value calculated and output by the second detector 14, that is, the chatter evaluation value calculation unit 73 of the chatter evaluation device 70. The chatter-related data acquisition unit 81d acquires the chatter evaluation value of the final machined surface (after spark-out). If necessary, the chatter-related data acquisition unit 81d can also acquire the chatter evaluation value for each grinding process.

The training data set storage unit 82 stores the training data set acquired by the training data set acquisition unit 81. Specifically, the training data set storage unit 82 has the feature amount of the state data extracted by the feature amount extraction unit 81c, that is, the amplitude A of the grindstone rotation frequency F (specific rotation frequency) and the chatter-related data acquisition unit 81d. It is stored in association with the frequency evaluation value obtained by.

The model generation unit 83 performs machine learning using the training data set stored in the training data set storage unit 82. Specifically, the model generation unit 83 uses the amplitude A of the grindstone rotation frequency F (specific rotation frequency), which is the feature amount for each grinding process extracted by the feature amount extraction unit 81c, as an explanatory variable, and aims at the chatter evaluation value. Perform machine learning as a variable. Then, the model generation unit 83 generates a trained model for predicting chatter vibration in the grinding process.

The prediction calculation device 30 predicts the chatter evaluation value as an evaluation value for evaluating the chatter vibration in the grinding process based on the state data during machining in the corresponding cylindrical grinding machine 60. The prediction calculation device 30 includes a model storage unit 91, a prediction data acquisition unit 92, an evaluation value prediction unit 93, and an output unit 94.

The model storage unit 91 stores the trained model generated by the model generation unit 83. The prediction data acquisition unit 92 acquires prediction data during processing in the grinding process to be predicted. The prediction data acquisition unit 92 includes a state data acquisition unit 92a, a state data division unit 92b, and a feature amount extraction unit 92c.

The state data acquisition unit 92a obtains state data detected by the first detector 13 and output via the control device 12, particularly drive load data and drive current data, during machining in the grinding process to be predicted. get. Specifically, the state data acquisition unit 92a acquires state data (drive current data) associated with the process information for each of the rough grinding process, the fine grinding process, the fine grinding process, and the sparkout process to be predicted. ..

The state data dividing unit 92b divides the state data acquired by the state data acquisition unit 92a for each grinding process. Specifically, the state data dividing unit 92b divides the state data for each grinding process based on the process information associated with the state data.

The feature amount extraction unit 92c extracts the amplitude A of the grindstone rotation frequency F (specific rotation frequency), which is the feature amount of the state data (driving current data) divided for each grinding process by the state data division unit 92b. Here, in this example, the feature quantity to be extracted is the amplitude A of the waveform at the grindstone rotation frequency F after performing FFT (Fast Fourier Transform) on the drive current data which is the state data. Therefore, the feature amount extracted by the feature amount extraction unit 92c is the same as the feature amount extracted by the feature amount extraction unit 81c of the training data set acquisition unit 81.

Here, the state data acquisition unit 92a, the state data division unit 92b, and the feature amount extraction unit 92c perform the same processing as the state data acquisition unit 81a, the state data division unit 81b, and the feature amount extraction unit 81c of the training data set acquisition unit 81. I do. In this example, the state data acquisition unit 92a, the state data division unit 92b, and the feature amount extraction unit 92c are the state data acquisition unit 81a, the state data division unit 81b, and the feature amount extraction unit 81c of the training data set acquisition unit 81. It will be explained as a separate element from.

However, it is also possible to use the elements 81a, 81b, 81c of the training data set acquisition unit 81 together with the elements 92a, 92b, 92c of the prediction data acquisition unit 92. That is, the functions of the elements 81a, 81b, and 81c in the learning processing device 20 are also used as a part of the functions of the prediction calculation device 30.

The evaluation value prediction unit 93 acquires the trained model stored in the model storage unit 91. Further, the evaluation value prediction unit 93 acquires the feature amount in the grinding process of the prediction target acquired by the prediction data acquisition unit 92, that is, the amplitude A of the grindstone rotation frequency F at the specific rotation frequency. As a result, the evaluation value prediction unit 93 predicts (estimates) the estimated chatter evaluation value based on the learned model and the feature amount (amplitude A of the grindstone rotation frequency F).

The output unit 94 outputs the estimated chatter evaluation value predicted by the evaluation value prediction unit 93 to the display devices 40 and 50. Here, the output unit 94 uses, for example, additional information output from the second detector 14 (chatter evaluation device 70), that is, the quality of the workpiece W (surface roughness on the ground surface, etc.) and the grindstone 66 (the grindstone 66 (). The worn state of the grindstone T) can also be output. Here, it can be determined that the wear state of the grindstone 66 (grindstone T) is advanced (worse) as the estimated chatter evaluation value becomes larger. Further, the output unit 94 displays, for example, the feature amount extracted by the feature amount extraction unit 92c, that is, the amplitude A of the grindstone rotation frequency F, or the state data detected by the first detector 13, that is, the cylinder. It is possible to output various information regarding the operation of the grinding machine 60 (grinding machine 11).

The display devices 40 and 50 display the information output from the output unit 94 of the prediction calculation device 30. Specifically, the display devices 40 and 50 display the estimated chatter evaluation value output from the prediction calculation device 30. As a result, the operator or the control device 12 of the cylindrical grinding machine 60 (grinding machine 11) can confirm the quality of the workpiece W processed through the grinding process.

In addition, the display devices 40 and 50 can display various information regarding the operation of the cylindrical grinding machine 60 (grinding machine 11) and the wear state of the grindstone 66 as well as the estimated chatter evaluation value. Thereby, the operator or the control device 12 can also determine whether or not an abnormality has occurred in the cylindrical grinding machine 60 (grinding machine 11) performing the grinding process.

The operator or the control device 12 can operate the grindstone correction device 68, for example, when the wear state of the grindstone 66 progresses and the estimated chatter evaluation value is larger than a preset reference value. .. As a result, the operator or the control device 12 can carry out the truing of the grindstone wheel 66 or replace the grindstone wheel 66.

Further, the display devices 40 and 50 can also display the feature amount for each grinding process, for example, so that the operator can grasp the improvement points. As a result, the operator can easily find a feature amount useful for improvement in the feature amount for each grinding process, that is, the magnitude of the amplitude of the grindstone rotation frequency (specific rotation frequency).

(6. First alternative example)
In this example described above, the trained model generated by the model generation unit 83 of the learning processing device 20 is stored in the model storage unit 91 of the prediction calculation device 30. By the way, the trained model is generated by performing machine learning using the training data set stored in the training data set storage unit 82. Therefore, as the number of training data sets increases, in other words, as the learning by machine learning progresses, a trained model with high accuracy can be generated, and it is preferable to update the trained model sequentially.

Therefore, in the first alternative example, as shown in FIG. 7, the learning processing device 20 can be configured to include the model update unit 84 as the update determination unit. As shown in detail in FIG. 8, the model update unit 84 includes a chatter evaluation value acquisition command unit 84a, a chatter evaluation value acquisition unit 84b, a model update determination unit 84c, and a model update command unit 84d.

The chatter evaluation value acquisition command unit 84a acquires the chatter evaluation value from the second detector 14 (chatter evaluation device 70) at a predetermined frequency, for example, every time the machining number N of the workpiece W becomes the reference machining number N1 or more. Is output to the chatter evaluation value acquisition unit 84b. When the chatter evaluation value acquisition unit 84b acquires the acquisition command information, the chatter evaluation value is acquired from the second detector 14 (chatter evaluation device 70). Then, the chatter evaluation value acquisition unit 84b outputs the acquired chatter evaluation value to the model update determination unit 84c. When the chatter evaluation value is acquired from the chatter evaluation device 70, the chatter evaluation value acquisition command unit 84a outputs the chatter evaluation value to the chatter evaluation device 70 so that the chatter evaluation value is output to the chatter evaluation value acquisition unit 84b. It is also possible to output.

The model update determination unit 84c acquires the chatter evaluation value output from the chatter evaluation value acquisition unit 84b, and also acquires the estimated chatter evaluation value predicted by the evaluation value prediction unit 93. The model update determination unit 84c predicts using the chatter evaluation value output based on the acceleration data actually measured by the sizing device 67 and the learned model stored in the model storage unit 91 in the evaluation value prediction unit 93. The difference is calculated by comparing with the estimated chatter evaluation value. Then, the model update determination unit 84c determines whether or not the magnitude of the calculated difference exceeds the preset determination value, that is, whether or not the deviation between the chatter evaluation value and the estimated chatter evaluation value is large.

When the difference exceeds the determination value (large deviation) in the determination, the model update determination unit 84c outputs the update request information requesting the update of the learned model to the model update command unit 84d. The model update determination unit 84c does not output update request information to the model update command unit 84d when the difference is equal to or less than the determination value (small deviation) in the determination.

When the model update command unit 84d acquires the update request information from the model update determination unit 84c, the model update command unit 84d requests the model generation unit 83 to generate the trained model again. When the model generation unit 83 acquires the update request information from the model update command unit 84d, the model generation unit 83 performs machine learning using the training data set stored in the training data set storage unit 82 in the same manner as the machine learning described above. Generate a new trained model. Then, the model storage unit 91 updates and stores the newly generated trained model.

By updating the trained model in this way, the accuracy of the trained model can be improved, and the estimated chatter evaluation value can be predicted with high accuracy.

(7. Second alternative example)
In the learning processing device 20 of this example described above, the feature amount extraction unit 81c performs an FFT on the drive current data acquired from the first detector 13 as state data, and sets the feature amount at a specific rotation frequency (grind rotation frequency F). The amplitude A of the drive current is extracted. In addition to this, the feature amount extraction unit 81c can acquire vibration data (acceleration data) for each grinding process from the first detector 13 as state data. In this case, the vibration data (acceleration data) also has frequency characteristics, and the feature amount extraction unit 81c performs frequency analysis (FFT) on the acquired vibration data (acceleration data), and the specific rotation frequency (grinding stone) is used as the feature amount. The amplitude of vibration (acceleration) at (rotation frequency) is extracted.

Then, the training data set storage unit 82 has the amplitude A of the drive current at the grindstone rotation frequency F (specific rotation frequency) extracted by the feature quantity extraction unit 81c, and the vibration (acceleration) at the grindstone rotation frequency (specific rotation frequency). And the chatter-related data (for example, the chatter evaluation value) are stored as a training data set. As a result, the model generation unit 83 performs machine learning using the amplitude A of the drive current and the amplitude of the vibration (acceleration) as explanatory variables and the chatter-related data (chatter evaluation value) as objective variables, and generates a trained model. be able to.

That is, in this case, machine learning can be performed by increasing the number of explanatory variables as compared with the case of this example described above. Therefore, the model generation unit 83 can improve the generation accuracy, and as a result, can generate a trained model with higher accuracy.

(8. Third alternative example)
In this example, the first example, and the second example described above, the chatter evaluation device 70 is used as the second detector 14, and the chatter evaluation device 70 is in the grinding process of the workpiece W or immediately after the process, so-called. , The chatter evaluation value is output as chatter-related data by in-process. That is, in this example, the first example, and the second example described above, the chatter evaluation device 70 is provided in each cylindrical grinding machine 60, that is, the chatter evaluation device 70 is provided in the machine of the grinding device 10. .. As a result, each chatter evaluation device 70 provided as the second detector 14 in the machine outputs the chatter evaluation value immediately after processing. The chatter evaluation device 70 can also output chatter evaluation values for each of the rough grinding process, the fine grinding process, the fine grinding process, and the spark-out process.

Instead of this, as shown in FIG. 1, as an inspection device 2 provided in-line outside the grinding device 10, for example, a chatter evaluation device 70 can be used. In this case, the chatter evaluation value after processing can be evaluated outside the machine.

Specifically, in the third alternative example, as shown by the broken lines in FIGS. 4 and 7, the chatter-related data acquisition section 81d of the training data set acquisition section 81 is provided in-line and chatter evaluation is performed from the chatter evaluation device 70. Get the value. Therefore, also in this third alternative example, the chatter prediction system 1 operates in the same manner as the above-mentioned present example, the first alternative example, and the second alternative example, and the same effect can be obtained.

(9. Others)
In this example, the first example, the second example, and the third example described above, the chatter-related data acquisition unit 81d of the training data set acquisition unit 81 was used as the second detector 14 or the inspection device 2. The chatter evaluation value is acquired from the chatter evaluation device 70 as chatter-related data. However, the chatter-related data output from the second detector 14 or the inspection device 2 is not limited to the chatter evaluation value, and for example, the roughness that directly detects the surface roughness of the ground surface of the workpiece W is detected. The data may be data, acceleration data in which acceleration generated due to the surface roughness of the ground surface is directly detected, or vibration data representing vibration indirectly detected from the acceleration data.

Regarding these roughness data, acceleration data, and vibration data, for example, the average value, maximum value, minimum value, and dispersion value obtained by statistically calculating after detecting the entire ground surface of the workpiece W, The standard deviation, etc. is output as an evaluation value. Then, the chatter-related data acquisition unit 81d acquires the output evaluation value as chatter-related data and outputs it to the training data set storage unit 82. As a result, the training data set storage unit 82 stores the training data set whose evaluation value is the objective variable, and the model generation unit 83 generates the trained model by performing machine learning using the training data set. Can be done.

Further, in this example and the first to third examples described above, the learning processing device 20 and the prediction calculation device 30 are used for each grinding process, that is, a rough grinding process, a fine grinding process, a fine grinding process, and a spark. The state data according to the out process and the chatter-related data immediately after processing are acquired. Then, the prediction calculation device 30 predicts the chatter amount (for example, the estimated chatter evaluation value).

However, the state data does not necessarily have to be divided for each grinding process. In this case, the learning processing device 20 continuously (time-series) acquires the state data, and trains the state data and the chatter-related data immediately after processing for the plurality of workpieces W acquired at an arbitrary time point. It is also possible to generate a trained model as a dataset. Further, the prediction arithmetic unit 30 continuously (time-series) acquires state data, and uses the acquired state data at an arbitrary time point and the trained model to obtain a chatter amount (for example, an estimated chatter evaluation value). It is also possible to predict and output.

In this case, since there is no distinction for each grinding process, for example, the state data division unit 81b in the training data set acquisition unit 81 of the learning processing device 20 and the state data division unit 92 in the prediction data acquisition unit 92 of the prediction calculation device 30. It is possible to omit 92b. Thereby, the configurations of the learning processing device 20 and the prediction calculation device 30 can be simplified. On the other hand, since there is no distinction between each grinding process, the accuracy of generating the trained model and the accuracy of predicting the chatter amount (estimated chatter evaluation value) are slightly worse depending on the grinding process, as compared with the case of this example described above. There is a risk.

Further, in this example and the first to third examples described above, the state data division unit 81b and the state data division unit 92b are the state data acquired by the state data acquisition unit 81a and the state data acquisition unit 92a. Was divided sequentially. Instead, the state data dividing unit 81b and the state data dividing unit 92b are based on the process information after the state data acquisition unit 81a and the state data acquisition unit 92a acquire the state data with the progress of all the grinding processes. It is also possible to divide the state data for each grinding process.

1 ... chatter prediction system, 2 ... inspection device, 10 ... grinding device, 11 ... grinding machine, 12 ... control device, 13 ... first detector, 14 ... second detector, 15 ... interface, 20 ... learning processing device, 21 ... Processor, 22 ... Storage device, 23 ... Interface, 30 ... Prediction calculation device, 31 ... Processor, 32 ... Storage device, 33 ... Interface, 40 ... Common display device, 50 ... Individual display device, 60 ... Cylindrical grinder, 61 ... bed, 62 ... headstock, 62a ... motor, 63 ... tailstock, 64 ... traverse base, 64a ... motor, 65 ... grinding table, 65a ... motor, 66 ... grinding wheel, 66a ... motor, 67 ... fixed size Device, 67a ... Feed mechanism, 67b ... Stylus, 67c ... Acceleration sensor, 68 ... Grinding wheel correction device, 69 ... Coolant device, 70 ... Chatter evaluation device, 71 ... Actual data acquisition unit, 72 ... Displacement conversion unit, 73 ... Evaluation value calculation unit, 74 ... Output unit, 81 ... Training data set acquisition unit, 81a ... State data acquisition unit, 81b ... State data division unit, 81c ... Feature amount extraction unit, 81d ... Chatter-related data acquisition unit, 82 ... Training Data set storage unit, 83 ... model generation unit, 84 ... model update unit, 84a ... chatter evaluation value acquisition command unit, 84b ... chatter evaluation value acquisition unit, 84c ... model update determination unit, 84d ... model update command unit, 91 ... Model storage unit, 92 ... prediction data acquisition unit, 92a ... state data acquisition unit, 92b ... state data division unit, 92c ... feature amount extraction unit, 93 ... evaluation value prediction unit, 94 ... output unit, A ... amplitude (feature) Amount), F ... Grinding stone rotation frequency (specific rotation frequency), N ... Number of machining, N1 ... Standard machining number, T ... Grinding stone, W ... Work piece

Claims (14)

A grinding device in which the grindstone grinds the workpiece through multiple grinding processes controlled by the control device.
A first detector that detects state data related to the operating state of the grindstone that can be observed in the grinding process by the grinding device and outputs the detected state data.
A second detector that detects chatter-related data related to the magnitude of chatter vibration generated in the grinding process by the grinding device and outputs the detected chatter-related data.
The feature amount of the state data acquired from the first detector is used as an explanatory variable, the chatter-related data acquired from the second detector is used as an objective variable, and a training data set including the explanatory variable and the objective variable is used. A trained model storage unit that stores the trained model generated by performing machine learning,
An evaluation value prediction unit that predicts and outputs an evaluation value for evaluating the chatter vibration generated in the grinding process by the grinding device using the learned model and the state data, and an evaluation value prediction unit.
A chatter prediction system equipped with. The state data has frequency characteristics and has frequency characteristics.
The chatter prediction system according to claim 1, further comprising a feature amount extraction unit that extracts the feature amount by frequency-analyzing the state data. The chatter according to claim 2, wherein the feature amount extraction unit divides the state data into a plurality of frequency components by frequency analysis, and extracts the amplitude of a specific frequency among the plurality of frequency components as the feature amount. Prediction system. The chatter prediction system according to claim 3, wherein the specific frequency is a grindstone rotation frequency corresponding to the rotation speed of the grindstone of the grinding device. The state data is at least one of drive current data corresponding to the drive load generated by the grinding process in the grinding device and vibration data representing the magnitude of vibration generated by the grinding process in the grinding device. The chatter prediction system according to any one of claims 2-4, including one. The chatter-related data is any one of claims 1-5, which is acceleration data or displacement data of the chatter vibration generated in response to the surface roughness of the grounded workpiece. The chatter prediction system described in. The chatter-related data is a chatter evaluation value representing the magnitude of the chatter vibration calculated by statistically calculating the acceleration data or the displacement data at a plurality of locations of the workpiece.
The chatter prediction system according to claim 6, wherein the evaluation value prediction unit outputs an estimated chatter evaluation value predicted using the trained model and the state data. When the workpiece has a peripheral surface around the axis
The chatter prediction system according to claim 7, wherein the chatter evaluation value is calculated by statistically calculating the acceleration data or the displacement data in the circumferential direction at a plurality of locations in the axial direction of the workpiece. It is provided with an update determination unit for determining whether or not the trained model needs to be updated by comparing the chatter-related data with the evaluation value predicted using the trained model.
The trained model storage unit stores the updated trained model when the trained model is updated based on the determination of the update determination unit, any one of claims 1-8. The chatter prediction system described in. The chatter prediction system according to any one of claims 1-9, wherein the second detector is provided integrally with the grinding device or separately. The trained model storage unit stores the trained model generated by performing machine learning using the feature amount of the state data generated for each grinding process and the chatter-related data. , The chatter prediction system according to any one of claims 1-9. The chatter prediction system according to claim 11, wherein the grinding step includes at least a rough grinding step, a fine grinding step, a fine grinding step, and a spark-out step. When the first detector outputs the state data via the control device, the state data is added with process information for identifying the grinding process by the control device.
The chatter prediction system according to claim 11 or 12, further comprising a state data dividing unit that divides the state data for each grinding process based on the process information. The feature amount of the state data acquired from the first detector is used as the explanatory variable, the chatter-related data acquired from the second detector is used as the objective variable, and the training data including the explanatory variable and the objective variable is used. The chatter prediction system according to any one of claims 1-11, comprising a trained model generation unit that generates the trained model by performing machine learning using the set.

Images