Machining state detection device, machining state detection method, program, dicing device, and learning model generation method

Abstract

To provide a machining state detection device, a machining state detection method, a program, a dicing device, and a learning model generation method capable of detecting the machining state in response to changes in a plurality of interrelated temperature factors.SOLUTION: A machining state detection device (140): acquires (150) an environmental temperature history, a blade supply water temperature history, and a heating element supply water temperature history; and derives (152) an environmental temperature history feature amount, a blade supply water temperature history feature amount, and a heating element supply water temperature history feature amount. When a machining error is predicted (154) on the basis of a temperature history feature amount including the environmental temperature history feature amount, the blade supply water temperature history feature amount, and the heating element supply water temperature history feature amount, a learned learning model (158) is applied, which is a trained learning model that has learned the temperature history feature amount and the machining error as learning data, and which outputs a predicted value of the machining error when the temperature history feature amount is input.SELECTED DRAWING: Figure 5

Description

[Technical field]

[0001] The present invention relates to a processing state detection device, a processing state detection method, a program, a dicing device, and a learning model generation method. [Background technology]

[0002] Dicing machines that process workpieces are required to achieve high-precision processing in a wide range of environments. In order to achieve high-precision processing in the dicing process, it is very important to manage the installation environment of the machine, such as room temperature and water temperature. For example, in an environment where the room temperature changes greatly, there is a concern that the processing accuracy will decrease due to the thermal expansion and thermal contraction of each unit equipped in the machine.

[0003] Generally, dicing machines do not guarantee processing accuracy over the entire temperature range, but rather have a specified operating temperature range. The specified processing accuracy of a dicing machine is guaranteed when it is operated within the specified operating temperature range.

[0004] Patent Document 1 describes a dicing device that uses a temperature sensor to monitor the ambient temperature near the processing position and corrects at least one of the feed rate of the spindle movement mechanism and the feed rate of the work table based on the monitored shaft temperature. The device described in this document is equipped with various data maps such as the displacement characteristics of the spindle over time, estimates the positional deviation of the processing point in accordance with the data maps, and performs cutting and dividing work while correcting the feed rate of the work. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] JP 2015-076516 A Summary of the Invention [Problem to be solved by the invention]

[0006] However, due to factors such as the facility environment, some users may find it difficult to use a machine that satisfies the specified operating temperature range. When the machine is used in an environment outside the operating temperature range, corrections such as cutter setting and kerf checking are performed at appropriate times, and it is determined whether the specified machining accuracy is satisfied in the corrected state. The user may set the correction timing and make the final decision on the machining accuracy.

[0007] Also, in cases where the room temperature changes significantly throughout the year, it may be necessary to take measures such as changing the operation of the device according to the time of year when the device is in operation. Furthermore, although correction may be performed by treating disturbance factors such as room temperature and various water temperatures as multiple independent parameters, in most cases multiple temperature factors are involved, making it difficult to achieve accurate correction.

[0008] The device described in Patent Document 1 has a correction map for each of a plurality of parameters and is capable of correcting each parameter, but it is difficult to deal with cases in which a plurality of temperature factors are involved.

[0009] The present invention has been made in consideration of the above circumstances, and aims to provide a processing state detection device, a processing state detection method, a program, a dicing device, and a learning model generation method that are capable of detecting the processing state in response to changes in multiple interrelated temperature factors. [Means for solving the problem]

[0010] A machining state detection device according to a first aspect of the present disclosure is a machining state detection device that detects a machining state when machining a workpiece using a blade, and includes an environmental temperature history acquisition unit that acquires an environmental temperature history representing the temperature history of the machining environment, a blade supply water temperature history acquisition unit that acquires a blade supply water temperature history representing the temperature history of the blade supply water supplied to the blade, a heating element supply water temperature history acquisition unit that acquires a heating element supply water temperature history representing the temperature history of the heating element supply water supplied to a heating element that generates heat when machining the workpiece, a temperature history feature derivation unit that derives environmental temperature history feature amounts that represent characteristics of the environmental temperature history, blade supply water temperature history feature amounts that represent characteristics of the blade supply water temperature history, and heating element supply water temperature history feature amounts that represent characteristics of the heating element supply water temperature history, and a machining error prediction unit that predicts machining errors based on temperature history feature amounts including the environmental temperature history feature amounts, the blade supply water temperature history feature amounts, and the heating element supply water temperature history feature amounts, wherein the machining error prediction unit is a learned learning model that has learned using the temperature history feature amounts and the machining errors as learning data, and the machining state detection device is applied with a learning model that outputs a machining error when the temperature history feature amounts are input.

[0011] According to the machining state detection device of the present disclosure, the machining state is detected in response to the change in the environmental temperature, the change in the temperature of the water supplied to the blade, and the change in the temperature of the water supplied to the heating element, which are all related to each other, and correction can be made according to the temperature history.

[0012] The blade supply water may include cutting water, blade cooling water, and cleaning water. The heating element supply water may include heating element cooling water. The heating element may include a spindle, a camera, and a support member that supports these.

[0013] A machining state detection device according to a second aspect is the machining state detection device of the first aspect, and includes a machining state history acquisition unit that acquires a machining state history representing a history of a machining state, and a machining state history feature amount derivation unit that derives a feature amount of the machining state history, and the machining error prediction unit may predict a machining error by taking the machining state history feature amount into account in the temperature history feature amount.

[0014] According to this aspect, the machining state taking into account the fluctuations in the actual machining state is detected, and correction can be performed in accordance with the fluctuations in the actual machining state.

[0015] A machining state detection device according to a third aspect is a machining state detection device according to the first aspect, wherein the temperature history feature derivation unit acquires, as the environmental temperature history feature, the temperature of the machining environment and the amount of change in the temperature of the machining environment per unit time, as the blade supply water temperature history feature, the temperature of the blade supply water and the amount of change in the temperature of the blade supply water per unit time, as the heating element supply water temperature history feature, and the temperature of the heating element supply water and the amount of change in the temperature of the heating element supply water per unit time, and the learning model may be generated by performing learning that uses as inputs a combination of the temperature of the machining environment and the amount of change in the temperature of the machining environment per unit time, a combination of the temperature of the blade supply water and the amount of change in the temperature of the blade supply water per unit time, and a combination of the temperature of the heating element supply water and the amount of change in the temperature of the heating element supply water per unit time, and outputs a machining error.

[0016] According to this aspect, it is possible to detect the machining state in consideration of the amount of change in various temperatures per unit time for various temperatures.

[0017] A machining state detection device according to a fourth aspect is a machining state detection device according to the first aspect, wherein the temperature history feature derivation unit acquires, as environmental temperature history feature values, the temperature of the machining environment, the amount of change in the temperature of the machining environment per unit time, and the direction of change in the temperature of the machining environment, and acquires, as blade supply water temperature history feature values, the temperature of the blade supply water, the amount of change in the temperature of the blade supply water per unit time, and the direction of change in the temperature of the blade supply water, and acquires, as heating element supply water temperature history feature values, the temperature of the heating element supply water, the amount of change in the temperature of the heating element supply water per unit time, and the direction of change in the temperature of the heating element supply water, and the learning model may be generated by performing learning using combinations of the temperature of the machining environment, the amount of change in the temperature of the machining environment per unit time, and the direction of change in the temperature of the machining environment, combinations of the temperature of the blade supply water, the amount of change in the temperature of the blade supply water per unit time, and the direction of change in the temperature of the blade supply water, and combinations of the temperature of the heating element supply water, the amount of change in the temperature of the heating element supply water per unit time, and the direction of change in the temperature of the heating element supply water, and

[0018] According to this aspect, it is possible to detect the machining state taking into consideration the amount of change in various temperatures per unit time and the direction of change in various temperatures.

[0019] A machining state detection device according to a fifth aspect is a machining state detection device according to the first aspect, wherein the temperature history feature derivation unit acquires the temperature of the machining environment and the accumulated temperature of the machining environment as the environmental temperature history feature, acquires the temperature of the blade supply water and the accumulated temperature of the blade supply water as the blade supply water temperature history feature, and acquires the temperature of the heating element supply water and the accumulated temperature of the heating element supply water as the heating element supply water temperature history feature, and the learning model may be generated by performing learning that uses as inputs a combination of the temperature of the machining environment and the accumulated temperature of the machining environment, a combination of the temperature of the blade supply water and the accumulated temperature of the blade supply water, and a combination of the temperature of the heating element supply water and the accumulated temperature of the heating element supply water, and outputs a machining error.

[0020] According to this aspect, it is possible to detect the machining state in consideration of the integration of various temperatures for various temperatures.

[0021] A machining state detection device according to a sixth aspect may be provided in a machining state detection device according to any one of the first to fifth aspects, further comprising a machining condition acquisition unit that acquires machining conditions to be applied to machining, and a learning model selection unit that selects a learning model to be applied to the machining error prediction unit according to the machining conditions from among a plurality of first learning models generated by performing learning for each machining condition.

[0022] According to this aspect, detection of the machining state is realized to which a learning model according to the machining conditions is applied.

[0023] A machining state detection method according to a seventh aspect of the present disclosure is a machining state detection method for detecting a machining state when machining an object to be machined using a blade, in which a control device includes an environmental temperature history acquisition step of acquiring an environmental temperature history representing the temperature history of the machining environment, a blade supply water temperature history acquisition step of acquiring a blade supply water temperature history representing the temperature history of blade supply water supplied to the blade, a heating element supply water temperature history acquisition step of acquiring a heating element supply water temperature history representing the temperature history of heating element supply water supplied to a heating element that generates heat when machining the object to be machined, and an environmental temperature history feature amount representing a feature of the environmental temperature history, The machining error prediction step executes a temperature history feature derivation step of deriving blade supply water temperature history feature values ​​which represent characteristics of the blade supply water temperature history and heating element supply water temperature history feature values ​​which represent characteristics of the heating element supply water temperature history, and a machining error prediction step of predicting a machining error based on temperature history feature values ​​including environmental temperature history feature values, blade supply water temperature history feature values, and heating element supply water temperature history feature values, wherein the machining error prediction step is a trained learning model which has trained using the temperature history feature values ​​and the machining error as learning data, and the learning model which outputs a machining error when the temperature history feature values ​​are input is applied.

[0024] According to the machining state detection method of the present disclosure, it is possible to obtain the same effects as those of the machining state detection device of the present disclosure.

[0025] In the machining state detection method according to the present disclosure, the same items as those specified in any one of the second to sixth aspects can be appropriately combined. In this case, the components performing the processes or functions specified in the machining state detection device can be understood as the components of the machining state detection method performing the corresponding processes or functions.

[0026] A program according to an eighth aspect of the present disclosure is a program for detecting a machining state when machining an object to be machined using a blade, wherein a control device realizes an environmental temperature history acquisition function for acquiring an environmental temperature history representing the temperature history of the machining environment, a blade supply water temperature history acquisition function for acquiring a blade supply water temperature history representing the temperature history of blade supply water supplied to the blade, a heating element supply water temperature history acquisition function for acquiring a heating element supply water temperature history representing the temperature history of heating element supply water supplied to a heating element that generates heat when machining an object to be machined, a temperature history feature derivation function for deriving environmental temperature history feature amounts representing characteristics of the environmental temperature history, blade supply water temperature history feature amounts representing characteristics of the blade supply water temperature history, and heating element supply water temperature history feature amounts representing characteristics of the heating element supply water temperature history, and a machining error prediction function for predicting machining errors based on temperature history feature amounts including the environmental temperature history feature amounts, the blade supply water temperature history feature amounts, and the heating element supply water temperature history feature amounts, wherein the machining error prediction function is a learned learning model that has been trained using temperature history feature amounts and machining errors as learning data, and the learning model that outputs a machining error when a temperature history feature amount is input is applied.

[0027] According to the program of the present disclosure, it is possible to obtain the same effects as those of the machining state detection device of the present disclosure.

[0028] In the program according to the present disclosure, the same items as those specified in any one of the second to sixth aspects may be appropriately combined. In this case, the components performing the processes or functions specified in the machining state detection device may be understood as the components of the program performing the corresponding processes or functions.

[0029] A dicing apparatus according to a ninth aspect of the present disclosure is a dicing apparatus including a processing unit that processes a workpiece using a blade, and a processing state detection unit that detects a processing state of the workpiece, the processing state detection unit including an environmental temperature history acquisition unit that acquires an environmental temperature history representing the temperature history of the processing environment, a blade supply water temperature history acquisition unit that acquires a blade supply water temperature history representing the temperature history of the blade supply water supplied to the blade, a heating element supply water temperature history acquisition unit that acquires a heating element supply water temperature history representing the temperature history of heating element supply water supplied to a heating element that generates heat when processing the workpiece, and a temperature history acquisition unit that acquires a characteristic of the environmental temperature history. the temperature history feature derivation unit deriving an environmental temperature history feature, a blade supply water temperature history feature representing a feature of the blade supply water temperature history, and a heating element supply water temperature history feature representing a feature of the heating element supply water temperature history; and a machining error prediction unit predicting a machining error based on temperature history feature including the environmental temperature history feature, the blade supply water temperature history feature, and the heating element supply water temperature history feature. The machining error prediction unit is a trained learning model that has trained using the temperature history feature and the machining error as learning data, and the dicing device is applied to the learning model that outputs a machining error when the temperature history feature is input.

[0030] According to the dicing apparatus of the present disclosure, it is possible to obtain the same operational effects as those of the processed state detection apparatus of the present disclosure.

[0031] In the dicing device according to the present disclosure, the same items as those specified in any one of the second to sixth aspects can be appropriately combined. In that case, the components performing the processes or functions specified in the processing state detection device can be understood as the components of the dicing device performing the corresponding processes or functions.

[0032] A dicing apparatus according to a tenth aspect may be the dicing apparatus according to the ninth aspect, further comprising a determination section that determines whether or not correction of the processing section is required, based on the processing error output from the processing error prediction section.

[0033] According to this aspect, it is possible to determine whether or not correction of the processed portion is required based on the processing error.

[0034] A dicing apparatus according to an eleventh aspect is a dicing apparatus according to the tenth aspect, wherein the processing error prediction unit predicts a processing error after a specified period of time has elapsed, and the judgment unit derives a judgment result that correction of the processing unit is necessary when the predicted processing error exceeds a specified threshold value.

[0035] According to this aspect, it is possible to determine whether or not correction of the processed portion is necessary based on a prediction of future processing errors.

[0036] A dicing apparatus according to a twelfth aspect is a dicing apparatus according to the eleventh aspect, wherein the processing error prediction unit predicts the processing error after a specified period of time has elapsed based on the amount of change in the processing error per unit time and the direction of the change in the processing error.

[0037] According to this aspect, it is possible to determine whether or not correction of the processed portion is necessary based on a prediction of the processing error that takes into account the direction of change in the processing error.

[0038] A learning model generation method according to a thirteenth aspect of the present disclosure is a learned learning model that has been trained using temperature history features including an environmental temperature history representing the temperature history of the processing environment, a blade supply water temperature history representing the temperature history of blade supply water supplied to a blade that processes the workpiece, and a heating element supply water temperature history representing the temperature history of heating element supply water supplied to a heating element that generates heat when processing the workpiece, as learning data, and a processing error when processing the workpiece, and is a learning model that outputs a processing error when the temperature history features are input.

[0039] According to the learning model generation method of the present disclosure, a trained learning model that can be applied to the processing state detection device, processing state detection method, program, and dicing device of the present disclosure can be generated. Effect of the Invention

[0040] According to the present invention, the machining state is detected with respect to the change in the environmental temperature, the change in the temperature of the water supplied to the blade, and the change in the temperature of the water supplied to the heating element, which are all related to each other, and correction can be made according to the temperature history. [Brief description of the drawings]

[0041] [Figure 1] FIG. 1 is a perspective view showing a schematic configuration of a dicing device according to an embodiment. [Diagram 2] FIG. 2 is a perspective view showing a schematic configuration of the processing unit shown in FIG. [Diagram 3] FIG. 3 is a perspective view showing an example of the configuration of the tip of the spindle. [Figure 4] FIG. 4 is a functional block diagram showing the electrical configuration of the dicing apparatus shown in FIG. [Diagram 5] FIG. 5 is a functional block diagram showing an electrical configuration of the machining state detection unit shown in FIG. [Figure 6] FIG. 6 is a flowchart showing the flow of the machining state detection method according to the embodiment. [Figure 7] FIG. 7 is a flowchart showing a procedure of processing applied to the analysis unit shown in FIG. [Figure 8] FIG. 8 is a schematic diagram of a learning model applied to the machining error prediction unit shown in FIG. [Figure 9] FIG. 9 is an explanatory diagram of the temperature history. [Figure 10] FIG. 10 is an explanatory diagram of the amount of change in temperature per unit time. [Figure 11] FIG. 11 is a schematic diagram of a learning model generation method. [Figure 12] FIG. 12 is an explanatory diagram of a first modified example of the temperature history feature amount. [Figure 13] FIG. 13 is an explanatory diagram of a second modified example of the temperature history feature amount. [Figure 14] FIG. 14 is an explanatory diagram of a third modified example of the temperature history feature amount. [Figure 15] FIG. 15 is a schematic diagram of a method for generating a learning model in which a temperature history feature amount according to the third modified example is used as learning data. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0042] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In this specification, the same components are given the same reference numerals, and duplicated descriptions will be omitted as appropriate.

[0043] [Configuration example of a dicing device according to the embodiment] Fig. 1 is a perspective view showing a schematic configuration of a dicing device according to an embodiment. The X and Y directions shown in Fig. 1 are mutually orthogonal and parallel to the surface on which the dicing device is placed. The Z direction is orthogonal to the X and Y directions and parallel to the direction orthogonal to the surface on which the dicing device is placed.

[0044] The dicing device 10 shown in FIG. 1 is a dicing device called a twin spindle dicer in which a pair of blades 12, 12 are arranged opposite to each other. The dicing device 10 includes a processing unit 18 including a pair of blades 12, 12, a pair of spindles 14, 14 with a built-in high-frequency motor and the blade 12 attached to the tip, and a work table 16 on which a workpiece W, which is an object to be processed, is placed and which suction-holds the workpiece W. The processing unit 18 moves the workpiece W and the blade 12 relatively on a plane parallel to the XY plane, and cuts the workpiece W using the blade 12. The cutting process may be referred to as a dicing process or the like. Examples of the material of the workpiece W include semiconductor materials such as silicon and silicon carbide. Other examples of the material of the workpiece W include materials other than semiconductor materials such as sapphire, glass, and quartz.

[0045] The dicing apparatus 10 includes a camera 19 that captures an image of the surface of the workpiece W, a cleaning unit 20 that spin-cleans the machined workpiece W, a load port 22 on which a cassette containing a plurality of workpieces W is placed, and a transport device 24 that transports the workpieces W. Each of the camera 19 and the like is disposed at a prescribed position. The dicing apparatus 10 includes a control device 26 that controls the overall operation of each part of the dicing apparatus 10. The control device 26 is disposed inside the dicing apparatus 10.

[0046] The control device 26 is an example of the device control means of the present invention, and is implemented by a computer. The computer that functions as the control device 26 includes one or more processors and one or more computer-readable media. The computer realizes various functions of the dicing device 10 by causing the processor to execute a program including one or more instructions stored in the computer-readable medium. The electrical configuration of the control device 26 will be described in detail later.

[0047] The dicing apparatus 10 includes a display unit 46. The display unit 46 is electrically connected to the control device 26, and displays various information such as the results of the dicing process, the state of the blade 12 such as partial loss, and various data. A display device with a touch panel can be used as the display unit 46. An operator of the dicing apparatus 10 can use the touch panel to input various information such as settings of processing conditions for the control device 26 and threshold values ​​for determining the state of the blade 12.

[0048] [Example of processing section configuration] 2 is a perspective view showing a schematic configuration of the processing unit shown in FIG. The processing unit 18 shown in the figure includes an X-table 34. The X-table 34 is guided by a pair of X-guides 30, 30 provided on an X-base 28, and is driven in the X-direction by a linear motor 32. A rotary table 36 that rotates in the θ-direction is erected on the upper surface of the X-table 34. A work table 16 is provided on the rotary table 36. The work table 16 moves in the X-direction by using the X-table 34, and rotates in the θ-direction by using the rotary table 36. The θ-direction represents the direction of rotation around a rotation axis that extends in a direction parallel to the Z-direction.

[0049] The processing unit 18 includes a gate-shaped Y base 38 that is disposed at a position straddling the X base 28. A pair of Y tables 42, 42 are provided on a front surface 38A of the Y base 38. The pair of Y tables 42, 42 are guided by a pair of Y guides 40, 40 fixed to the front surface 38A of the Y base 38, and are driven in the Y direction by a Y drive device. The Y drive device includes a motor and a feed screw device.

[0050] A pair of Z tables 44, 44 are provided on each of the pair of Y tables 42, 42. The Z table 44 is guided by a Z guide provided on the Y table 42, and is driven in the Z direction by a Z drive device. The Z drive device may have a similar configuration to the Y drive device. Note that illustration of the Y drive device, Z guide, and Z drive device is omitted.

[0051] A pair of spindles 14, 14 are fixed to each of the pair of Z tables 44, 44 in an opposing state. A pair of blades 12, 12 attached to the respective tips of the pair of spindles 14, 14 are disposed facing each other along the Y direction.

[0052] Each of the pair of blades 12, 12 is indexed in the Y direction and cut in the Z direction. The work table 16 is cut in the X direction and rotates in the θ direction. These operations are controlled by the control device 26 shown in FIG.

[0053] FIG. 3 is a perspective view showing a configuration example of the tip of the spindle. A wheel cover 54 that covers the blade 12 is attached to the tip of the spindle 14 shown in FIG. 3. The wheel cover 54 includes a front cover portion 56, a rear cover portion 58, a nozzle block 60, and a guide block 72. A hose 62 is connected to the rear cover portion 58. Cutting water supplied through the hose 62 is sprayed from a nozzle 63 provided on the wheel cover 54 toward the rotating blade 12. A hose 64 is connected to the nozzle block 60. Blade cooling water supplied through the hose 64 is sprayed from a pair of nozzles 66, 66 provided on the wheel cover 54 toward a processing point of the workpiece W that is processed using the blade 12.

[0054] Although not shown in the figures, the dicing apparatus 10 uses heater cooling water to cool the heaters inside the apparatus, such as the spindle 14, the microscope including the camera 19, the Z-axis structure, and the scale support unit. That is, the dicing apparatus 10 includes a second cooling water supply unit that supplies heater cooling water to the heaters other than the blade 12. The dicing apparatus 10 also includes a cleaning water supply unit that supplies cleaning water.

[0055] The flow paths of blade supply water, including cutting water, blade cooling water, and cleaning water, form an open flow path that branches off to each supply destination and is discharged into the device. On the other hand, the heating element cooling water is industrial water such as city water that flows to cool heating elements such as the spindle motor, the microscope (camera 19), the Z-axis structure, the scale support, and other heating elements and parts that need to be kept at the same temperature (Z-axis structure, scale support), and the flow path of the heating element cooling water forms a closed flow path.

[0056] The dicing apparatus 10 is equipped with a blade breakage detection device 48 that detects partial damage that occurs on the cutting edge 12A of the blade 12. The control device 26 provided in the dicing apparatus 10 grasps the state of the blade 12 based on the detection result of the blade breakage detection device 48.

[0057] The blade breakage detection device 48 includes an optical detection unit 50. The optical detection unit 50 is attached to the wheel cover 54 at the center position of the wheel cover 54 in the X direction, and is disposed at a position opposite in the Z direction to the processing position of the blade 12. The optical detection unit 50 includes a light-projecting unit 68 and a light-receiving unit 70. A light-projecting wiring 80 is connected to the light-projecting unit 68, and a light-receiving wiring 84 is connected to the light-receiving unit 70.

[0058] The light-projecting section 68 and the light-receiving section 70 included in the light detection unit 50 are disposed at positions facing each other across the blade 12 in the Y direction. The light-projecting section 68 includes a light-projecting element such as an LED. The light-projecting section 68 may include an LED array including a plurality of LEDs. The light-projecting element is supplied with power from a power supply device via light-projection side wiring 80. Note that an illustration of the power supply device is omitted.

[0059] The light receiving unit 70 receives at least a portion of the light emitted from the light projecting unit 68 toward the blade 12. The light receiving unit 70 includes a light receiving element such as a photodiode. The light receiving unit 70 may include a photodiode array including a plurality of photodiodes. The light receiving element outputs a detection signal having a current value according to the amount of received light. The detection signal output from the light receiving element is transmitted to the control device 26 via the light receiving side wiring 84.

[0060] [Electrical configuration of the dicing device] Fig. 4 is a functional block diagram showing the electrical configuration of the dicing apparatus shown in Fig. 1. The control device 26 includes a system control unit 100. The system control unit 100 comprehensively controls various control units included in the dicing apparatus 10. The system control unit 100 controls writing of data, etc. to the computer-readable medium 120 and reading of data, etc. from the computer-readable medium 120.

[0061] The control device 26 includes a processing control unit 102. The processing control unit 102 controls the operation of each unit of the processing unit 18 based on a command signal transmitted from the system control unit 100. The processing control unit 102 functions as a drive control unit that controls the operation of the X drive unit, Y drive unit, Z drive unit, and θ drive unit provided in the processing unit 18. The processing control unit 102 also functions as a spindle control unit that controls the operation of the spindle 14.

[0062] The control device 26 includes an image capturing data acquisition unit 104. The image capturing data acquisition unit 104 acquires image capturing data of the workpiece W captured and generated using the camera 19. The image capturing data of the workpiece W acquired using the image capturing data acquisition unit 104 is used to grasp processing errors of the workpiece W during alignment, kerf checking, and the like.

[0063] The control device 26 includes a blade supply water temperature acquisition unit 106, a heating element supply water temperature acquisition unit 108, and a room temperature acquisition unit 110. The blade supply water temperature acquisition unit 106 acquires the temperatures of the cutting water, blade cooling water, and washing water detected using a blade supply water temperature detection unit 130. The blade supply water temperature acquisition unit 106 acquires time-series data of the temperatures of the cutting water, etc., by applying a specified sampling period.

[0064] The heating element supply water temperature acquisition unit 108 acquires the temperature of the heating element cooling water detected using the heating element supply water temperature detection unit 132. The heating element supply water temperature acquisition unit 108 acquires time-series data of the temperature of the heating element cooling water by applying a specified sampling period.

[0065] The room temperature acquisition unit 110 detects the room temperature detected by the room temperature detection unit 134. The room temperature is understood as the environmental temperature in the environment in which the dicing apparatus 10 is installed. The room temperature acquisition unit 110 acquires time-series data of the room temperature by applying a specified sampling period. The sampling periods applied to the blade supply water temperature acquisition unit 106, the heating element supply water temperature acquisition unit 108, and the room temperature acquisition unit 110 may be the same or different from each other. The blade supply water temperature detection unit 130, the heating element supply water temperature detection unit 132, and the room temperature detection unit 134 are each equipped with a temperature sensor. The temperature sensor may be of a contact type or a non-contact type.

[0066] The control device 26 includes a liquid supply control unit 112. The liquid supply control unit 112 controls the operation of the liquid supply device 136 based on a command signal transmitted from the system control unit 100. In other words, the liquid supply control unit 112 controls the supply timing and amount of the blade supply water and the heating element supply water.

[0067] The liquid supply device 136 may include a blade supply water supply unit that supplies blade supply water, and a heating element supply water supply unit that supplies heating element supply water. The blade supply water supply unit may include a cutting water supply unit that supplies cutting water, a blade cooling water supply unit that supplies blade cooling water, and a cleaning water supply unit that supplies cleaning water.

[0068] The liquid supply control unit 112 may include a blade supply water supply control unit and a heating element supply water supply control unit. The blade supply water supply control unit may include a cutting water supply control unit, a blade cooling water supply control unit, and a cleaning water supply control unit.

[0069] The control device 26 includes a blade information acquisition unit 114. The blade information acquisition unit 114 acquires the detection result of the blade 12 output from the blade breakage detection device 48. The control device 26 determines whether or not there is partial damage to the blade 12 based on the detection result of the blade 12.

[0070] The control device 26 includes a machining state acquisition unit 116. The machining state acquisition unit 116 acquires parameters representing the actual machining state, such as the rotation speed of the spindle 14. The machining state acquisition unit 116 may acquire parameters representing the actual machining state based on the machining conditions set using the machining control unit 102. The machining state acquisition unit 116 may acquire the time since the machining state changed. For example, the time from the start of the supply of the heating element cooling water and the time from the start of machining of each channel may be acquired. In addition, when the rotation speed of the spindle 14 is changed, the cutter set during machining (wear amount measurement) and the kerf check are performed, the time from the change in the rotation speed of the spindle 14 and the time from the implementation of the cutter set during machining may be acquired.

[0071] The control device 26 includes a machining state detection unit 140. The machining state detection unit 140 predicts a machining error of the workpiece W based on various temperature information acquired during machining of the workpiece W and the actual machining state. The machining state detection unit 140 may predict a machining error after a prescribed period has elapsed based on the amount of change in the machining error per unit time and the direction of the amount of change. The machining state detection unit 140 predicts a dangerous state in which the machining error of the workpiece W is likely to become large.

[0072] The machining state detection unit 140 can function as a judgment unit that judges whether or not correction of the machining unit 18 is necessary when a dangerous state in which the machining error of the workpiece W is likely to increase is predicted. The machining state detection unit 140 may derive a judgment result that correction of the machining unit 18 is necessary when the predicted machining error exceeds a specified threshold value.

[0073] The control device 26 includes a learning model storage unit 164. The learning model storage unit 164 stores a learning model used for predicting a machining error applied to the machining state detection unit 140. The prediction of the machining error will be described in detail later.

[0074] One or more processors are applied to the hardware of various processing units such as the system control unit 100 shown in Fig. 4. One processing unit or two or more processing units may be realized using one processor, and one processing unit may be realized using multiple processors.

[0075] The processor may be a general-purpose processing device such as a central processing unit (CPU), or a processing device specialized for a specific process. The multiple processors may be of the same type, or may be of different types.

[0076] The control device 26 includes a computer-readable medium 120. The computer-readable medium 120, which is a non-transitory tangible entity, includes a memory 122 that is a primary storage device and a storage 124 that is a secondary storage device. The computer-readable medium 120 may use a semiconductor memory, a hard disk device, a solid-state drive device, or the like. The computer-readable medium 120 may use any combination of multiple devices.

[0077] A hard disk device may be referred to as an HDD, which is an abbreviation of the English term Hard Disk Drive, and a solid state drive device may be referred to as an SSD, which is an abbreviation of the English term Solid State Drive.

[0078] The control device 26 includes an input / output interface 126 and a communication interface 128. The display unit 46 is connected to the control device 26 via the input / output interface 126. The input / output interface 126 may be compatible with various standards such as USB (Universal Serial Bus).

[0079] The control device 26 performs data communication with an external device via a communication interface 128. The communication form of the communication interface 128 may be either wired communication or wireless communication. The control device 26 may be connected to a network via the communication interface 128 and communicably connected to an external device. The network may be a LAN (Local Area Network) or the like. The network is not shown in the figure.

[0080] Fig. 4 illustrates main components such as a processing section related to processing of the dicing apparatus 10 shown in Fig. 1 etc. The dicing apparatus 10 may include components such as a processing section that are not illustrated in Fig. 4.

[0081] [Example of machining state detection unit configuration] Fig. 5 is a functional block diagram showing an electrical configuration of the machining state detection unit shown in Fig. 4. The machining state detection unit 140 includes a temperature history generation unit 150, a machining state history generation unit 151, a temperature history feature amount derivation unit 152, a machining state history feature amount derivation unit 153, a machining error prediction unit 154, and an analysis unit 156. The machining error prediction unit 154 includes a learned learning model 158.

[0082] The temperature history generating unit 150 generates a temperature history of the blade supply water from the temperature of the blade supply water, such as cutting water, transmitted from the blade supply water temperature acquiring unit 106. The temperature history of the blade supply water is based on time series data of the temperature of the blade supply water.

[0083] Furthermore, the temperature history generating unit 150 generates a temperature history of the heating element coolant from the temperature of the heating element coolant transmitted from the heating element supply water temperature acquiring unit 108. The temperature history of the heating element coolant is updated using time series data of the temperature of the heating element coolant. Furthermore, the temperature history generating unit 150 generates a room temperature history from the room temperature transmitted from the room temperature acquiring unit 110. The room temperature history is updated using time series data of the room temperature.

[0084] The temperature history generation unit 150 described in the embodiment is an example of a blade supply water temperature history acquisition unit that acquires blade supply water temperature history representing the temperature history of blade supply water, and is an example of hardware in which a program realizes the blade supply water temperature history acquisition function.

[0085] In addition, the temperature history generating unit 150 described in the embodiment is an example of a heating element supply water temperature history acquisition unit that acquires the temperature history of heating element supply water supplied to a heating element that generates heat when processing an object to be processed, and is an example of hardware in which a program realizes the heating element supply water temperature history acquisition function.

[0086] Furthermore, the temperature history generating unit 150 described in the embodiment is an example of an environmental temperature history acquiring unit that acquires environmental temperature history that represents the temperature history of the processing environment, and is an example of hardware in which a program realizes an environmental temperature history acquiring function.

[0087] The machining state history generating unit 151 generates a machining state history from an actual machining state acquired by using the machining state acquiring unit 116. An example of the machining state history is time-series data of the rotation speed of the spindle 14. The machining state history generating unit 151 described in the embodiment is an example of a machining state history acquiring unit that acquires a machining state history that represents the history of the machining state.

[0088] The temperature history feature value deriving unit 152 derives temperature history feature values ​​including feature values ​​of the temperature history of the water supplied to the blade, feature values ​​of the temperature history of the water supplied to the heating element, and feature values ​​of the history of the room temperature. For example, the feature values ​​of the temperature history of the water supplied to the blade may be a pair of the temperature of the water supplied to the blade and the slope of the temperature of the water supplied to the blade.

[0089] That is, the feature quantities of the temperature history of the blade supply water include a pair of the cutting water temperature and the slope of the cutting water temperature, a pair of the blade cooling water temperature and the slope of the blade cooling water temperature, and a pair of the wash water temperature and the slope of the wash water temperature. Also, the feature quantities of the temperature history of the heating element supply water include a pair of the heating element cooling water temperature and the slope of the heating element cooling water temperature.

[0090] The gradient of the temperature of the blade supply water is derived as a temperature change per unit time at the sampling timing when the temperature of the blade supply water is acquired. For example, the sampling period of the temperature of the blade supply water may be applied as a unit time, and the gradient of the temperature change of the blade supply water may be derived using the temperatures of the blade supply water acquired at successive sampling timings.

[0091] The feature quantity of the temperature history of the heating element supply water may be a pair of the temperature of the heating element supply water and the slope of the temperature change of the heating element supply water. The slope of the temperature change of the heating element supply water is derived as the temperature change per unit time at the sampling timing when the temperature of the heating element supply water is acquired. Furthermore, the feature quantity of the room temperature history may be a pair of the room temperature and the slope of the room temperature change. The slope of the room temperature change is derived as the temperature change per unit time at the sampling timing when the room temperature is acquired.

[0092] The temperature history feature amount derivation unit 152 described in the embodiment is an example of hardware in which a program realizes a temperature history feature amount derivation function.

[0093] The machining state history feature amount derivation unit 153 derives the feature amount of the machining state history. The feature amount of the machining state history includes a pair of the rotation speed of the spindle 14 and the slope of the rotation speed of the spindle 14. For example, when the rotation speed of the spindle 14 is set to 40 krpm (kilo revolutions per minute) as the machining condition of channel 1 and the rotation speed of the spindle 14 is set to 50 krpm as the machining condition of channel 2, and machining is performed in the order of channel 1 and channel 2, the rotation speed of the spindle 14 changes during machining. Due to the change in the rotation speed of the spindle 14, position fluctuation may occur. In other words, the machining state also has a time constant and can be used to predict machining errors.

[0094] The machining error prediction unit 154 predicts machining errors when machining the workpiece W, based on the feature values ​​of the temperature history and the feature values ​​of the machining state history, using the trained learning model 158. The temperature history here is a general term for the temperature history of the water supplied to the blade, the temperature history of the water supplied to the heating element, and the history of room temperature.

[0095] The learning model 158 is a trained learning model that has been trained to output a predicted value of a processing error when a feature of the temperature history is input, using a pair of the feature of the temperature history and the processing error as training data. That is, in generating the learning model 158, supervised learning is performed using a pair of the feature of the temperature history and the processing error as training data. In other words, the learning model 158 is trained to output a predicted value of a processing error when a feature of the temperature history is input.

[0096] When the feature of the temperature history derived from the temperature acquired at an arbitrary sampling timing is input to the learning model 158, the learning model 158 outputs a predicted value of the processing error at the sampling timing at which the temperature was acquired.

[0097] A neural network such as a deep neural network is applied to the learning model 158. Values ​​derived by learning using software are applied to the weights and biases of the neural network. Note that the deep neural network may be referred to as DNN, which is an abbreviation of Deep Neural Network in English.

[0098] The learning model 158 may be configured as a trained learning model that has been trained by adding a set of the feature amount of the machining state history and the machining error as training data. When the feature amount of the temperature history and the feature amount of the machining state history are input, the learning model 158 outputs a predicted value of the machining error.

[0099] The machining error prediction unit 154 described in the embodiment is an example of hardware in which a program realizes a machining error prediction function.

[0100] The analysis unit 156 predicts the future transition of the machining error during machining based on the prediction result of the machining error output from the learning model 158 included in the machining error prediction unit 154. When the analysis unit 156 predicts the occurrence of a machining error exceeding a specified allowable range, it outputs a signal indicating that fact as the analysis result. When the occurrence of a machining error exceeding a specified allowable range is predicted, the control device 26 can perform machining correction at an appropriate timing.

[0101] The machining state detection unit 140 includes a machining condition acquisition unit 160. The machining condition acquisition unit 160 acquires machining conditions to be applied to the machining unit 18, such as the type of the workpiece W and the type of the blade 12, from the machining control unit 102.

[0102] The machining state detection unit 140 includes a learning model selection unit 162. The learning model selection unit 162 selects a learning model corresponding to the machining conditions acquired by the machining condition acquisition unit 160 from among a plurality of learning models generated by performing learning for each machining condition and stored in the learning model storage unit 164. The learning model selected by the learning model selection unit 162 is applied to the machining error prediction unit 154. The learning model storage unit 164 may be provided in an external device of the control device 26. Note that the plurality of learning models generated by performing learning for each machining condition described in the embodiment are an example of a plurality of first learning models.

[0103] [Processing state detection method procedure] Fig. 6 is a flow chart showing the flow of the machining state detection method according to the embodiment. The machining state detection method shown in Fig. 6 is realized by the control device 26 to which a computer is applied, executing a prescribed program.

[0104] In the history information generating step S10, the temperature history generating unit 150 shown in Fig. 5 generates a temperature history of the water supplied to the blade, a temperature history of the water supplied to the heating element, a room temperature history, and the like. In the history information generating step S10, the machining state history generating unit 151 generates a machining state history. After the history information generating step S10, a history feature value derivation step S12 is executed. Note that the history information generating step S10 described in the embodiment is an example of an environmental temperature history acquisition step, a blade supply water temperature history acquisition step, and a heating element supply water temperature history acquisition step.

[0105] In the history feature amount derivation step S12, the temperature history feature amount derivation unit 152 generates a temperature history feature amount of the water supplied to the heating element, a temperature history feature amount of the water supplied to the heating element, and a history feature amount of the room temperature. In the history feature amount derivation step S12, the machining state history feature amount generation unit 153 generates a feature amount of the machining state history. After the history feature amount derivation step S12, a machining error prediction step S14 is executed. Note that the history feature amount derivation step S12 described in the embodiment is an example of a temperature history feature amount derivation step.

[0106] In the machining error prediction step S14, the machining error prediction unit 154 predicts machining errors corresponding to the cutting water temperature history feature amount, the cooling water temperature history feature amount, and the room temperature history feature amount by applying the learning model 158. The machining error prediction step S14 is understood as an inference step of the learning model 158. The prediction result of the machining error is output to the analysis unit 156 shown in Fig. 5. After the machining error prediction step S14, the process proceeds to an end determination step S16.

[0107] In the end determination step S16, the machining state detection unit 140 determines whether or not to end detection of the machining state. In the end determination step S16, if it is determined that the specified end condition is not satisfied and that machining state detection is to be continued, the result is No. In the case of No, the process proceeds to the history information generation step S10, and each step from the history information generation step S10 to the end determination step S16 is repeatedly executed until the end determination step S16 is Yes.

[0108] On the other hand, in the end determination step S16, if it is determined that the specified end condition is satisfied and the machining state detection is to be ended, a Yes determination is made. If the Yes determination is made, a specified end process is carried out and the machining state detection is ended.

[0109] Prior to the machining error prediction step S14, a machining condition acquisition step for acquiring machining conditions and a learning model selection step for selecting a learning model according to the machining conditions may be executed, and the machining error prediction step S14 may be executed using the selected learning model.

[0110] [Processing procedure of the analysis part] Fig. 7 is a flowchart showing the procedure of the process applied to the analysis unit 156 shown in Fig. 5. In the machining error acquisition step S20 shown in Fig. 7, the analysis unit 156 shown in Fig. 5 acquires the machining error output from the machining error prediction unit 154. Specifically, in the machining error acquisition step S20, the prediction result of the machining error predicted in the machining error prediction step S14 shown in Fig. 6 is acquired. After the machining error acquisition step S20, the process proceeds to the machining error transition prediction step S22.

[0111] In the machining error transition prediction step S22, the analysis unit 156 predicts the future transition of the machining error. That is, in the machining error transition prediction step S22, a dangerous state in which the machining error is likely to increase is predicted. The dangerous state in which the machining error is likely to increase is a state in which an error will occur if the temperature change continues as it is.

[0112] Specifically, in the machining error transition prediction step S22, it is predicted whether the machining error will increase, and if it is predicted that the machining error will increase, it is predicted whether the machining error will exceed the allowable range. After the machining error transition prediction step S22, the process proceeds to the machining error determination step S24.

[0113] In the machining error determination step S24, if the machining error is predicted to remain within the allowable range, a No determination is made, and the process proceeds to an end determination step S28. On the other hand, in the machining error determination step S24, if the machining error is predicted to exceed the allowable range, a Yes determination is made, and the process proceeds to a machining position correction step S26.

[0114] In the processing error determination step S24, if the determination is Yes, an alarm may be output. Examples of the alarm include a mode in which text information is displayed on the display unit 46 shown in Fig. 1, and a mode in which a sound is used.

[0115] In the processing position correction step S26, a kerf check is performed separately from the recipe conditions, the current processing position is checked, and the processing position is corrected based on the current processing position. That is, in the processing position correction step S26, the processing position is checked, which is provided in the dicing device 10, and a function of correcting the processing position when the error of the processing position exceeds the allowable range is applied, and the processing position is corrected in a timely manner according to the recipe conditions. After the processing position correction step S26, the process proceeds to the end determination step S28.

[0116] In the end determination step S28, it is determined whether the analysis process is to be ended. Examples of the end of the analysis process include the end of machining of the workpiece W and end due to the occurrence of an error. If it is determined that the analysis process is to be continued in the end determination step S28, a No determination is made. If a No determination is made, the process proceeds to the machining error acquisition step S20, and each step from the machining error acquisition step S20 to the end determination step S28 is repeatedly executed until a Yes determination is made in the end determination step S28.

[0117] On the other hand, if it is determined in the termination determination step S28 that the analysis process is to be terminated, a Yes determination is made. If the Yes determination is made, a prescribed termination process is carried out and the procedure of the analysis process is terminated.

[0118] [Specific examples of learning models] Fig. 8 is a schematic diagram of a learning model applied to the machining error prediction unit shown in Fig. 5. When the blade supply water temperature history feature value, the heating element supply water temperature history feature value, and the room temperature history feature value are input, the learning model 158 outputs a predicted value of the machining error at the timing when the blade supply water temperature, the heating element supply water temperature, and the room temperature are acquired. A mode in which a score representing the probability of correct answer is assigned to the predicted value of the machining error may be applied. The learning model 158 may output a predicted value of the machining error that takes into account the actual machining state.

[0119] The blade supply water temperature history feature may be a pair of the blade supply water temperature at the timing of acquiring the blade supply water temperature and the amount of change in the blade supply water temperature per unit time at the timing of acquiring the blade supply water temperature.

[0120] The heating element supply water temperature history characteristic may be a pair of the heating element supply water temperature at the acquisition timing of the heating element supply water temperature that is the same as the acquisition timing of the blade supply water temperature, and the amount of change in the heating element supply water temperature per unit time at the acquisition timing of the heating element supply water temperature.

[0121] As the room temperature history feature, a pair of the room temperature at the same room temperature acquisition timing as the blade supply water temperature acquisition timing and the room temperature change per unit time at the room temperature acquisition timing may be applied. As the machining state history feature, a pair of the machining state at the machining state acquisition timing and the change in the machining state per unit time at the machining state acquisition timing may be applied. The "same timing" here may include a deviation within an allowable range. For example, a deviation of less than the sampling period of each temperature may be considered to be substantially the same timing.

[0122] 8 indicates that the probability that the machining error is 1.5 micrometers is 90 percent, the probability that the machining error is 1.0 micrometers is 5 percent, and the probability that the machining error is 0.5 micrometers is 2 percent. Note that the room temperature history feature value described in the embodiment is an example of an environmental temperature history feature value that indicates the characteristics of the environmental temperature history.

[0123] [Time until temperature saturates for each part due to various temperature changes] The time it takes for the temperature of each part to reach saturation in response to a change in room temperature is as follows. Work: 3 minutes Table: 5 minutes Spindle: 10 minutes Microscope: 10 minutes Z-axis structure: 1 hour Casting: 8 hours The time until the temperature of each part becomes saturated in response to a change in the temperature of the water supplied to the blades or when the water supply to the blades is switched from off to on is as follows. Work: 1 minute or less Table: 3 minutes Spindle: 10 minutes Microscope: 5 minutes Z-axis structure: 20 minutes Casting: 1 hour The time until the temperature of each part becomes saturated in response to a temperature change in the heating element supply water or switching the heating element supply water from off to on is as follows. Spindle: 2 minutes Z-axis structure: 10 minutes Scale: 2 minutes Microscope: 1 minute The above temperature values ​​are an estimate of the time it takes for the temperature changes in each part to roughly converge when each temperature is changed by 2°C under the standard temperature conditions.

[0124] For example, the microscope is affected by changes in room temperature, temperature changes in the water supplied to the blade, and temperature changes in the water supplied to the heating element. When the room temperature rises, the microscope itself shifts to the negative side in the Y-axis direction relative to the workpiece W due to the expansion of the members supporting the microscope. However, a shift to the positive side in the Y-axis direction also occurs due to the expansion of the workpiece W, and when the temperature changes of the Z-axis structure and the casting catch up, the elongation relative to the workpiece W is somewhat alleviated.

[0125] The rise in blade supply water occurs only in the cutting state, and is caused by the expansion of the member supporting the microscope, causing the microscope itself to shift to the negative side in the Y-axis direction relative to the workpiece W. However, this is the case when the temperature of the blade supply water shifts from 2°C higher than the reference room temperature to 4°C higher, and if the temperature after the rise is lower than the reference room temperature plus 2°C, it will shift to the positive side in the Y-axis direction. Also, because a shift in position in the Y-axis direction occurs due to the expansion of the workpiece W and a shift in position in the Y-axis direction occurs due to the expansion of the spindle 14 due to the temperature change in the blade supply water, even a temperature rise of 2°C will result in different behavior depending on the temperature difference from room temperature.

[0126] The rise in temperature of the heating element supply water causes the expansion of the members supporting the microscope and the expansion of the members supporting the camera, shifting the microscope to the negative side in the Y-axis direction relative to the workpiece W. The spindle 14, which uses the same heating element supply water as the microscope, also expands in the same manner, causing a shift in the position of the spindle 14 in the Y-axis direction. On the other hand, if the scale is water-cooled, the shift of the spindle 14 is canceled.

[0127] The time constants for various temperature changes are different, and each time constant is also different under conditions other than the reference temperature conditions. This is caused by changes in the temperature gradient of each part and reversals in the direction of the gradient. For this reason, it is difficult to estimate the machining error caused by the relative positional deviation between the workpiece W and the microscope and the relative positional deviation between the workpiece W and the blade 12 based on the temperature measurement value at a certain point in time.

[0128] In this embodiment, the fluctuations of various temperatures are monitored for a certain period of time from a predetermined point after the start of processing, and the processing error is predicted using an appropriate learning model 158. This makes it possible to perform correction at an appropriate timing and reduce the processing error. In addition, when the learning model 158 that takes into account the history of the processing state is used, the reduction of the processing error caused by the fluctuation of the processing state is realized.

[0129] [Specific examples of temperature history] FIG. 9 is an explanatory diagram of the temperature history. In FIG. 9, the temperature history is illustrated in a graph format with the horizontal axis representing time and the vertical axis representing temperature. The temperature history is understood as time-series data of a plurality of temperatures that are continuously acquired at a specified sampling period. As shown in the figure, the temperatures that make up the temperature history may include not only temperatures within the operating temperature range of the device, but also temperatures outside the operating temperature range.

[0130] An example of the reference temperature of the room temperature and the reference temperature of the heating element supply water is 22°C. An example of the reference temperature of the blade supply water is 24°C. The reference temperature of the heating element supply water may be the same as the reference temperature of the room temperature. The reference temperature of the blade supply water may be specified as 2°C higher than the reference temperature of the room temperature.

[0131] Fig. 10 is an explanatory diagram of the amount of change in temperature per unit time. Fig. 10 shows an enlarged view of region 200 in the temperature history shown in Fig. 9. Note that Fig. 10 shows a time axis and a temperature axis for convenience.

[0132] [Specific examples of temperature history features] 10 are timings for acquiring temperatures Te1, Te2, Te3, Te4, and Te5, respectively. The time interval dt between each timing is the temperature sampling period.

[0133] The amount of temperature change per unit time at each timing is calculated as the average rate of change of a function representing the temperature history. For example, the amount of temperature change per unit time at timing t2 is (Te2-Te1) / dt. Similarly, the amount of temperature change per unit time at timing t3 is (Te3-Te2) / dt. In other words, when i is an integer equal to or greater than 1, the amount of temperature change per unit time at the i-th timing ti is expressed as {Tei-Te(i-1)} / dt.

[0134] That is, when the temperature history is expressed as a function f(t), the amount of change in temperature per unit time at the i-th timing ti is expressed as the first derivative df(ti) / dt of the temperature history f(t).

[0135] When the temperature history feature derivation unit 152 shown in FIG. 5 acquires the temperature history shown in FIG. 9 from the temperature history generation unit 150, it derives the temperature for each temperature acquisition timing and the amount of change in temperature per unit time for each temperature acquisition timing as temperature history feature values.

[0136] The machining state history feature amount can be expressed in the same manner as the temperature history feature amount by applying the rotation speed of the spindle 14 or the like instead of the temperature of the temperature history feature amount. That is, the machining state history feature amount derivation unit 153 can acquire the machining history from the machining state history generation unit 151 and derive the machining state history feature amount.

[0137] [Learning model generation method] Fig. 11 is a schematic diagram of a learning model generation method. The generation of the trained learning model 158 shown in Fig. 5 includes a plurality of learning data collection steps S100, a correct answer label assignment step S102, and a learning step S104.

[0138] In the multiple learning data collection step S100, power usage information and image information stored in the control device 26 are collected as learning data. The power usage information includes the cutting water temperature history, the cooling water temperature history, and the room temperature history during the operation period of the device. A temperature history feature value Tef(t) is derived from the cutting water temperature history, the cooling water temperature history, and the room temperature history. The temperature history feature value Tef(t) represents multiple temperature history feature values ​​at different timings.

[0139] The temperature history feature amount Tef(t) represents a plurality of temperature history feature amounts at any timing t. The temperature history feature amount Tef(t) includes the feature amount of the cutting water temperature history at any timing t, the feature amount of the cooling water temperature history at any timing t, and the feature amount of the room temperature history at any timing t.

[0140] The image information is photographed data of the workpiece W acquired when a kerf check is performed by the device. A machining error Er(t) is derived from the photographed data of the workpiece W. The machining error Er(t) represents a plurality of machining errors at an arbitrary timing t.

[0141] In the correct label assignment process S102, the processing error Er(t) is assigned as a correct label to each of the multiple temperature history feature values ​​Tef(t), and a pair of the temperature history feature value Tef(t) and the processing error Er(t) corresponding to the temperature history feature value Tef(t) is generated as learning data.

[0142] In the learning process S104, using the learning data generated in the correct label assignment process S102, supervised learning is performed in which the temperature history feature Tef(t) is used as an input to the learning model 159 and the processing error Er(t) is used as correct data, and a trained learning model 158 shown in Figure 5 is generated.

[0143] Using the power usage information and image information stored when the dicing apparatus 10 is operated, re-learning may be performed by applying the procedure for generating a learning model shown in FIG. 11 to the trained learning model 158 shown in FIG. 5.

[0144] When the processing state is taken into consideration, learning data regarding the processing state is collected in a learning data collection step S100, and a learning model 158 in which the processing state is taken into consideration is generated through a correct label assignment step S102 and a learning step S104.

[0145] [First modified example of temperature feature amount] FIG. 12 is an explanatory diagram of a first modified example of the temperature history feature quantity. In the temperature history feature quantity according to the first modified example, the direction of temperature change is added to the temperature and the amount of temperature change per unit time. The direction of temperature change indicates whether the amount of temperature change per unit time is increasing or decreasing, and can be calculated as the average rate of change of a function that indicates the amount of temperature change per unit time. In other words, when the temperature history is expressed as f(t) using a function, the gradient of the temperature change is expressed as the second derivative coefficient d of the temperature history f(t) at the i-th timing ti. 2 f(ti) / dt 2 This is expressed as:

[0146] That is, the blade supply water temperature history feature values ​​relating to the first modified example are the blade supply water temperature, the amount of change in the blade supply water temperature per unit time, and the direction of change in the blade supply water temperature. The heating element supply water temperature history feature values ​​relating to the first modified example are the heating element supply water temperature, the amount of change in the heating element supply water temperature per unit time, and the direction of change in the heating element supply water temperature. The room temperature history feature values ​​relating to the first modified example are the room temperature, the amount of change in the room temperature per unit time, and the direction of change in the room temperature. The first modified example of the temperature history feature values ​​can also be applied to the modified example of the machining state history feature values.

[0147] [Second modified example of temperature feature quantity] 13 is an explanatory diagram of a second modified example of the temperature history feature quantity. The temperature history feature quantity according to the second modified example is a sum of temperatures instead of a change in the temperature feature quantity per unit time. When the temperature history is expressed as f(t) using a function, the sum of temperatures is calculated as an interval integral of the temperature history f(t) from the processing start timing to the timing ti.

[0148] That is, the blade supply water temperature history feature value according to the second modified example is the sum of the blade supply water temperature and the blade supply water temperature. The heating element supply water temperature history feature value according to the second modified example is the sum of the heating element supply water temperature and the heating element supply water temperature. The room temperature history feature value according to the second modified example is the room temperature and the sum of the room temperature. The second modified example of the temperature history feature value can also be applied to the modified example of the machining state history feature value.

[0149] The sum of the blade supply water temperatures described in the embodiments is an example of an integrated blade supply water temperature. The sum of the heating element supply water temperatures described in the embodiments is an example of an integrated heating element supply water temperature. The sum of the room temperature described in the embodiments is an example of an integrated machining environment temperature.

[0150] [Third modified example of temperature feature quantity] FIG. 14 is an explanatory diagram of a third modified example of the temperature history feature quantity. For the temperature history feature quantity according to the third modified example, the temperature history for each hour is applied as the temperature feature quantity. That is, for the blade supply water temperature feature quantity according to the third modified example, the temperature history of the blade supply water for each hour is applied. For the heating element supply water temperature feature quantity according to the third modified example, the temperature history of the heating element supply water for each hour is applied. For the room temperature feature quantity according to the third modified example, the room temperature history for each hour is applied.

[0151] The period of the temperature history may be 1 hour, 2 hours, 3 hours, etc., and may be determined according to the time constant of the change in the processing error. The number of samples of the temperature history may be 3 or more, but the more samples there are, the more accurate the processing error can be derived.

[0152] Fig. 15 is a schematic diagram of a method for generating a learning model in which a temperature history feature value according to the third modified example is used as learning data. In the learning data collection step S101 shown in the figure, a one-hour temperature history is applied as the temperature history feature value Tef(t) obtained from the power usage information. In addition, a processing error Er(t) for each hour is obtained from the image information.

[0153] In the correct label assignment step S102, the processing error Er(t) for each hour is assigned to the temperature history feature Tef(t) to which the temperature history for one hour is applied. For example, the processing error at the timing when one hour has passed from the start of processing is assigned to the temperature history feature Tef(t) for the period from the start of processing to the time when one hour has passed.

[0154] In the learning process S104, a pair of the temperature history feature Tef(t) and the processing error Er(t) is used as learning data, and supervised learning is performed in which the temperature history feature Tef(t) is used as the input to the learning model 159 and the processing error Er(t) is used as the correct answer data, thereby generating a learned learning model 158 shown in Figure 5.

[0155] [Examples of machining state detection devices and their applications to machining state detection] 4 and 5 can function as a processing state detection device applied to the dicing apparatus 10. A computer is applied to the processing state detection device. Various functions of the processing state detection device are realized by the computer executing a program.

[0156] Further, the history information generating step S10, the history feature amount deriving step S12 and the machining error predicting step S14 shown in FIG. 6 are understood as a machining state predicting method in which the machining state detecting device, which is a computer, carries out the above-mentioned respective steps.

[0157] [Effects of the embodiment] The dicing apparatus 10 according to the embodiment can provide the following advantageous effects.

[0158] [1] Based on the temperature history feature value at any timing during machining, the machining error at the timing when the temperature during machining is acquired is predicted. The temperature history feature value includes the temperature history feature value of the water supplied to the blade, the temperature history feature value of the water supplied to the heating element, and the room temperature history feature value. This makes it possible to predict the machining error in response to temperature change for multiple temperature factors that have mutual causal relationships that are difficult for an operator to determine and have different time constants of change.

[0159] [2] The prediction of the machining error is applied to a trained learning model 158. The learning model 158 is trained using a set of temperature history feature value and machining error as training data, and is trained and generated so that a predicted value of the machining error is output when the temperature history feature value is input. This is expected to improve the prediction accuracy of the machining error.

[0160] [3] The temperature history feature value is the temperature at the timing when the temperature history is acquired and the amount of change in temperature per unit time at the timing when the temperature history is acquired. This makes it possible to predict the machining error taking into account the amount of change in temperature per unit time.

[0161] [4] The temperature history feature amount is the temperature at the timing when the temperature history is acquired, the amount of change in temperature per unit time at the timing when the temperature history is acquired, and the direction of the temperature change. This makes it possible to predict the machining error taking into account the direction of the temperature change.

[0162] [5] The temperature history feature amount is the temperature at the timing when the temperature history is acquired and the sum of the temperatures at the timing when the temperature history is acquired. This makes it possible to predict the machining error taking the sum of the temperatures into consideration.

[0163] [6] The machining error that takes into account the change in the machining state during machining is predicted based on the machining state history feature amount at any timing during machining, thereby reducing the machining error caused by the change in the machining state.

[0164] [7] The system is equipped with an analysis unit that predicts the transition of machining errors from predicted values ​​of machining errors, which allows the machining position to be corrected at the appropriate time without relying on the operator's empirical judgment.

[0165] [8] The learning model 158 is re-learned using the power usage information and image information stored when the dicing device 10 is operated. This eliminates the need for settings and conditions corresponding to the device environment, and allows accurate correction of the processing position as the number of times the dicing device 10 is operated increases.

[0166] The above-described embodiment of the present invention may be modified, added, or deleted as appropriate within the scope of the gist of the present invention. The present invention is not limited to the above-described embodiment, and many modifications may be made by a person having ordinary knowledge in the relevant field within the technical concept of the present invention. In addition, the embodiment, modified example, and application example may be implemented in combination as appropriate. [Explanation of symbols]

[0167] 10... dicing device, 12... blade, 14... spindle, 18... processing unit, 26... control device (device control means), 100... system control unit, 102... processing control unit, 106... blade supply water temperature acquisition unit, 108... heating element supply water temperature acquisition unit, 110... room temperature acquisition unit, 120... computer readable medium, 122... memory, 130... blade supply water temperature detection unit, 132... heating element supply water temperature detection unit, 134... room temperature detection unit, 140... processing state detection unit, 150... temperature history generation unit, 152... temperature history feature derivation unit, 154... processing error prediction unit, 156... analysis unit, 158... learning model, 160... processing condition acquisition unit, 162... learning model selection unit, 164... learning model memory unit, Er... processing error, Tef... temperature history feature

Claims

1. A machining state detection device that detects a machining state when machining an object using a blade, an environmental temperature history acquisition unit that acquires an environmental temperature history representing the temperature history of the processing environment; a blade supply water temperature history acquisition unit that acquires a blade supply water temperature history that indicates a temperature history of the blade supply water supplied to the blade; a heating element supply water temperature history acquisition unit that acquires a heating element supply water temperature history that indicates a temperature history of heating element supply water that is supplied to a heating element that generates heat when processing the object to be processed; a temperature history feature value deriving unit that derives an environmental temperature history feature value that represents a feature of the environmental temperature history, a blade supply water temperature history feature value that represents a feature of the blade supply water temperature history, and a heating element supply water temperature history feature value that represents a feature of the heating element supply water temperature history; a machining error prediction unit that predicts a machining error based on temperature history feature amounts including the environmental temperature history feature amount, the blade supply water temperature history feature amount, and the heating element supply water temperature history feature amount; Equipped with The machining error prediction unit is a trained learning model that has trained using the temperature history feature value and the machining error as training data, and the machining state detection device applies a learning model that outputs the machining error when the temperature history feature value is input.

2. a machining state history acquisition unit that acquires a machining state history representing a history of the machining state; a machining state history feature amount derivation unit that derives feature amounts of the machining state history; Equipped with 2. The machining state detection device according to claim 1, wherein the machining error prediction unit predicts the machining error by taking into account the temperature history feature amount and the machining state history feature amount.

3. The temperature history feature amount derivation unit As the environmental temperature history feature amount, a temperature of the machining environment and a change amount of the temperature of the machining environment per unit time are acquired; a temperature of the blade supply water and a change in the temperature of the blade supply water per unit time are acquired as the blade supply water temperature history feature amount; As the heating element supply water temperature history feature amount, the temperature of the heating element supply water and a change amount of the heating element supply water per unit time are acquired; The learning model is The machining state detection device described in claim 1 is generated by performing learning using a combination of the temperature of the machining environment and the amount of change in the temperature of the machining environment per unit time, a combination of the temperature of the blade supply water and the amount of change in the temperature of the blade supply water per unit time, and a combination of the temperature of the heating element supply water and the amount of change in the temperature of the heating element supply water per unit time as inputs, and outputting the machining error.

4. The temperature history feature amount derivation unit As the environmental temperature history feature amount, the temperature of the machining environment, the amount of change in the temperature of the machining environment per unit time, and the direction of change in the temperature of the machining environment are acquired; As the blade supply water temperature history feature amount, the temperature of the blade supply water, the amount of change in the temperature of the blade supply water per unit time, and the direction of change in the temperature of the blade supply water are acquired; As the heating element supply water temperature history feature amount, the temperature of the heating element supply water, the amount of change in the temperature of the heating element supply water per unit time, and the direction of change in the temperature of the heating element supply water are acquired; The learning model is The machining state detection device described in claim 1 is generated by performing learning using as inputs a combination of the temperature of the machining environment, the amount of change in the temperature of the machining environment per unit time, and the direction of change in the temperature of the machining environment, a combination of the temperature of the blade supply water, the amount of change in the temperature of the blade supply water per unit time, and the direction of change in the temperature of the blade supply water, and a combination of the temperature of the heating element supply water, the amount of change in the temperature of the heating element supply water per unit time, and the direction of change in the temperature of the heating element supply water, and the machining error as output.

5. The temperature history feature amount derivation unit As the environmental temperature history feature amount, the temperature of the processing environment and an integrated value of the temperature of the processing environment are acquired; acquiring the temperature of the blade supply water and an integrated value of the temperature of the blade supply water as the blade supply water temperature history feature amount; acquiring the temperature of the heating element supply water and an integrated value of the temperature of the heating element supply water as the heating element supply water temperature history feature amount; The learning model is The machining state detection device described in claim 1 is generated by performing learning using a combination of the temperature of the machining environment and the integrated temperature of the machining environment, a combination of the temperature of the blade supply water and the integrated temperature of the blade supply water, and a combination of the temperature of the heating element supply water and the integrated temperature of the heating element supply water as inputs, and outputting the machining error.

6. a processing condition acquisition unit that acquires processing conditions to be applied to the processing; a learning model selection unit that selects a learning model to be applied to the machining error prediction unit according to the machining conditions from among a plurality of first learning models generated by performing the learning for each machining condition; The machining state detection device according to any one of claims 1 to 5, comprising:

7. A machining state detection method for detecting a machining state when machining an object using a blade, comprising: The control device an environmental temperature history acquisition step of acquiring an environmental temperature history representing the temperature history of the processing environment; a blade supply water temperature history acquisition step of acquiring a blade supply water temperature history representing a temperature history of the blade supply water supplied to the blade; a heating element supply water temperature history acquisition step of acquiring a heating element supply water temperature history representing a temperature history of heating element supply water supplied to a heating element that generates heat when processing the object to be processed; a temperature history feature value derivation process for deriving an environmental temperature history feature value representing a feature of the environmental temperature history, a blade supply water temperature history feature value representing a feature of the blade supply water temperature history, and a heating element supply water temperature history feature value representing a feature of the heating element supply water temperature history; a machining error prediction step of predicting a machining error based on temperature history feature values ​​including the environmental temperature history feature value, the blade supply water temperature history feature value, and the heating element supply water temperature history feature value; Run The machining error prediction step is a trained learning model that has been trained using the temperature history feature value and the machining error as training data, and the training model that outputs the machining error when the temperature history feature value is input is applied to the machining state detection method.

8. A program for detecting a processing state when processing an object using a blade, The control device An environmental temperature history acquisition function that acquires the environmental temperature history that represents the temperature history of the processing environment; a blade supply water temperature history acquisition function for acquiring a blade supply water temperature history representing a temperature history of the blade supply water supplied to the blade; a heating element supply water temperature history acquisition function for acquiring a heating element supply water temperature history representing the temperature history of heating element supply water supplied to a heating element that generates heat when processing the object to be processed; a temperature history feature value derivation function for deriving an environmental temperature history feature value representing a feature of the environmental temperature history, a blade supply water temperature history feature value representing a feature of the blade supply water temperature history, and a heating element supply water temperature history feature value representing a feature of the heating element supply water temperature history; and a program for realizing a machining error prediction function for predicting a machining error based on temperature history feature amounts including the environmental temperature history feature amount, the blade supply water temperature history feature amount, and the heating element supply water temperature history feature amount, The machining error prediction function is a trained learning model that has been trained using the temperature history feature and the machining error as training data, and is a program that applies a learning model that outputs the machining error when the temperature history feature is input.

9. a processing unit that processes the workpiece using a blade; a processing state detection unit that detects a processing state of the object to be processed, The machining state detection unit an environmental temperature history acquisition unit that acquires an environmental temperature history representing the temperature history of the processing environment; a blade supply water temperature history acquisition unit that acquires a blade supply water temperature history that indicates a temperature history of the blade supply water supplied to the blade; a heating element supply water temperature history acquisition unit that acquires a heating element supply water temperature history that indicates a temperature history of heating element supply water that is supplied to a heating element that generates heat when processing the object to be processed; a temperature history feature value deriving unit that derives an environmental temperature history feature value that represents a feature of the environmental temperature history, a blade supply water temperature history feature value that represents a feature of the blade supply water temperature history, and a heating element supply water temperature history feature value that represents a feature of the heating element supply water temperature history; a machining error prediction unit that predicts a machining error based on temperature history feature amounts including the environmental temperature history feature amount, the blade supply water temperature history feature amount, and the heating element supply water temperature history feature amount; Equipped with The processing error prediction unit is a trained learning model that has trained using the temperature history feature and the processing error as training data, and the dicing device to which the training model that outputs the processing error when the temperature history feature is input is applied.

10. 10. The dicing device according to claim 9, further comprising a determination unit that determines whether or not correction of the processing unit is required based on the processing error output from the processing error prediction unit.

11. the machining error prediction unit predicts a machining error after a specified period of time has elapsed, The dicing device according to claim 10 , wherein the determination unit derives a determination result indicating that correction of the processing unit is necessary when the predicted processing error exceeds a specified threshold value.

12. 12. The dicing device according to claim 11, wherein the processing error prediction unit predicts the processing error after a specified period of time has elapsed based on the amount of change in the processing error per unit time and the direction of the change in the processing error.

13. A learning model generation method for generating a learning model that has been trained using temperature history features including an environmental temperature history that represents the temperature history of the processing environment, a blade supply water temperature history that represents the temperature history of blade supply water supplied to a blade that processes the workpiece, and a heating element supply water temperature history that represents the temperature history of heating element supply water supplied to a heating element that generates heat when processing the workpiece, as learning data, and a processing error when processing the workpiece, and that outputs the processing error when the temperature history features are input.

Images