Machine learning method, simulation device, laser processing system, and program used in a laser processing system

Abstract

The present invention aims to learn the degree of processing of a processing part before and after irradiation of laser light during laser processing, the material of a processing object, and the parameters of the irradiated laser light. [Solution] Deep learning is performed using the material of the workpiece, laser light parameters that indicate the characteristics of the laser light irradiated onto the workpiece, and pre-processing section data and post-processing section data that reflect the three-dimensional shape of the workpiece resulting from laser processing of the processing section before and after irradiating the laser light, and a first relationship is obtained as one of the learning results, where when the material of the workpiece, pre-processing section data, and laser light parameters are used as input data, the post-processing section data after irradiating the laser light related to the input data is used as output data. [Selected Figure] Figure 1

Description

Technical Field

[0001] The present invention relates to a machine learning method, a simulation device, a laser processing system, and a program used in a laser processing system.

Background Art

[0002] Conventionally, as a technique related to machine learning used in this type of laser processing system, there are a state quantity observation unit that observes the state quantity of the laser processing system, an operation result acquisition unit that acquires the processing result by the laser processing system, and an output from the state quantity observation unit and an output from the operation result acquisition unit, and a learning unit that learns laser processing condition data in association with the state quantity and the processing result of the laser processing system, and a decision-making unit that outputs the laser processing condition data with reference to the laser processing condition data learned by the learning unit. A machine learning device has been proposed (see, for example, Patent Document 1). In this device, it is said that laser processing condition data that can obtain an optimal processing result can be determined by such machine learning.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] In laser processing, as shown in Fig. 13, the removed volume (ablation volume) removed from the workpiece with respect to the number of irradiation pulses increases non-linearly, and it is also understood that the ablation volume increases non-linearly with respect to the fluence of the laser light irradiated (pulse energy per unit area). For this reason, even if parameters are set for the workpiece and laser light is irradiated, it has been extremely difficult to estimate the extent of ablation processing due to the irradiation of the laser light, or to determine how to set the parameters of the laser light for performing specific processing on the workpiece. In the above-described machine learning apparatus, it is said that laser processing condition data that can obtain an optimal processing result can be determined, but it is only learning in relation to the result of laser processing, and learning in relation to the degree of processing of the processed part before and after irradiation of the laser light during laser processing has not been performed.

[0005] The machine learning method used in the laser processing system of the present invention mainly aims to learn the degree of processing of the processed part before and after irradiation of the laser light during laser processing, the material of the workpiece, and the parameters of the laser light to be irradiated.

[0006] The simulation apparatus used in the laser processing system of the present invention mainly aims to perform a simulation using the learning results of the degree of processing of the processed part before and after irradiation of the laser light during laser processing, the material of the workpiece, and the parameters of the laser light to be irradiated.

[0007] The laser processing system of the present invention mainly aims to improve the accuracy of laser processing.

[0008] The program of the present invention mainly aims to function a computer as a machine learning apparatus that learns the degree of processing of the processed part before and after irradiation of the laser light during laser processing, the material of the workpiece, and the parameters of the laser light to be irradiated.

Means for Solving the Problems

[0009] The machine learning method, simulation device, laser processing system, and program used in the laser processing system of the present invention have adopted the following means to achieve the above main object.

[0010] The machine learning method of the present invention is a machine learning method used in a laser processing system that performs ablation processing by irradiating a processing object with a laser beam, performing deep learning using the material of the processing object, laser beam parameters indicating the characteristics of the laser beam irradiated on the processing object, pre-processing part data and post-processing part data that reflect the three-dimensional shape associated with the laser processing of the processing part before and after irradiating the processing object with the laser beam, and obtaining, as one of the learning results, a first relationship in which the post-processing part data after the laser beam related to the input data is irradiated when the material of the processing object, the pre-processing part data, and the laser beam parameters are used as input data is used as the output data. It is characterized by this.

[0011] In the machine learning method of the present invention, deep learning is performed using the material of the object to be processed, laser light parameters indicating the characteristics of the laser light irradiated on the object to be processed, pre-processing part data and post-processing part data reflecting the three-dimensional shape associated with the laser processing of the processing part before and after irradiating the object to be processed with the laser light. As a result of this deep learning (learning result), a first relationship is obtained in which the post-processing part data after irradiating the laser light related to the input data is used as the output data when the material of the object to be processed, the pre-processing part data, and the laser light parameters are used as the input data. Here, since the pre-processing part data and the post-processing part data are data reflecting the three-dimensional shape associated with the laser processing of the processing part before and after irradiating the object to be processed with the laser light, data before and after irradiation of the laser light before laser processing and during laser processing, data before and after irradiation of the laser light during laser processing, data before and after irradiation of the laser light during laser processing and after laser processing (after processing completion), etc. are also included. Thereby, it is possible to learn the degree of processing of the processing part before and after irradiation of the laser light during laser processing, the material of the object to be processed, and the parameters of the laser light to be irradiated, and the first relationship in which the post-processing part data after irradiating the laser light related to the input data is used as the output data when the material of the object to be processed, the pre-processing part data, and the laser light parameters are used as the input data can be obtained as the learning result. "Deep learning" in the machine learning method of the present invention uses the pre-processing part data and the post-processing part data and obtains the first relationship in which the post-processing part data is used as the output data as the learning result, so basically it is supervised learning. Also, as the "laser light parameters", at least a part of the wavelength, pulse width, pulse intensity, spot diameter, number of pulses, fluence (pulse energy per unit area) can be used, and as the "pre-processing part data" and "post-processing part data", at least a part of the three-dimensional shape data of the processing part, the surface temperature distribution data of the processing part, and the color distribution data of the processing part can be used. This is because the surface temperature distribution data of the processing part and the color distribution data of the processing part indicate the change in temperature due to the irradiation of the laser light and are considered to reflect the ease of change in the three-dimensional shape and the change caused thereby. Here, the color distribution data includes not only the data of the normal color distribution but also Raman spectrum data, light reflectance spectrum data, etc.

[0012] In the machine learning method of the present invention, based on the first relationship, when the material of the object to be processed, the pre-processing part data, and the post-processing part data are used as input data, the state of the processing part of the post-processing part data is obtained from the state of the processing part of the pre-processing part data related to the input data. It is also possible to obtain, as one of the learning results, a second relationship in which the laser light parameters of the laser light to be irradiated, which are necessary for this, are used as output data. In this way, when the material of the object to be processed, the pre-processing part data, and the post-processing part data are used as input data, the laser light parameters of the laser light to be irradiated, which are necessary for changing the state of the processing part of the post-processing part data from the state of the processing part of the pre-processing part data related to the input data, can be obtained as the learning result.

[0013] The simulation device of the present invention is a simulation device used in a laser processing system that performs ablation processing by irradiating an object to be processed with laser light, the learning results obtained by the machine learning method according to any of the above aspects of the present invention, that is, the material of the object to be processed, the laser light parameters indicating the characteristics of the laser light irradiated on the object to be processed, the pre-processing part data and the post-processing part data reflecting the three-dimensional shape associated with the laser processing of the processing part before and after irradiating the object to be processed with laser light. Using these, a first relationship in which, when the material of the object to be processed, the pre-processing part data, and the laser light parameters are used as input data, the post-processing part data after the laser light related to the input data is irradiated is used as output data, and based on the first relationship, when the material of the object to be processed, the pre-processing part data, and the post-processing part data are used as input data, the state of the processing part of the post-processing part data is obtained from the state of the processing part of the pre-processing part data related to the input data. Using the second relationship in which the laser light parameters of the laser light to be irradiated, which are necessary for this, are used as output data, the output data is output for the input data. It is characterized by this.

[0014] In the simulation device of the present invention, when the material of the object to be processed, the pre-processing part data, and the laser beam parameters are used as input data, the post-processing part data after the laser beam related to the input data is irradiated is output, or when the material of the object to be processed, the pre-processing part data, and the post-processing part data are used as input data, the laser beam parameters of the laser beam to be irradiated necessary to change the state of the post-processing part from the state of the pre-processing part related to the input data can be output. As a result, simulation can be performed using the learning results of the degree of processing of the processing part before and after the irradiation of the laser beam during laser processing, the material of the object to be processed, and the parameters of the laser beam to be irradiated.

[0015] In the simulation device of the present invention, data obtained by inputting the shape of the object to be processed before processing and the target shape, and adjusting the laser beam parameters and the laser beam irradiation position according to the difference between the shape during processing and the target shape is used as the input data, and the output data as the result of the machining simulation obtained by applying the learning result to the input data is used as the shape during processing, and the machining simulation may be repeated until the difference between the shape during processing and the target shape falls within a predetermined range. For example, as initial values, arbitrary laser beam parameters and an arbitrary number of laser irradiation positions are set, and the material of the object to be processed, the shape before processing, etc. are added thereto to obtain input data, and machining simulation is applied to obtain output data (post-processing part data after the laser beam related to the input data is irradiated). The obtained output data is used as the shape during processing to determine whether the difference from the target shape is within a predetermined range. When the difference between the shape during processing and the target shape is not within the predetermined range, the laser beam parameters and the laser irradiation position are adjusted according to the difference to obtain input data. Then, machining simulation is applied to the adjusted input data to obtain output data. The process of adjusting the input data according to the difference between the intermediate shape and the target shape, the process of applying machining simulation to the adjusted input data to obtain output data, and the process are repeated until the difference between the shape during processing (output data) and the target shape falls within a predetermined range. The number of repetitions of such a repetitive process and the laser beam parameters and the laser irradiation position adjusted for each repetition become the simulation results. In this case, by changing the initial value and obtaining simulation results a plurality of times, the optimum values for laser processing can be obtained by comparing the total energy, processing time, processing accuracy, etc. due to laser irradiation in each simulation result.

[0016] The first laser processing system of the present invention is a laser processing system including a processing laser beam irradiation device that irradiates a laser beam onto an object to be processed to perform ablation processing, a processing part data measuring device that measures processing part data reflecting a three-dimensional shape associated with the laser processing of the object to be processed, and a control device that controls the processing laser beam irradiation device. The control device performs learning using the machine learning method according to any of the above-described aspects of the present invention, that is, the material of the object to be processed, the laser light parameters indicating the characteristics of the laser light irradiated on the object to be processed, the pre-processing part data and the post-processing part data reflecting the three-dimensional shape associated with the laser processing of the processing part before and after irradiating the object to be processed with the laser light. Deep learning is performed using these data, and when the material of the object to be processed, the pre-processing part data, and the laser light parameters are used as input data, the post-processing part data after the laser light related to the input data is irradiated is used as output data, and a first relationship is obtained as one of the learning results. Based on the first relationship, when the material of the object to be processed, the pre-processing part data, and the post-processing part data are used as input data, the laser light parameters of the laser light to be irradiated necessary to change the state of the processing part of the post-processing part data from the state of the processing part of the pre-processing part data related to the input data are used as output data, and a second relationship is obtained as one of the learning results. The laser processing laser light irradiation device is controlled based on the output data for the input data using the learning result. It is characterized by this.

[0017] In the first laser processing system of the present invention, since learning using the machine learning method according to any of the above-described aspects of the present invention is performed, laser processing can be performed using the learning results of the degree of processing of the processing part before and after irradiation of the laser light during laser processing, the material of the object to be processed, and the parameters of the laser light to be irradiated. Further, since the first relationship and the second relationship are learned every time laser processing is performed, the accuracy of laser processing can be improved.

[0018] The second laser processing system of the present invention is A laser processing system including a laser processing laser light irradiation device that irradiates a laser light to an object to be processed to perform ablation processing, and a control device that controls the laser processing laser light irradiation device, The control device performs deep learning using the learning results obtained by the machine learning method according to any aspect of the present invention described above, that is, the material of the object to be processed, the laser light parameters indicating the characteristics of the laser light irradiated on the object to be processed, the pre-processing part data and the post-processing part data reflecting the three-dimensional shape associated with the laser processing of the processing part before and after irradiating the object to be processed with the laser light. A first relationship in which the post-processing part data after irradiating the input data with the laser light related to the input data is used as the output data when the material of the object to be processed, the pre-processing part data, and the laser light parameters are used as the input data, and based on the first relationship, when the material of the object to be processed, the pre-processing part data, and the post-processing part data are used as the input data, the laser light parameters of the laser light to be irradiated necessary to change the state of the processing part of the post-processing part data from the state of the processing part of the pre-processing part data related to the input data are used as the output data. A second relationship, and controls the laser light irradiation device for processing based on the output data with respect to the input data using the first relationship and the second relationship. It is characterized by this.

[0019] In the second laser processing system of the present invention, since the learning results obtained by the machine learning method according to any aspect of the present invention described above are used, laser processing can be performed using the learning results of the degree of processing of the processing part before and after irradiation of the laser light during laser processing, the material of the object to be processed, and the parameters of the laser light to be irradiated. As a result, the accuracy of laser processing can be improved.

[0020] The program of the present invention is A program that causes a computer to function as a machine learning device used in a laser processing system, a step of inputting a plurality of data including the material of the object to be processed, the laser light parameters indicating the characteristics of the laser light irradiated on the object to be processed, the pre-processing part data and the post-processing part data reflecting the three-dimensional shape associated with the laser processing of the processing part before and after irradiating the object to be processed with the laser light, Using the input plurality of data, through deep learning, a step of obtaining, as one of the learning results, a first relationship in which, when the material of the object to be processed, the pre-processing part data, and the laser beam parameters are input data, the post-processing part data after the laser beam related to the input data is irradiated is used as the output data; It is characterized by having.

[0021] In the program of the present invention, a plurality of data including the material of the object to be processed, the laser beam parameters indicating the characteristics of the laser beam irradiated on the object to be processed, the pre-processing part data reflecting the three-dimensional shape associated with the laser processing of the processing part before and after irradiating the object to be processed with the laser beam, and the post-processing part data are input. Using the input plurality of data, through deep learning, a first relationship in which, when the material of the object to be processed, the pre-processing part data, and the laser beam parameters are input data, the post-processing part data after the laser beam related to the input data is irradiated is used as the output data is obtained as one of the learning results. Here, since the pre-processing part data and the post-processing part data are data reflecting the three-dimensional shape associated with the laser processing of the processing part before and after irradiating the object to be processed with the laser beam, data before and after irradiation of the laser beam before laser processing and during laser processing, data before and after irradiation of the laser beam during laser processing, data before and after irradiation of the laser beam during laser processing and after laser processing (after processing completion), etc. are also included. Thereby, the computer can be made to function as a machine learning device that learns the degree of processing of the processing part before and after irradiation of the laser beam during laser processing, the material of the object to be processed, and the parameters of the laser beam to be irradiated. Note that "deep learning" uses the pre-processing part data and the post-processing part data and obtains, as the learning result, a first relationship in which the post-processing part data is the output data, so basically it is supervised learning. Also, for the "laser beam parameters", at least a part of the wavelength, pulse width, pulse intensity, number of pulses, fluence (pulse energy per unit area) can be used, and for the "pre-processing part data" and the "post-processing part data", at least a part of the three-dimensional data of the processing part, the surface temperature distribution data of the processing part, the color distribution data of the processing part can be used.

[0022] In the program of the present invention, based on the first relationship, when the material of the object to be processed, the pre-processing part data, and the post-processing part data are used as input data, the laser light parameters of the laser light to be irradiated, which are necessary for changing the state of the processing part of the post-processing part data from the state of the processing part of the pre-processing part data related to the input data, are obtained as one of the learning results in a step of obtaining a second relationship with the output data. By doing so, the computer can be made to function as a machine learning device that obtains, as a learning result, a second relationship in which the laser light parameters of the laser light to be irradiated, which are necessary for changing the state of the processing part of the post-processing part data from the state of the processing part of the pre-processing part data related to the input data, are used as output data when the material of the object to be processed, the pre-processing part data, and the post-processing part data are used as input data.

Brief Description of Drawings

[0023]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Embodiments for Carrying Out the Invention

[0024] Next, embodiments for carrying out the present invention will be described using examples.

Examples

[0025] FIG. 1 is a configuration diagram showing an outline of the configuration of a laser processing system 20 as an embodiment of the present invention. As shown in the figure, the laser processing system 20 of the embodiment includes a processing laser irradiation device 30 that outputs a processing laser beam, and an electric stage 46 on which a processing object 10 irradiated with the laser beam whose focus and irradiation position are adjusted by a focus lens 42 and a mirror 44 is output from the processing laser irradiation device 30, a three-dimensional data measuring device 48 that measures the three-dimensional shape data of the processing part of the processing object 10, and a system control unit 50 that controls the entire system.

[0026] The laser irradiation device 30 for processing includes a processing laser irradiator 32 that outputs a processing laser beam, a pulse picker 34 that extracts an arbitrary number of pulses from a pulse train of the laser beam from the processing laser irradiator 32 at an arbitrary timing, a half-wave plate 35 that adjusts the polarization direction of the laser beam from the pulse picker 34, a polarization beam splitter 36 that reflects the S-polarized light of the laser beam and transmits the P-polarized light, and a laser control unit 38 that controls these components.

[0027] The processing laser irradiator 32 is configured as a titanium sapphire laser irradiator that can output laser light (pulsed laser light) with a wavelength of, for example, 800 nm, a pulse width that can be changed from 35 fs to 10 ps, a repetition frequency of 1 kHz, a maximum output of 6 W, a maximum pulse energy of 6 mJ, and a fluence of 0.1 to 100 J / cm 2 ².

[0028] Instead of the half-wave plate 35 and the polarization beam splitter 36, an acousto-optic element (AOM), a neutral density filter, etc. can be used.

[0029] The laser control unit 38 is configured as a microcomputer centered on a CPU (not shown), and includes a ROM, a RAM, a flash memory, an input / output port, a communication port, etc. in addition to the CPU. The laser control unit 38 communicates with the system control unit 50 via the communication port. The laser control unit 38 controls the processing laser irradiator 32 so that laser light with laser beam parameters based on a control signal from the system control unit 50 is output. The laser beam parameters can use at least a part of wavelength, pulse width, pulse intensity, spot diameter, number of pulses, fluence (pulse energy per unit area). Also, the laser control unit 38 controls the pulse picker 34 so that the timing and the number of pulses of the laser beam pulse train are based on a control signal from the system control unit 50, or controls the half-wave plate 35 and the polarization beam splitter 36 so that the polarization direction of the laser beam is based on a control signal from the system control unit 50.

[0030] The electric stage 46 is a stage that moves the object to be processed 10 to the measurement position of the three-dimensional data measuring instrument 48. In the embodiment, one with a position accuracy of 0.5 μm and a movable distance of 150 mm was used.

[0031] The three-dimensional measuring instrument 48 can use, for example, an apparatus capable of measuring a three-dimensional shape such as a white interference microscope, a scanning laser microscope, an X-ray CT (Computed Tomography), a step gauge, an AMF (Atomic Force Microscope), or a Raman microscope. In the embodiment, a white interference microscope with a vertical resolution of 1 nm, a horizontal resolution of 0.2 μm, and a measurement time of 1 to 10 seconds was used as the measurement accuracy. If the same optical system as the laser light is used for the three-dimensional measuring instrument 48, the electric stage 46 becomes unnecessary. The three-dimensional shape data measured by the three-dimensional measuring instrument 48 can include not only the three-dimensional shape data of the processed part but also the surface temperature distribution data and color distribution data of the processed part. This is because the surface temperature distribution data and color distribution data of the processed part indicate the temperature change due to the irradiation of the laser light and are considered to reflect the ease of change in the three-dimensional shape and the cause of the change. Here, the color distribution data includes not only the data of the normal color distribution but also Raman spectrum data, light reflectance spectrum data, and the like.

[0032] The system control unit 50 is configured as a microcomputer centered on a CPU, although not shown in the figure. In addition to the CPU, it includes a ROM, a RAM, a flash memory, a GPU (Graphics Processing Unit), input / output ports, communication ports, and the like. Functionally, the system control unit 50 has an input unit 52 such as a keyboard and a mouse, and a machine learning unit 54. Input data from the input unit 52, the position signal of the workpiece 10 on the electric stage 46, the three-dimensional measurement data from the three-dimensional measuring instrument 48, and the like are input to the system control unit 50 via the input port. Further, drive control signals to the electric stage 46, drive control signals to the three-dimensional measuring instrument 48, and the like are output from the system control unit 50 via the output port. In addition, the system control unit 50 communicates with the laser control unit 38 and acquires the laser light parameters of the laser light output from the processing laser irradiation device 30.

[0033] The machine learning unit 54 performs deep learning using learning data including the material of the workpiece 10, the laser light parameters of the laser light irradiated on the processed portion of the workpiece 10, and the three-dimensional shape data of the processed portion before and after the irradiation of the laser light. When the material of the workpiece 10, the laser light parameters of the laser light irradiated on the processed portion of the workpiece 10, and the three-dimensional shape data of the processed portion before the irradiation of the laser light are given as the first input data, the relationship for estimating the three-dimensional shape data of the processed portion after the irradiation of the laser light as the first output data is obtained as the first learning result and stored as the learning result 56. FIG. 2 is a schematic diagram schematically showing the deep learning of the embodiment. As shown in the figure, the deep learning of the embodiment is based on a convolutional neural network with a teacher because the three-dimensional shape data of the processed portion after the irradiation of the laser light is input, and the input laser light parameters become the feature vectors. Further, in the embodiment, based on the first learning result, when the material of the workpiece 10 and the three-dimensional shape data of the processed portion before and after the irradiation of the laser light are given as the second input data, the relationship for estimating the laser light parameters of the laser light to be irradiated on the processed portion of the workpiece 10 as the second output data is obtained as the second learning result and also stored as the learning result 56.

[0034] The laser processing system 20 of the embodiment functions as a laser processing system that performs laser processing by irradiating a laser beam onto the processing portion of the object 10 placed on the electric stage 46. In addition, every time the object 10 placed on the electric stage 46 is irradiated with a laser beam, the electric stage 46 and the three-dimensional measuring instrument 48 are operated to acquire three-dimensional shape data, and it also functions as a machine learning system used in a laser processing system that performs deep learning in the machine learning unit 54. Further, the laser processing system 20 of the embodiment inputs the first input data and the second input data from the input unit 52 or the like, applies the first learning result and the second learning result in the learning result 56 of the machine learning unit 54 to the first input data and the second input data, and also functions as a simulation device used in a laser processing system that outputs the first output data and the second output data.

[0035] Next, the learning process when the laser processing system 20 of the embodiment is made to function as a machine learning system used in a laser processing system will be described. FIG. 3 is a flowchart showing an example of the learning process executed by the system control unit 50 of the laser processing system 20 of the embodiment. In the learning process, first, the workpiece 10 is set on the electric stage 46 (step S100), and three-dimensional shape data of the processing part before laser processing is acquired by the three-dimensional measuring instrument 48 (step S110). Random laser light parameters are set at random irradiation positions within a predetermined range (for example, a range where laser light irradiation is possible) (step S120), the laser light with the set laser light parameters is irradiated at the set irradiation positions (step S130), and three-dimensional shape data after the laser light is irradiated is acquired by the three-dimensional measuring instrument 48 (step S140). The processes of steps S120 to S140 are repeatedly executed until the number of irradiation times Nf reaches a threshold value Nfref (for example, 100 times or 200 times) (step S150). FIG. 4 shows an example of the three-dimensional shape data for each laser light irradiation when the laser light with random laser light parameters is irradiated seven times at random irradiation positions when acquiring learning data. In the figure, the color density indicates the depth of the processing part. As shown in the figure, it can be seen that the laser light with random laser light parameters is irradiated at random irradiation positions on the workpiece 10. Note that the pulse energy of the laser light used in the experiment of FIG. 4 was random within the range of 0.1 μJ to 100 μJ.

[0036] Such processing (the processing in steps S120 to S150) is repeatedly executed from the process of setting the new workpiece 10 in step S100 on the electric stage 46 until the number of repetitions Nn of the workpiece 10 of the same material reaches the threshold value Nnref (for example, 20 times or 30 times) (step S160), and until the number of materials Nm of the workpiece 10 reaches the threshold value Nmref (for example, 5 or 10) (step S170). Examples of the material of the workpiece 10 include quartz, copper, aluminum, carbon fiber reinforced plastic (CFRP), sapphire, and silicon. Now, if the threshold value Nfref of the irradiation number Nf is 200 times, the threshold value Nnref of the number of repetitions Nn of the workpiece 10 of the same material is 20 times, and the threshold value Nmref of the number of materials Nm of the workpiece 10 is 5, the number of data of the learning data is 20,000 (200×20×5). Note that the learning data is the laser light parameter and the three-dimensional shape data before and after irradiating the laser light.

[0037] After obtaining the learning data in this way, supervised deep learning is performed on the obtained learning data, with the laser light parameter as the feature vector z and the three-dimensional shape data after irradiating the laser light as the solution of the example (step S180). Then, when the material of the workpiece 10, the laser light parameter of the laser light irradiated to the processed part of the workpiece 10, and the three-dimensional shape data of the processed part before irradiating the laser light are given as the first input data, a relationship for estimating the three-dimensional shape data of the processed part after irradiating the laser light as the first output data is obtained as the first learning result and stored (step S190). Subsequently, when the material of the workpiece 10 and the three-dimensional shape data of the processed part before and after irradiating the laser light are given as the second input data based on the first learning result, a relationship for estimating the laser light parameter of the laser light to be irradiated to the processed part of the workpiece 10 as the second output data is derived as the second learning result (step S200), stored (step S210), and the learning process is terminated.

[0038] Next, the simulation results and experimental results based on the learning results obtained by the laser processing system 20 of the embodiment will be described. In the simulation and experiment, laser light with a wavelength of 800 nm, a spot diameter of 26 μm, a pulse width of 35 fs, and pulse energies of 40 μJ, 50 μJ, and 60 μJ was moved 10 μm at a time to the workpiece 10 made of quartz as shown in FIG. 5, and irradiated seven times in a folded-back manner in sequence, and the processing depth at each position in the folded-back direction of the processed part for each irradiation of the laser light was obtained. The experimental results are the measured values by the three-dimensional measuring instrument 48. FIG. 6 shows the simulation results and experimental results of the number of irradiations of the laser light when the pulse energy is 50 μJ and the processing depth at each position in the folded-back direction of the processed part. As shown in the figure, the simulation results and the experimental results are in good agreement. FIG. 7 shows the simulation results and experimental results of the processing depth at each position in the folded-back direction of the processed part when the laser light is irradiated 100 times with pulse energies of 40 μJ, 50 μJ, and 60 μJ. In FIG. 7, the scale of the vertical axis when the pulse energies are 40 μJ, 50 μJ, and 60 μJ is changed, but the simulation results and the experimental results are in good agreement. From this, it can be understood that the machine learning and learning results by the laser processing system 20 of the embodiment are appropriate.

[0039] FIG. 8 is an explanatory diagram showing the simulation results and experimental results of the processing depth at each position when laser light with a wavelength of 800 nm, a spot diameter of 26 μm, a pulse width of 35 fs, and a pulse energy of 50 μJ was irradiated 500 times while moving the irradiation position 12.5 μm along a star shape with a side length of 125 μm to the workpiece 10 made of silicon. In the figure, the darkness of the color represents the processing depth. As shown in the figure, in the simulation results and the experimental results, the processing depth shape (average depth, shape of the outer peripheral part, points where it becomes deeper at the intersections of the line segments) is in good agreement.

[0040] In the laser processing system 20 of the embodiment described above, deep learning is performed using the material of the object 10 to be processed, the laser light parameters indicating the characteristics of the laser light irradiated on the object 10 to be processed, and the three-dimensional shape data of the processed part before and after irradiating the object 10 to be processed with the laser light. When the material of the object 10 to be processed, the laser light parameters of the laser light irradiated on the processed part of the object 10 to be processed, and the three-dimensional shape data of the processed part before irradiation with the laser light are given as the first input data, a relationship for estimating the three-dimensional shape data of the processed part after irradiation with the laser light as the first output data is obtained as the first learning result. Thereby, the relationship among the three-dimensional shape data of the processed part before and after irradiation with the laser light during laser processing, the material of the object 10 to be processed, and the laser light parameters of the laser light can be learned. Also, the three-dimensional shape of the processed part of the object 10 to be processed that has changed due to irradiation with the laser light can be estimated.

[0041] Also, in the laser processing system 20 of the embodiment, when the material of the object 10 to be processed and the three-dimensional shape data of the processed part before and after irradiation with the laser light are given as the second input data based on the first learning result, a relationship for estimating the laser light parameters of the laser light to be irradiated on the processed part of the object 10 to be processed as the second output data is derived as the second learning result. Thereby, the laser light parameters of the laser light necessary to make the processed part of the object 10 to be processed into a desired shape can be estimated.

[0042] In the learning process in the laser processing system 20 of the embodiment, although learning data is to be acquired for the object 10 to be processed with different materials in addition to the object 10 to be processed with the same material, it may be possible to acquire learning data only for the object 10 to be processed with the same material.

[0043] In the learning process of the laser processing system 20 of the embodiment, although the three-dimensional shape data of the processing part is acquired every time the laser beam is irradiated, it is also possible to acquire the three-dimensional shape data of the processing part every time the laser beam is irradiated multiple times, or every time the laser beam is irradiated a random number of times. In this case, the laser beam irradiated multiple times or the laser beam irradiated a random number of times may have the same laser beam parameters or different laser beam parameters.

[0044] In the machine learning of the laser processing system 20 of the embodiment, although 800 nm is used as the wavelength of the laser beam, it is also applicable to laser beams in the wavelength range in which photoinduced electron excitation affects the processing, that is, the wavelength range of 193 nm to 5 μm.

[0045] In the machine learning in the laser processing system 20 of the embodiment, although 35 fs is used as the pulse width of the laser beam, it is also applicable to laser beams in the range from short pulses to ultrashort pulses, that is, the pulse width range of 10 fs to 100 ns.

[0046] In the laser processing system 20 of the embodiment, although machine learning is performed when the learning process in FIG. 3 is executed, after the learning process in FIG. 3 is completed, machine learning may be performed using the learning data accumulated so far every time laser processing is performed or at a predetermined timing.

[0047] In the laser processing system 20 of the embodiment, a processing laser irradiation device 30, electric stages 46, a three-dimensional data measuring instrument 48, and a system control unit 50 are provided, and three-dimensional shape data of the processing part is acquired every time laser light is irradiated to acquire learning data. However, when the present invention is in the form of a machine learning device, in addition to having the same hardware configuration as the laser processing system 20 of the embodiment, it may also have a hardware configuration including only the system control unit 50. In this case, for the processing in steps S100 to S170 of the learning process in FIG. 3, the material of the processing object 10, the laser light parameters indicating the characteristics of the laser light irradiated to the processing object 10, and the three-dimensional shape data of the processing part before and after irradiating the processing object 10 with laser light are acquired in advance by another device, and these learning data may be inputted.

[0048] When the present invention is in the form of a program that causes a computer to function as a machine learning device, in addition to the system control unit 50 that functions as this machine learning device, it may also be a program to be executed by the system control unit 50 when configuring the laser processing system 20 of the embodiment including a processing laser irradiation device 30, electric stages 46, a three-dimensional data measuring instrument 48, etc. As the program in this case, it may be the flowchart of machine learning in FIG. 3. Also, if it is configured without a processing laser irradiation device 30, electric stages 46, a three-dimensional data measuring instrument 48, etc., that is, if it includes only a microcomputer corresponding to the system control unit 50, for the processing in steps S100 to S170 of the learning process in FIG. 3, the material of the processing object 10, the laser light parameters indicating the characteristics of the laser light irradiated to the processing object 10, and the three-dimensional shape data of the processing part before and after irradiating the processing object 10 with laser light are acquired in advance by another device, and these learning data may be inputted.

[0049] When the present invention is in the form of a simulation device, it may perform deep learning by inputting learning data, or it may store only the learning results obtained by deep learning.

[0050] An example of an application case when the present invention is in the form of a simulation device will be described. FIG. 9 is a flowchart showing an example of an optimization simulation process for simulating the optimization of laser beam parameters and laser irradiation positions in the laser processing of an object to be processed. In this process, first, the material and target shape of the object to be processed are set (step S300), and initial values of processing conditions are set (step S310). The processing conditions are the laser beam parameters and the laser irradiation position. As the initial values of the laser beam parameters, the wavelength, pulse width, pulse intensity, spot diameter, number of pulses, and fluence are determined, and the laser irradiation position may be determined according to the number of pulses to be irradiated and the target shape.

[0051] When the initial values of the processing conditions are set, the material of the object to be processed, the laser beam parameters, and the laser irradiation position are used as input data, and a machining simulation is performed based on the input data (step S320). This machining simulation applies the first learning result obtained in the learning process of FIG. 3 to the input data to calculate the three-dimensional shape data (shape during machining) of the machined part after irradiation with the laser beam based on the input data. Next, the difference between the obtained three-dimensional shape data (shape during machining) of the machined part and the target shape is calculated, and it is determined whether this difference is within the allowable range (step S330). When it is determined that the difference is not within the allowable range, it is determined whether the cutting by the laser processing of the machined part is overcut (step S340). When it is determined that it is not overcut, the processing conditions are adjusted based on the difference between the shape during machining and the target shape (step S360). The adjustment of the processing conditions may be, for example, using the initial value for the laser beam parameters and adjusting only the laser irradiation position so that the laser beam is irradiated to a location where the difference between the shape during machining and the target shape is large, or both the laser beam parameters and the laser irradiation position may be adjusted. Then, when the processing conditions are adjusted, the process returns to step S320 where the machining simulation is applied with the adjusted processing conditions as the input data. Such adjustment of the processing conditions and the machining simulation are repeated until the difference between the shape during machining and the target shape is within the allowable range.

[0052] If it is determined in step S340 that overcutting has occurred, the machining simulation performed immediately before is regarded as not having been performed (step S350). Based on the difference between the shape during machining and the target shape as the result of the machining simulation performed before that, an adjustment different from the adjustment of the previous machining conditions is made (step S360), and the process returns to step S320. This can prevent overcutting from occurring.

[0053] When it is determined in step S330 that the difference between the shape during machining and the target shape is within the allowable range, it is determined whether the number of repetitions from step S310 to S370 has reached a predetermined number of calculations (step S370). When it is determined that the predetermined number of calculations has not been reached, the process returns to the process of setting the initial value of the machining conditions in step S310. In this case, as the initial value of the machining conditions, it is conceivable to increase (or decrease) the fluence of the laser beam parameters by a predetermined amount each time, or to increase (or decrease) the number of pulses by a predetermined number, etc.

[0054] When it is determined in step S370 that the number of repetitions from step S310 to S370 has reached the predetermined number of calculations, each simulation result obtained by repeating steps S310 to S370 a predetermined number of times is output (step S380), and the optimization simulation process ends. As for the optimization, it can be determined based on the degree of difference (machining accuracy) between the three-dimensional shape data by laser processing and the target shape, the energy required for laser processing, the time required for laser processing, etc. for each simulation result.

[0055] A simulation was performed when laser processing a cylindrical hole with a diameter of 80 μm and a depth of 5 μm as the target shape on a workpiece 10 made of silicon. The laser irradiation locations were made to be approximately uniform on the plane. Fig. 10 shows an example of the simulation results of the relationship between the pulse energy and the total energy and the relationship between the pulse energy and the roughness of the bottom surface. As shown in the figure, the total energy of the laser processing is maximum when the pulse energy is near 25 μJ. The roughness of the bottom surface is minimum when the pulse energy is near 30 μJ. Fig. 11 shows the total fluence irradiated by the laser. In the figure, the outermost side is the value 0 and it is maximum at the thinnest location slightly inside the circular shape of the dashed line. Fig. 12 shows the 3D shape data based on the simulation results and the 3D shape data of the workpiece laser processed using the processing conditions of the simulation results. The left side is the 3D shape data based on the simulation results, and the right side is the 3D shape data of the workpiece actually laser processed. As shown in the figure, it can be understood that the 3D shape data based on the simulation results well matches the 3D shape data of the workpiece actually laser processed.

[0056] From the above description, even when the present invention is in the form of a simulation apparatus, since the 3D shape data based on the simulation results accurately matches the 3D shape data of the workpiece actually laser processed, the optimization of the laser processing can be performed.

[0057] As described above, the embodiments for implementing the present invention have been described using examples. However, the present invention is not limited to such examples at all, and it goes without saying that the present invention can be implemented in various forms without departing from the gist of the present invention.

Industrial Applicability

[0058] The present invention can be used in the manufacturing industry of laser processing systems and the like.

Claims

1. A machine learning method used in a laser processing system for performing ablation processing by irradiating a workpiece with a laser beam, comprising: performing deep learning using the material of the workpiece, laser beam parameters indicating the characteristics of the laser beam irradiated on the workpiece, pre-processing part data and post-processing part data reflecting the three-dimensional shape associated with the laser processing of the processing part before and after irradiating the workpiece with the laser beam, and obtaining, as one of the learning results, a first relationship in which the post-processing part data after irradiating the laser beam related to the input data is used as the output data when the material of the workpiece, the pre-processing part data, and the laser beam parameters are used as the input data; the laser beam parameters at least partially include each of wavelength, pulse width, pulse intensity, number of pulses, and fluence; the pre-processing part data and the post-processing part data include either a part of a set of numerical values indicating the processing state at each position in the three-dimensional space including the workpiece, or a part of a set of representative values obtained as a result of projecting the processing state at each position in the three-dimensional space including the workpiece onto a specific coordinate axis; A machine learning method characterized by the above.

2. A machine learning method used in a laser processing system for performing ablation processing by irradiating a workpiece with a laser beam, comprising: performing deep learning using the material of the workpiece, laser beam parameters indicating the characteristics of the laser beam irradiated on the workpiece, pre-processing part data and post-processing part data reflecting the three-dimensional shape associated with the laser processing of the processing part before and after irradiating the workpiece with the laser beam, and obtaining, as one of the learning results, a first relationship in which the post-processing part data after irradiating the laser beam related to the input data is used as the output data when the material of the workpiece, the pre-processing part data, and the laser beam parameters are used as the input data; the laser beam parameters at least partially include each of wavelength, pulse width, pulse intensity, number of pulses, and fluence; the pre-processing part data and the post-processing part data include a part of a set of numerical values representing the processing depth at each position of the workpiece; A machine learning method characterized by the above.

3. The machine learning method according to Claim 1 or 2, Based on the first relationship, when the material of the object to be processed, the pre-processing part data, and the post-processing part data are used as input data, the laser beam parameters of the laser beam to be irradiated, which are necessary to change the state of the processing part of the post-processing part data from the state of the processing part of the pre-processing part data related to the input data, are used as output data, and a second relationship is obtained as one of the learning results. Machine learning method.

4. The machine learning method according to any one of Claims 1 to 3, wherein the pre-processing part data and the post-processing part data include a part of either the surface temperature distribution data of the processing part or the color distribution data of the processing part. Machine learning method.

5. A simulation device used in a laser processing system that irradiates an object to be processed with a laser beam to perform ablation processing, wherein the output data is output for the input data by using the learning result obtained by the machine learning method according to any one of Claims 1 to 4. The simulation device is characterized in that.

6. The simulation device according to Claim 5, wherein the shape of the object to be processed before processing and the target shape are input, data obtained by adjusting the laser beam parameters and the laser beam irradiation location according to the difference between the shape during processing and the target shape is used as the input data, the output data as a result of the processing simulation obtained by applying the learning result to the input data is used as the shape during processing, and the processing simulation is repeated until the difference between the shape during processing and the target shape is within a predetermined range. Simulation device.

7. A laser processing system including a laser beam irradiation device for irradiating an object to be processed with a laser beam to perform ablation processing, a processing part data measurement device for measuring processing part data reflecting a three-dimensional shape associated with the laser processing of the object to be processed, and a control device for controlling the laser beam irradiation device for processing, wherein the control device performs learning using the machine learning method according to any one of Claims 1 to 4, and controls the laser beam irradiation device for processing based on the output data for the input data by using the learning result. The laser processing system is characterized in that.

8. A laser processing system comprising a laser beam irradiation device for performing ablation processing by irradiating a workpiece with a laser beam, and a control device for controlling the laser beam irradiation device for processing. The control device controls the laser beam irradiation device for processing based on the output data for the input data using the learning result obtained by the machine learning method according to any one of claims 1 to 4. A laser processing system characterized by this.

9. A program that causes a computer to function as a machine learning device used in a laser processing system, A step of inputting a plurality of data including the material of the workpiece, laser beam parameters indicating the characteristics of the laser beam irradiated on the workpiece, pre-processing part data and post-processing part data reflecting the three-dimensional shape associated with the laser processing of the processing part before and after irradiating the workpiece with the laser beam, A step of obtaining, as one of the learning results, a first relationship in which, by deep learning using the input plurality of data, when the material of the workpiece, the pre-processing part data, and the laser beam parameters are input data, the post-processing part data after the laser beam related to the input data is irradiated is used as the output data, having The laser beam parameters include at least a part of each of wavelength, pulse width, pulse intensity, number of pulses, and fluence. The pre-processing part data and the post-processing part data include either a part of a set of numerical values indicating the processing state at each position in the three-dimensional space including the workpiece, or a part of a set of representative values obtained as a result of projecting the processing state at each position in the three-dimensional space including the workpiece onto a specific coordinate axis. A program characterized by this.

10. A program that causes a computer to function as a machine learning device used in a laser processing system, A step of inputting a plurality of data including the material of the workpiece, laser beam parameters indicating the characteristics of the laser beam irradiated on the workpiece, pre-processing part data and post-processing part data reflecting the three-dimensional shape associated with the laser processing of the processing part before and after irradiating the workpiece with the laser beam, Using the input plurality of data, through deep learning, obtaining, as one of the learning results, a first relationship in which, when the material of the object to be processed, the pre-processing part data, and the laser beam parameters are input data, the post-processing part data after the laser beam related to the input data is irradiated is used as output data; having; the laser beam parameters at least partially include each part of wavelength, pulse width, pulse intensity, number of pulses, and fluence; the pre-processing part data and the post-processing part data include a part of a set of numerical values representing the processing depth at each position of the object to be processed; A program characterized by the above.

11. A program according to Claim 9 or 10, based on the first relationship, when the material of the object to be processed, the pre-processing part data, and the post-processing part data are input data, obtaining, as one of the learning results, a second relationship in which the laser beam parameters of the laser beam to be irradiated necessary to change the state of the processing part of the post-processing part data from the state of the processing part of the pre-processing part data related to the input data are used as output data; A program having the above.

Images