JP7451948B2 - Machining quality prediction system - Google Patents

Description

The present invention relates to a processing quality prediction system.

When grinding a workpiece using a grinding wheel, it is required that the grinding quality of the workpiece satisfies predetermined conditions. For example, it is necessary to prevent a workpiece from forming a damaged layer, to ensure that the surface properties of the workpiece (for example, surface roughness) are within a specified value, and to prevent chatter patterns from forming on the workpiece. It will be done.

An operator inspects the workpiece after grinding to confirm whether the grinding quality satisfies predetermined conditions, and a workpiece that satisfies the predetermined conditions is determined to be a good product. Here, Patent Document 1 describes that it is determined whether or not a work-affected layer has occurred in a workpiece based on a grinding load measured during grinding.

In addition, when grinding a workpiece using a grinding wheel, truing (used to include dressing) is performed on the surface of the grinding wheel in order to maintain the sharpness of the grinding wheel. If the sharpness of the grinding wheel decreases, the quality of the workpiece may deteriorate. Therefore, truing is performed every time the number of ground workpieces reaches a predetermined number, and the predetermined number is determined so that the quality of the workpieces does not deteriorate. However, since it is determined by the operator, there is a risk that grinding will continue despite the reduction in sharpness, leading to a risk of deterioration in the quality of the workpiece.

Therefore, Patent Document 2 discloses that the vibration of the spindle head is detected by a vibration detector attached to the spindle head, and the amplitude of the spindle reaches a preset value according to the grinding accuracy of the grinding surface of the workpiece. Afterwards, it is described that the grinding operation is stopped and the grinding wheel is dressed.

By the way, in recent years, with the improvement of computer processing speed, artificial intelligence has been rapidly developing. For example, Patent Document 3 describes generating laser processing condition data by machine learning.

Japanese Patent Application Publication No. 2013-129028 Japanese Patent Application Publication No. 2002-307304 Japanese Patent Application Publication No. 2017-164801

The formation of a process-affected layer on a workpiece is caused by the high grinding point temperature during grinding. Here, it is known that the higher the grinding load, the higher the grinding point temperature. However, in addition to the case where the grinding load is high, the grinding point temperature also becomes high, for example, when the cooling capacity of the coolant is low. Therefore, it is not possible to grasp the state of the processed damaged layer with high precision using only the grinding load.

As a means for measuring the grinding point temperature, it is conceivable to use a non-contact temperature measuring device such as a thermography to measure the grinding point temperature. However, grinding is generally performed while supplying coolant, and temperature measuring instruments can only measure the temperature of the coolant flowing near the grinding point, making it easy to measure the temperature at the grinding point. isn't it. As a result, by using the temperature measuring device, it is not possible to grasp the state of the process-affected layer.

An object of the present invention is to provide a machining quality prediction system that can predict the state of a machining-affected layer generated on a workpiece with higher accuracy.

The machining quality prediction system includes a grinding machine body that grinds the workpiece with the grinding wheel while rotating the workpiece and the grinding wheel , and an observable state of the grinding machine body during the grinding process of the workpiece. A detector for detecting data, and a machining process generated on the workpiece by increasing the temperature of the grinding point where the grinding wheel and the workpiece come into contact, using the state data detected by the detector. The method further includes a prediction unit that predicts the presence or absence of an affected layer, the depth of the processed affected layer, the surface hardness of the workpiece, or the burnt color of the surface of the workpiece . The detector includes a heat flow sensor that detects heat flow data, which is a magnitude of heat flux in a predetermined direction, as the state data, and the heat flow sensor is arranged on an outer circumferential surface of at least one of the grinding wheel and the workpiece. It is arranged opposite to a predetermined position different from the grinding point, and detects the magnitude of the heat flux in the predetermined direction, which is the direction from the predetermined position toward the heat flow sensor .

Here, the heat generated at the grinding point during the grinding process is transferred to the workpiece, grinding wheel, chips, and coolant. The heat flow sensor detects heat radiation data from a position different from the grinding point on the surface of at least one of the grinding wheel and the workpiece. In other words, the heat flow sensor does not directly detect the temperature at the grinding point. However, since the heat generated at the grinding point is transferred to the workpiece and the grinding wheel, heat remains on the surfaces of the workpiece and the grinding wheel even if the surfaces are different from the grinding point. Therefore, the heat flow sensor can detect heat radiation data corresponding to the heat generated at the grinding point even at a position different from the grinding point on the surface of at least one of the grinding wheel and the workpiece. As a result, by using the heat radiation data detected by the heat flow sensor, it is possible to predict the state of the process-affected layer with high accuracy.

FIG. 1 is a diagram showing the configuration of a processing quality prediction system. It is a top view showing a grinding machine. It is a diagram showing the arrangement of a grinding wheel, a workpiece, and a heat flow sensor, and is an enlarged view of the grinding wheel and the workpiece viewed from the axial direction. It is a diagram showing a functional block configuration of a processing quality prediction system. FIG. 3 is a diagram showing the relationship between types of state data, types of prediction targets, and learned models. FIG. 7 is a diagram showing the degree of influence on the state of a process-affected layer depending on the type of state data. It is a graph showing heat radiation data obtained by a heat flow sensor and vibration data generated in a support shaft of a grinding wheel.

(1. Configuration of processing quality prediction system 1)
The hardware configuration of the machining quality prediction system 1 will be explained with reference to FIG. 1. The machining quality prediction system 1 includes at least one grinding machine 10 and one arithmetic device 20. The grinding machine 10 may be used for one machine or for a plurality of machines. In this example, the machining quality prediction system 1 includes one grinder 10. In this example, the processing quality prediction system 1 further includes a display device 30.

The grinding machine 10 includes at least a grinding machine main body 11 that grinds a workpiece W using a grinding wheel T, a control device 12, a detector 13, and an interface 14. The grinding machine main body 11 has a grinding wheel T, supports a workpiece W, and has a configuration for relatively moving the grinding wheel T and the workpiece W. The grinding machine main body 11 grinds a workpiece W using a grinding wheel T.

The control device 12 controls the grinding machine main body 11. The control device 12 includes a CNC device, a PLC device, and the like. The detector 13 detects observable state data on the grinding machine body 11 during grinding of the workpiece W. Detector 13 includes at least a heat flow sensor. The interface 14 is a device that allows the grinding machine body 11, the control device 12, and the detector 13 to communicate with the outside.

The arithmetic device 20 uses the state data detected by the detector 13 to predict the machining quality of the workpiece W by applying machine learning. In particular, the calculation device 20 predicts the state of the machining-affected layer as the machining quality of the workpiece W. Furthermore, the calculation device 20 can also predict the chatter state, roundness, and surface roughness as the machining quality of the workpiece W.

The arithmetic device 20 is placed close to the grinding machine 10 and functions as a so-called edge computer. The arithmetic device 20 includes a processor 21, a storage device 22, an interface 23, and the like. The computing device 20 is communicatively connected to the grinding machine 10 .

(2. Example of grinding machine 10)
As an example of the grinder 10, a cylindrical grinder will be described with reference to FIG. The grinding machine 10 is a machine tool for grinding a workpiece W. As the grinding machine 10, grinding machines of various configurations can be applied, such as a cylindrical grinding machine and a cam grinding machine. In this example, the grinding machine 10 is a cylindrical grinding machine of a grindstone traverse type. However, the grinding machine 10 can also be of a table traverse type.

The grinding machine 10 includes a grinding machine main body 11, a control device 12, a detector 13, and an interface 14 (shown in FIG. 1). The grinding machine main body 11 mainly includes a bed 41, a headstock 42, a tailstock 43, a traverse base 44, a grindstone head 45, a grinding wheel T, a sizing device 46, a grinding wheel correction device 47, and a coolant nozzle 48.

Bed 41 is fixed on the installation surface. The headstock 42 is provided on the upper surface of the bed 41 on the front side in the X-axis direction (lower side in FIG. 2) and on one end side in the Z-axis direction (left side in FIG. 2). The headstock 42 supports the workpiece W rotatably around the Z-axis. The workpiece W is rotated by the drive of a motor 42a provided on the headstock 42. The tailstock 43 is located on the upper surface of the bed 41 at a position opposite to the headstock 42 in the Z-axis direction, that is, on the front side in the X-axis direction (lower side in FIG. 2) and on the other end side in the Z-axis direction ( (right side in FIG. 2). That is, the headstock 42 and the tailstock 43 rotatably support the workpiece W at both ends.

The traverse base 44 is provided on the upper surface of the bed 41 so as to be movable in the Z-axis direction. The traverse base 44 is moved by the drive of a motor 44a provided on the bed 41. The grindstone head 45 is provided on the upper surface of the traverse base 44 so as to be movable in the X-axis direction. The grindstone head 45 is moved by the drive of a motor 45a provided on the traverse base 44. The grinding wheel T is rotatably supported by the grinding wheel head 45. The grinding wheel T is rotated by the drive of a motor 45b provided on the grinding wheel head 45. The grinding wheel T is configured by fixing a plurality of abrasive grains with a bond material.

The sizing device 46 measures the dimensions (diameter) of the workpiece W. The grinding wheel correction device 47 corrects the shape of the grinding wheel T. The grinding wheel correction device 47 is a device that performs truing (including dressing) of the grinding wheel T. Furthermore, the grinding wheel correction device 47 also has a function of measuring the dimensions (diameter) of the grinding wheel T.

Here, truing is a reshaping operation, such as shaping the grinding wheel T to match the shape of the workpiece W when the grinding wheel T is worn out due to grinding, and removing runout of the grinding wheel T due to one-sided wear. It is. Dressing is a refinishing (sharpening) work that involves adjusting the protruding amount of abrasive grains and creating a cutting edge of the abrasive grains. Dressing is an operation to correct blindness, clogging, spillage, etc., and is usually performed after truing. However, since truing and dressing may be performed without any particular distinction, the term truing is used herein to include dressing.

The coolant nozzle 48 is provided near the grinding wheel T on the grinding wheel head 45. The coolant nozzle 48 supplies coolant to the grinding point of the workpiece W by the grinding wheel T. The coolant supplied from the coolant nozzle 48 to the grinding point is collected, cooled to a predetermined temperature, and then supplied from the coolant nozzle 48 to the grinding point again.

The control device 12 controls each drive device based on an NC program generated based on operation command data such as the shape of the workpiece W, processing conditions, the shape of the grinding wheel T, and coolant supply timing information. That is, the control device 12 inputs operation command data, generates an NC program based on the operation command data, and operates each motor 42a, 44a, 45a, 45b, a coolant device (not shown), etc. based on the NC program. Grinding of the workpiece W is performed by controlling the grinding process. In particular, the control device 12 performs grinding until the workpiece W has a finished shape based on the outer diameter of the workpiece W measured by the sizing device 46. Further, the control device 12 corrects the grinding wheel T (truing and dressing) by controlling each motor 44a, 45a, 45b, the grinding wheel correction device 47, etc. at the timing of correcting the grinding wheel T. .

Furthermore, the grinding machine 10 includes a plurality of detectors 13a to 13g that detect observable state data in the grinding machine body 11 during the grinding process of the workpiece W. However, the grinding machine 10 only needs to include at least the detector 13d, which is a heat flow sensor, among the plurality of detectors 13a to 13g.

The detector 13a is a rotational power sensor that detects rotational power data of the motor 42a that rotationally drives the workpiece W as state data. The detector 13b is an axial power sensor that detects axial power data for moving the grindstone head 45 in the X-axis direction as state data. The detector 13c is a rotational power sensor that detects rotational power data of the motor 45b that rotationally drives the grinding wheel T as state data.

The detector 13d is a heat flow sensor that detects heat radiation data from a position different from the grinding point on the surface of at least one of the grinding wheel T and the workpiece W as state data. In this example, the detector 13d as a heat flow sensor is arranged to face the surface of the grinding wheel T, and the case where heat radiation data from the surface of the grinding wheel T is detected is taken as an example. Here, the heat flux sensor can detect the magnitude of heat flux and the direction of heat flux. The heat radiation data is a value corresponding to the magnitude of heat flux. When the surface of the grinding wheel T is hot, heat is radiated from the surface of the grinding wheel T, so the detector 13d as a heat flow sensor measures the magnitude of the heat flux radiated from the grinding wheel T (heat radiation). data) can be detected. Note that the direction of the heat flow here is only the direction from the surface of the grinding wheel T toward the detector 13d.

The detector 13e is a vibration sensor that detects vibration data generated on a support shaft (support device) that supports the grinding wheel T as state data. The detector 13f is a vibration sensor that detects vibration data generated in the main shaft (support device) of the headstock 42 that supports the workpiece W or the tailstock shaft (support device) of the tailstock 43 as state data. As the vibration sensors as the detectors 13e and 13f, an acceleration sensor, an AE (acoustic emission) sensor, or the like can be used. The detector 13g is a processing sound sensor that detects sound data generated by the grinding process as state data. The machining sound sensor as the detector 13g can be placed at any position within the machining area.

(3. Position of heat flow sensor 13d)
The position of the heat flow sensor 13d will be explained with reference to FIG. 3. In this example, the heat flow sensor 13d is arranged to face the surface of the grinding wheel T, and detects heat radiation data from the surface of the grinding wheel T.

In FIG. 3, an example is given in which the grinding wheel T rotates clockwise and the workpiece W rotates counterclockwise. A grinding point Pa is formed at which the side surface of the grinding wheel T and the side surface of the workpiece W come into contact. In the grinding wheel T, Pa is the grinding point, Pb is the lower end, and Pc is the upper end. That is, in this example, the rotation direction of the grinding wheel T is such that a predetermined position on the outer peripheral surface of the grinding wheel T moves in the order of grinding point Pa (lateral position on the workpiece W side), lower end Pb, upper end Pc, and grinding point Pa. The direction is to Further, in the workpiece W, Pa is the grinding point, Pd is the lower end, and Pe is the upper end. That is, in this example, the rotation direction of the workpiece W is such that a predetermined position on the outer peripheral surface of the workpiece W moves in the order of grinding point Pa (lateral position on the grinding wheel T side), lower end Pd, upper end Pe, and grinding point Pa. The direction is to

Moreover, the coolant nozzle 48 supplies coolant to the grinding point Pa of the workpiece W by the grinding wheel T. In particular, as shown in FIG. 3, the workpiece W is ground by the grinding wheel T while the coolant nozzle 48 supplies coolant from a position above the grinding point Pa toward the grinding point Pa.

The heat flow sensor 13d is disposed on the outer peripheral surface of the grinding wheel T at a position opposite to the grinding point Pa. Specifically, the heat flow sensor 13d is disposed on the outer circumferential surface of the grinding wheel T at a position above the grinding point Pa so as to face each other. Therefore, the position of the grinding wheel T that the heat flow sensor 13d faces is a position that is unlikely to be directly exposed to coolant. As a result, the heat flow sensor 13d can detect heat radiation data on the surface of the grinding wheel T because the influence of the coolant is small.

Further, by being arranged as described above, the heat flow sensor 13d is located at a position shifted by 180 degrees or more from the grinding point Pa in the rotation direction of the grinding wheel T. Therefore, after the coolant is applied to the vicinity of the grinding point Pa of the grinding wheel T, the coolant is dispersed from the surface of the grinding wheel T by the centrifugal force accompanying the rotation of the grinding wheel T. Therefore, the amount of coolant adhering to the surface of the grinding wheel T facing the heat flow sensor 13d is reduced. As a result, the heat flow sensor 13d is less affected by the coolant, and can therefore detect heat radiation data on the surface of the grinding wheel T with higher accuracy.

In particular, the heat flow sensor 13d is disposed on the outer peripheral surface of the grinding wheel T at a position between the grinding point Pa and the upper end Pc of the grinding wheel T, facing each other. When the grinding wheel T is in the above-mentioned rotational direction, by arranging the heat flow sensor 13d at the position, the coolant can be dispersed more effectively by the centrifugal force accompanying the rotation of the grinding wheel T. Therefore, the heat flow sensor 13d is further less affected by the coolant, and can detect heat radiation data on the surface of the grinding wheel T with even higher accuracy.

Here, in the above, the heat flow sensor 13d was arranged at a position facing the surface of the grinding wheel T. In addition, the heat flow sensor 13d may be arranged at a position facing the surface of the workpiece W to detect heat radiation data from the surface of the workpiece W. However, if the outer diameter of the workpiece W is small, coolant may adhere to the entire surface of the workpiece W, so it is difficult to detect heat radiation data from the workpiece W with high accuracy. may not be possible. On the other hand, since the grinding wheel T generally has a large diameter, if the outer diameter of the workpiece W is small, it is recommended that the heat flow sensor 13d be placed at a position facing the surface of the grinding wheel T as described above. good.

When the heat flow sensor 13d is disposed at a position facing the surface of the workpiece W, it is preferably disposed at substantially the same position relative to the grinding wheel T described above. The heat flow sensor 13d is preferably disposed on the outer circumferential surface of the workpiece W at a position above the grinding point Pa so as to face each other. In particular, the heat flow sensor 13d is preferably disposed on the outer peripheral surface of the workpiece W at a position between the grinding point Pa and the upper end Pe of the workpiece W, facing each other.

(4. Functional block configuration of processing quality prediction system 1)
The functional block configuration of the machining quality prediction system 1 will be explained with reference to FIG. 4. The processing quality prediction system 1 includes a plurality of detectors 13a to 13g, a calculation device 20, and a display device 30. As described above, the detectors 13a-13g detect observable state data in the grinding machine body 11 during the grinding process of the workpiece W.

Further, an example will be given in which the processing quality prediction system 1 further includes a processing quality inspection device 50. The machining quality inspection device 50 is a device for inspecting the target machining quality, and may be a device separate from the grinding machine 10 or a device built into the grinding machine 10.

The calculation device 20 includes a learning processing device 60 and a prediction calculation device 70. The learning processing device 60 performs processing to generate a learned model by performing machine learning. In other words, the learning processing device 60 functions as a learning phase in machine learning. The prediction calculation device 70 uses the learned model to perform calculation processing to predict machining quality. In other words, the prediction calculation device 70 functions as an inference phase in machine learning. Note that the learning processing device 60 and the prediction calculation device 70 may constitute independent devices or may constitute one device.

The learning processing device 60 generates a learned model for predicting the machining quality of the workpiece W based on the state data detected by the detectors 13a to 13g and the machining quality obtained by the machining quality inspection device 50. generate. The learning processing device 60 includes a training dataset acquisition section 61 , a training dataset storage section 62 , a model generation section 63 , and an effective feature amount extraction section 64 .

The training data set acquisition unit 61 acquires a training data set for performing machine learning. Here, the training data set acquisition section 61 includes a state data acquisition section 61a, a feature amount calculation section 61b, and a processing quality data acquisition section 61c. The state data acquisition unit 61a acquires state data detected by each of the detectors 13a to 13g during the grinding process of the workpiece W.

The feature quantity calculation unit 61b calculates a plurality of feature quantities for the state data acquired by the state data acquisition unit 61a. Here, various statistical quantities in the state data are used as the feature quantity. For example, the feature amounts include the maximum value, minimum value, average value, variance, standard deviation, skewness, kurtosis, median value, etc. in the state data. Further, the feature amount may include a statistic for data obtained by differentiating the state data, a statistic for data obtained by frequency analysis of the state data, and the like. The feature amount calculation unit 61b may calculate all of the above feature amounts, or may calculate some of the feature amounts.

For example, if eight types of statistics are applied to three types of feature amounts: original data, differential data, and frequency analysis data, there will be 24 types of feature amounts. If 24 types of feature amounts are calculated for each of the seven types of detectors 13a to 13g, there will be a total of 168 types of feature amounts.

The machining quality data acquisition unit 61c acquires the machining quality data acquired by the machining quality inspection device 50 in association with the target workpiece W. The machining quality data includes at least altered layer data representing the state of a machining-affected layer produced on the workpiece W, and includes at least one of chatter state data, roundness data, and surface roughness data produced on the workpiece W.

That is, the training data set acquisition unit 61 uses a plurality of feature amounts for each of a plurality of state data as explanatory variables, and uses processing quality data as an objective variable, and acquires a training dataset including explanatory variables and objective variables. The training data set storage unit 62 stores the training data set acquired by the training data set acquisition unit 61.

The model generation unit 63 performs machine learning using the training data set stored in the training data set storage unit 62. Specifically, the model generation unit 63 performs machine learning using a plurality of feature quantities (for example, 168 types of feature quantities) for each of the plurality of state data as a first explanatory variable and processing quality data as a first objective variable. I do. That is, the model generation unit 63 performs machine learning using the first explanatory variable and the first objective variable as the first training data set. Then, the model generation unit 63 generates a first trained model for predicting machining quality. Here, the number of first trained models is generated depending on the type of processing quality. Therefore, the plurality of first learned models correspond to the state of the processed damaged layer, the chatter state, the roundness, and the surface roughness, respectively, as the processing quality.

Regression is preferably used as a machine learning method. For example, decision trees, linear regression, ridge regression, Lasso regression, elastic nets, random forests, etc. are useful. In particular, by using these methods, it is possible to understand the degree of influence of a plurality of feature quantities. Note that machine learning methods other than regression are also applicable.

The effective feature amount extraction unit 64 extracts effective feature amounts that have a high degree of influence on the first target variable from among the plurality of feature amounts by performing the above-described machine learning. The effective feature amount is not limited to one, but it is preferable to extract a plurality of effective feature amounts. When the state of the process-affected layer is used as the objective variable, it was confirmed that the degree of influence of the heat radiation data from the heat flow sensor 13d is high. Therefore, when the state of the process-affected layer is used as the objective variable, it is preferable to use heat radiation data as the effective feature quantity. Furthermore, among the plurality of feature quantities related to heat radiation data, a plurality of feature quantities having a high degree of influence may be set as effective feature quantities. Further, the effective feature amount extracting unit 64 also extracts effective feature amounts for each of the chatter state, roundness, and surface roughness.

After a plurality of effective feature quantities have been extracted by the effective feature quantity extraction unit 64, the model generation unit 63 performs machine learning again. Specifically, the model generation unit 63 uses a plurality of effective feature quantities from among the plurality of feature quantities stored in the training data set storage unit 62 as second explanatory variables, and calculates the processing quality corresponding to the effective feature quantity. Perform machine learning using data as the second objective variable. That is, the model generation unit 63 performs machine learning using the second explanatory variable and the second objective variable as the second training data set. Then, the model generation unit 63 generates a second trained model for predicting machining quality. Here, the number of second trained models is generated depending on the type of processing quality. The second trained model generated here is a different model from the first trained model generated first.

The prediction calculation device 70 predicts the machining quality of the workpiece W based on state data detected during the grinding process. The prediction calculation device 70 includes a model storage section 71 , a prediction data acquisition section 72 , a prediction section 73 , a display output section 74 , a workpiece post-processing section 75 , and an improvement processing section 76 .

The model storage unit 71 stores a plurality of second learned models generated by the model generation unit 63. The model storage unit 71 stores a plurality of second trained models corresponding to each of a plurality of machining qualities.

The prediction data acquisition unit 72 acquires prediction data during the grinding process. The prediction data acquisition section 72 includes a state data acquisition section 72a and an effective feature quantity calculation section 72b. The condition data acquisition unit 72a acquires condition data detected by the plurality of detectors 13a to 13g during the grinding process.

The effective feature quantity calculation unit 72b calculates one or more effective feature quantities according to the type of processing quality based on the plurality of detected state data. For example, when predicting the state of the process-affected layer, the effective feature amount calculation unit 72b uses, as the effective feature amount, at least the feature amount related to the heat radiation data detected by the heat flow sensor 13d.

The prediction unit 73 predicts the corresponding machining quality based on the second trained model stored in the model storage unit 71 and the data acquired by the prediction data acquisition unit 72. For example, the prediction unit 73 uses the second learned model corresponding to the state of the processed damaged layer and the feature amounts related to the state data detected by the detectors 13a to 13g in the effective feature amount calculation unit 72b of the processed changed layer. The state of the process-affected layer is predicted based on the effective feature amount corresponding to the state.

Here, the prediction unit 73 may predict the presence or absence of a work-affected layer. Furthermore, the prediction unit 73 may predict whether the degree of generation of the process-affected layer is greater than or equal to a predetermined threshold. The degree of formation of a work-affected layer corresponds to the generation temperature of the work-affected layer, and for example, it depends on the surface hardness of the workpiece W, the burnt color of the surface of the workpiece W, the depth of the work-affected layer, etc. This is the corresponding indicator. Further, the prediction unit 73 may predict the degree of generation of the process-affected layer.

The display output unit 74 outputs the predicted processing quality to the display device 30. The workpiece post-processing unit 75 sends the target workpiece W to the post-processing device 80 when the machining quality (state of machining-affected layer, etc.) predicted by the prediction unit 73 is in a predetermined state (bad state). to be sorted, discarded, or subjected to detailed inspection.

The improvement processing unit 76 causes the control device 12 to change the grinding process conditions when the machining quality (state of the process-affected layer, etc.) predicted by the prediction unit 73 is in a predetermined state (bad state), or , to execute truing of the grinding wheel T. Changes in the grinding conditions include, for example, slowing down the feed rate (speed in the cutting direction) for rough grinding, or increasing the machining allowance for finish grinding (fine grinding, fine grinding). In particular, the improvement processing unit 76 may change the grinding conditions according to the predicted depth of the process-affected layer. At that time, it is preferable to take cycle time into consideration and decide to slow down the feed rate in rough grinding and increase the machining allowance in finish grinding, depending on the predicted depth of the machining-affected layer.

(5. Relationship between type of state data, type of prediction target, and trained model)
The relationship between the state data detected by the detectors 13a to 13g, the type of prediction target, and the learned model will be explained with reference to FIG. As shown in FIG. 5, explanatory variables in machine learning (input data in the inference phase) are rotational power data, shaft power data, rotational power data, heat radiation data, vibration data, and vibration data detected by the detectors 13a to 13g. , as sound data.

The learned model A is applied when the objective variable (output data in the inference phase) in machine learning is the state data of the process-altered layer. When the target variable in machine learning is chatter state data, trained model B is applied. When the objective variable in machine learning is roundness data, the learned model C is applied. When the objective variable in machine learning is surface roughness data, the learned model D is applied.

(6. Degree of influence on the state of the processed damaged layer depending on the type of state data)
Next, the degree of influence on the state of the process-affected layer depending on the type of state data will be explained with reference to FIG. 6. FIG. 6 shows the degree of influence of the state data detected by each of the detectors 13a to 13g on the process-affected layer.

As shown in FIG. 6, it can be seen that the heat radiation data detected by the heat flow sensor 13d has the greatest influence on the process-affected layer. From this, it can be said that it is effective to use heat radiation data in order to predict the state of the process-affected layer.

(7. Behavior of heat radiation data)
Next, the behavior of heat radiation data during grinding will be described with reference to FIG. In FIG. 7, vibration data generated on the support shaft that supports the grinding wheel T is also illustrated.

The vibration data shows a high value as soon as the grinding process starts, and shows a low value as soon as the grinding process ends. Furthermore, like the vibration data, the heat radiation data also shows a high value as soon as the grinding process starts, and shows a low value as soon as the grinding process ends. Thus, it can be seen that the heat radiation data has very good followability to the grinding process. This shows that the heat radiation data is affected by the heat generated by the grinding process. It is thought that this good followability improves the accuracy of predicting the state of the process-affected layer.

(8. Effect)
Heat generated at the grinding point during grinding is transferred to the workpiece, grinding wheel, chips, and coolant. The heat flow sensor 13d detects heat radiation data from a position different from the grinding point Pa on the surface of at least one of the grinding wheel T and the workpiece W. That is, the heat flow sensor 13d does not directly detect the temperature at the grinding point Pa. However, since the heat generated at the grinding point Pa is transferred to the workpiece W and the grinding wheel T, heat remains on the surfaces of the workpiece W and the grinding wheel T even if the surfaces are different from the grinding point Pa. ing. Therefore, the heat flow sensor 13d can detect heat radiation data corresponding to the heat generated at the grinding point Pa even if the position is different from the grinding point Pa on the surface of at least one of the grinding wheel T and the workpiece W. . As a result, by using the heat radiation data detected by the heat flow sensor 13d, the state of the process-affected layer can be predicted with high accuracy.

1: Machining quality prediction system, 10: Grinding machine, 11: Grinding machine body, 12: Control device, 13, 13a-13g: Detector, 20: Arithmetic device, 30: Display device, 45: Grinding head, 48: Coolant Nozzle, 50: processing quality inspection device, 60: learning processing device, 61: training data set acquisition unit, 61a: state data acquisition unit, 61b: feature amount calculation unit, 61c: processing quality data acquisition unit, 62: training data set Storage unit, 63: Model generation unit, 64: Effective feature amount extraction unit, 70: Prediction calculation device, 71: Model storage unit, 72: Prediction data acquisition unit, 72a: State data acquisition unit, 72b: Effective feature amount calculation 73: Prediction section, 75: Workpiece post-processing section, 76: Improvement processing section, Pa: Grinding point, Pb: Lower end of grinding wheel, Pc: Upper end of grinding wheel, Pd: Lower end of workpiece, Pe: Machining Upper end of object, T: Grinding wheel, W: Workpiece

Claims (12)

a grinding machine body that grinds the workpiece using the grinding wheel while rotating the workpiece and the grinding wheel;
a detector that detects observable state data on the grinding machine body during grinding of the workpiece;
Using the state data detected by the detector , it is possible to determine the presence or absence of a machining-altered layer that is generated on the workpiece due to an increase in the temperature of the grinding point where the grinding wheel and the workpiece contact, and the machining-altered layer. a prediction unit that predicts the depth of the layer, the surface hardness of the workpiece, or the burnt color of the surface of the workpiece ;
Equipped with
The detector includes a heat flow sensor that detects heat flow data that is the magnitude of heat flux in a predetermined direction as the state data,
The heat flow sensor is disposed opposite to a predetermined position different from the grinding point on the outer circumferential surface of at least one of the grinding wheel and the workpiece , and is arranged in the predetermined direction, which is a direction from the predetermined position toward the heat flow sensor. A processing quality prediction system that detects the magnitude of heat flux . The processing quality prediction system further includes:
The state data detected by the detector is used as an explanatory variable, and the presence or absence of the machining - affected layer generated on the workpiece , the depth of the machining-affected layer, the surface hardness of the workpiece, or the workpiece a learning processing device that generates a trained model by performing machine learning using altered layer data representing the burnt color of the surface of the object as an objective variable and using a training data set including the explanatory variable and the objective variable;
a trained model storage unit that stores the trained model generated by the learning processing device;
Equipped with
The prediction unit uses the learned model and the state data to determine the presence or absence of the work-affected layer, the depth of the work-affected layer, the surface hardness of the workpiece, or the burnt color of the workpiece surface. The processing quality prediction system according to claim 1, which predicts. The prediction unit predicts whether the depth of the work-affected layer, the surface hardness of the workpiece, or the burnt color of the surface of the workpiece is equal to or higher than a predetermined threshold. Machining quality prediction system. The grinding machine body performs the grinding process while supplying coolant from a position above the grinding point toward the grinding point,
The processing according to any one of claims 1 to 3 , wherein the heat flow sensor is disposed on an outer circumferential surface of at least one of the grinding wheel and the workpiece, facing the predetermined position above the grinding point. Quality prediction system. The rotation direction of the grinding wheel is a direction in which the outer peripheral surface of the grinding wheel moves in the order of the side of the grinding wheel, which is the grinding point, the lower end, the upper end, and the grinding point,
The machining quality prediction system according to claim 4 , wherein the heat flow sensor is disposed on the outer peripheral surface of the grinding wheel to face the predetermined position above the grinding point. The processing quality prediction system according to claim 5 , wherein the heat flow sensor is disposed on the outer circumferential surface of the grinding wheel to face the predetermined position between the grinding point and the upper end of the grinding wheel. The rotation direction of the workpiece is a direction in which the outer peripheral surface of the workpiece moves in the order of the side of the workpiece, which is the grinding point, the lower end, the upper end, and the grinding point,
The machining quality prediction system according to claim 4 , wherein the heat flow sensor is disposed on the outer peripheral surface of the workpiece so as to face the predetermined position above the grinding point. The learning processing device includes:
A plurality of feature quantities related to the state data are used as first explanatory variables, and represent the presence or absence of the process-affected layer, the depth of the process-affected layer, the surface hardness of the workpiece, or the burnt color of the surface of the workpiece. By using the altered layer data as a first objective variable and performing machine learning using a first training data set including the first explanatory variable and the first objective variable, the first Extract effective features that have a high degree of influence on the objective variable,
The extracted effective feature amount is used as a second explanatory variable, and the alteration represents the presence or absence of the process-affected layer, the depth of the process-affected layer, the surface hardness of the workpiece, or the burnt color of the surface of the workpiece . generating the learned model by performing machine learning using the layer data as a second objective variable and using a second training data set including the second explanatory variable and the second objective variable;
The prediction unit uses the learned model and the effective feature amount related to the state data to predict the presence or absence of the work-affected layer, the depth of the work-affected layer, the surface hardness of the workpiece, or the workpiece. The processing quality prediction system according to claim 2, which predicts the burnt color of the surface of the workpiece . The detector further includes:
a rotational power sensor that detects rotational power data of at least one of the grinding wheel and the workpiece as the state data;
a shaft power sensor that detects shaft power data for moving the shaft in the cutting direction between the grinding wheel and the workpiece as the state data;
a processing sound sensor that detects sound data generated by the grinding process as the state data;
a vibration sensor that detects vibration data generated in a support device that supports at least one of the grinding wheel and the workpiece as the state data;
The processing quality prediction system according to any one of claims 1 to 8 , comprising at least one of the following. The processing quality prediction system further includes:
If the prediction unit predicts that the work-affected layer exists based on the presence or absence of the work-affected layer, or the depth of the work-affected layer, the surface hardness of the workpiece, or the Any one of claims 1 to 9 , further comprising a workpiece post-processing unit that performs any one of sorting, discarding, and detailed inspection of the target workpiece when it is predicted that the burnt color of the surface is equal to or higher than a predetermined threshold value. The processing quality prediction system according to item 1. The processing quality prediction system further includes:
If the prediction unit predicts that the work-affected layer exists based on the presence or absence of the work-affected layer, or the depth of the work-affected layer, the surface hardness of the workpiece, or the Any one of claims 1 to 10 , further comprising an improvement processing unit that changes grinding processing conditions or executes truing of the grinding wheel when it is predicted that the burnt color of the surface is equal to or higher than a predetermined threshold. Processing quality prediction system described in . The learning processing device determines the presence or absence of the machining-affected layer, the depth of the machining-affected layer, the surface hardness of the workpiece, or the burnt color of the surface of the workpiece , as well as the chatter state occurring in the workpiece. generating the learned model using at least one of data, roundness data, and surface roughness data as the objective variable;
The prediction unit uses the learned model and the state data to predict the presence or absence of the work-affected layer, the depth of the work-affected layer, the surface hardness of the workpiece, or the burnt surface of the workpiece. 3. The processing quality prediction system according to claim 2, which predicts color and also predicts at least one of a chatter state, roundness, and surface roughness occurring in the workpiece.

Images