Machine learning device, machine learning method, processing system and processing method

Abstract

To train a processing model so that the model can determine a processing condition corresponding to a desired processed result, with high accuracy.SOLUTION: A storing device 13 stores simulated results showing behavior which is simulated of work-pieces 5a and 5b in a step of processing the work-pieces 5a and 5b. A processor 11 controls a processing device 2 so that the device processes at least either of the work-pieces 5a and 5b under a plurality of processing conditions. The processor 11 makes an inspecting device 3 inspect a work-piece 5c processed under the plurality of processing conditions to obtain a plurality of processed results of the work-pieces 5a and 5b. The processor 11 trains a processing model so the model outputs corresponding behavior of the work-piece being processed and a corresponding processing condition when one processed result is inputted, on the basis of the plurality of processing conditions, the plurality of processed results and a plurality of behavior of the work-piece being processed including the simulated results.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a machine learning device, a machine learning method, a processing system, and a processing method. [Background technology]

[0002] In order to obtain a desired processing result by processing a workpiece (a workpiece) using a processing device, it is necessary to determine the optimal processing conditions to be set for the processing device. Since manually determining the optimal processing conditions takes a very long time, a technique is known that uses, for example, machine learning to search for processing conditions that correspond to the desired processing result.

[0003] Patent Document 1 discloses a machine learning method used in a laser processing system that performs ablation processing by irradiating a workpiece with laser light. This method learns the degree of processing of the processed part before and after irradiation with laser light during laser processing, the material of the workpiece, and parameters of the irradiated laser light. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Publication No. 2023-171479 Summary of the Invention [Problem to be solved by the invention]

[0005] When controlling the machining of a workpiece by a machining device using a machining model trained by machine learning, in order to control the machining of the workpiece by the machining device with high precision, it is necessary to train the machining model so that the machining conditions corresponding to the desired machining results are determined with high precision.

[0006] An object of the present disclosure is to provide a machine learning device and a machine learning method that can train a machining model so as to determine machining conditions corresponding to desired machining results with higher accuracy than conventional methods.

[0007] Another object of the present disclosure is to provide a processing system that controls the processing of a workpiece by a processing device using a processing model trained by machine learning, and that can determine processing conditions corresponding to desired processing results with higher accuracy than conventional processing systems.

[0008] Another object of the present disclosure is to provide a processing method that controls the processing of a workpiece by a processing device using a processing model trained by machine learning, and that can determine processing conditions corresponding to a desired processing result with higher accuracy than conventional methods. [Means for solving the problem]

[0009] A machine learning device according to one aspect of the present disclosure includes: a first communication device connected to a processing device that processes at least one workpiece under variable processing conditions and an inspection device that inspects the workpiece processed by the processing device; a first storage device that stores a simulation result indicating a simulated behavior of the workpiece during the process of machining the workpiece; a first processor; The first processor Controlling the machining device so as to machine at least one workpiece under each of a plurality of machining conditions; The inspection device inspects each of the workpieces machined under the plurality of machining conditions to obtain a plurality of machining results of the workpieces; A machining model is trained based on the plurality of machining conditions, the plurality of machining results, and a plurality of in-machining behaviors that indicate the behavior of the workpiece during the process of machining the workpiece, the plurality of in-machining behaviors including the simulation results, so that when one machining result is input, a corresponding in-machining behavior and corresponding machining conditions are output. [Effects of the Invention]

[0010] A machine learning device according to one aspect of the present disclosure can train a machining model so as to determine machining conditions corresponding to desired machining results with higher accuracy than conventional methods. [Brief explanation of the drawings]

[0011] [Figure 1] 1 is a block diagram showing a configuration of a processing system 100 according to a first embodiment. [Figure 2] 2 is a perspective view showing the dimensions of workpieces 5a and 5b machined by the machining device 2 of FIG. 1. FIG. [Figure 3] FIG. 2 is a block diagram showing the configuration of a control device 1 in FIG. [Figure 4] 4 is a flowchart showing a machine learning process executed by the processor 11 of FIG. 3. [Figure 5] FIG. 5 is a schematic diagram showing a machining model generated by executing the machine learning process of FIG. 4. [Figure 6] 4 is a flowchart showing a processing process executed by the processor 11 of FIG. 3. [Figure 7] FIG. 10 is a block diagram showing a configuration of a processing system 100A according to a second embodiment. [Figure 8] 8 is a diagram showing waveforms of visible light, thermal radiation light, and reflected light detected by welding process monitor 6 of FIG. 7, and a waveform of laser light generated by laser source device 21. FIG. [Figure 9] 8 is a flowchart showing a machine learning process executed by the control device 1A of FIG. 7. [Figure 10] 8 is a flowchart showing a processing process executed by the control device 1A of FIG. 7. DETAILED DESCRIPTION OF THE INVENTION

[0012] Hereinafter, the embodiments will be described in detail with reference to the drawings as appropriate. However, more detailed explanation than necessary may be omitted. For example, detailed explanation of already well-known matters, redundant explanation of substantially identical configurations, etc. may be omitted. This is to avoid unnecessary redundancy in the following explanation and to facilitate understanding by those skilled in the art. Note that components with the same reference numerals have the same functions in each embodiment.

[0013] It should be noted that the present disclosure provides the accompanying drawings and the following description for the understanding of those skilled in the art, and is not intended to limit the subject matter described in the claims.

[0014] [First embodiment] [Configuration of the first embodiment] 1 is a block diagram showing the configuration of a processing system 100 according to a first embodiment. The processing system 100 includes a control device 1, a processing device 2, an inspection device 3, and a conveyance device 4. The processing system 100 controls the processing of workpieces (machined objects) 5a and 5b by the processing device 2 using a processing model trained by machine learning.

[0015] The control device 1 controls the machining device 2 so as to machine the workpieces 5a and 5b under predetermined machining conditions. The machining conditions include, for example, a machining position, a machining power, and a machining profile.

[0016] The processing device 2 operates under the control of the control device 1 and processes at least one workpiece 5a, 5b under variable processing conditions. In this embodiment, the processing device 2 is, for example, a laser welding device that mechanically connects two workpieces 5a, 5b to each other by welding them with laser light, thereby creating one processed workpiece 5c.

[0017] The inspection device 3 inspects the workpiece 5c machined by the processing device 2, obtains the processing results of the workpiece 5c, and notifies the control device 1 of the processing results. The processing results include, for example, the surface shape, tensile strength, and cross-sectional shape of the workpiece 5c.

[0018] The conveying device 4 conveys unprocessed workpieces 5a, 5b and processed workpiece 5c in the processing system 100. The conveying device 4 supplies the unprocessed workpieces 5a, 5b to the processing device 2, supplies the workpiece 5c processed by the processing device 2 to the inspection device 3, and discharges the processed workpiece 5c outside the processing system 100 after inspection.

[0019] FIG. 2 is a perspective view showing the dimensions of workpieces 5a and 5b processed by the processing apparatus 2 of FIG. 1. Each of workpieces 5a and 5b has a length of 15.0 mm in the y direction, a width of 5.5 mm in the x direction, and a thickness of 0.3 mm in the z direction. Workpieces 5a and 5b are overlapped by 10 mm along the y direction and each has a 5 mm gripping portion at the end in the y direction. Workpieces 5a and 5b are welded to each other by irradiating the overlapping area with a laser beam from above. The laser beam spot moves by 2 mm along the x direction. The gripping portions of workpieces 5a and 5b are gripped when inspection apparatus 3 inspects the tensile strength of workpiece 5c. Workpieces 5a and 5b may be members made of the same material with the same properties, or members made of different materials with different properties. In the latter case, for example, the workpiece 5a may be a member made of iron having a laser light absorption rate of 40% (for a wavelength of 1070 nm) and a melting point of 1700 K, and the workpiece 5b may be a member made of copper having a laser light absorption rate of 20% (for a wavelength of 1070 nm) and a melting point of 1358 K.

[0020] When the processing device 2 is a laser welding device, the processing conditions include the position where the laser beam is irradiated on the workpieces 5a, 5b, the power of the laser beam, and the profile of the laser beam. In this case, the processing conditions may also include the scanning speed of the laser beam. In the example of FIG. 2, the laser beam moves over 2 mm in the x direction. The power of the laser beam varies, for example, in the range of 70 to 4500 W. The profile of the laser beam includes, for example, the beam spot diameter, shape, and intensity distribution. In this case, the beam spot diameter may have a fixed value, for example, 102 μm, or may have a variable value due to defocusing. The scanning speed of the laser beam varies in the range of 100 mm per second to 2000 mm per second.

[0021] The processing conditions may include the dimensions and / or material of the workpieces 5a and 5b.

[0022] FIG. 3 is a block diagram showing the configuration of the control device 1 of FIG. 1. The control device 1 includes a bus 10, a processor 11, a memory 12, a storage device 13, an input device 14, a display device 15, and a communication device 16. The processor 11 controls the overall operation of the control device 1. The memory 12 temporarily stores programs and data necessary for the operation of the control device 1. The storage device 13 is a non-volatile storage medium that stores programs and data necessary for the operation of the control device 1. The input device 14 receives user input to control the operation of the machining system 100. The input device 14 includes, for example, a keyboard and / or a pointing device. The display device 15 displays the status of the machining system 100, such as the machining conditions set for the machining device 2 and the machining results acquired by the inspection device 3. The communication device 16 is connected to the machining device 2 and the inspection device 3 via a communication line, sets the machining conditions for the machining device 2, and acquires the machining results from the inspection device 3. The processor 11, the memory 12, the storage device 13, the input device 14, the display device 15, and the communication device 16 are connected to each other via a bus 10.

[0023] The processor 11 executes the machine learning process of FIG. 4 , thereby training a machining model based on a plurality of machining conditions, a plurality of machining results, and a plurality of in-machining behaviors so that when one machining result is input, a corresponding in-machining behavior and corresponding machining conditions are output. The in-machining behavior indicates the behavior of the workpieces 5a, 5b in the process of machining the workpieces 5a, 5b. The in-machining behavior includes, for example, simulation results indicating simulated behavior of the workpieces 5a, 5b in the process of welding the workpieces 5a, 5b. The simulation results may be calculated in advance by the processor 11 or another external computing device and stored in the storage device 13. The processor 11 trains the machining model using deep learning or any other machine learning method. The trained machining model is stored in the storage device 13.

[0024] If the processing device 2 is a laser welding device, the processing behavior may indicate, for example, the state of the workpieces 5a, 5b melted by laser light. For example, if the workpieces 5a, 5b are welded by scanning laser light, the processing behavior may indicate the state of the workpieces 5a, 5b in a cross section including the direction of laser light propagation and a direction perpendicular to the surface of the workpieces 5a, 5b irradiated with the laser light. The processing behavior may include the contour shape or temperature distribution in the cross section of the welded portion. The processing behavior may also include the distribution of melted and unmelted portions in the cross section of the welded portion. In this case, the processing behavior may include the distribution of melted and unmelted portions at each of multiple time points, or may include the distribution of portions that are at least temporarily melted and not melted at all between the start and end of processing. For example, the depth of the melt can be determined from an image showing the cross-sectional state of the welded portion.

[0025] As described above, the behavior during processing includes simulation results. If the processing device 2 is a laser welding device, the behavior during processing may include simulation results from a welding simulation. The simulation results may include images or numerical data showing the state of the workpieces 5a and 5b melted by the laser beam, such as cross-sectional images of the welded portion. The welding simulation is a fluid simulation that theoretically calculates the behavior of the molten pool during laser welding based on processing conditions, including the position where the laser beam is irradiated on the workpieces 5a and 5b, the power of the laser beam, the profile of the laser beam, and the dimensions and material of the workpieces 5a and 5b. Performing a welding simulation allows observation of the behavior of the molten pool, which is not actually observable, and also allows cross-sectional images of the workpiece to be obtained at any time step. The time step may be set to, for example, 1 / 100 of the total processing time, i.e., 0.07 seconds if the total processing time is 7 seconds.

[0026] The processor 11 also executes the processing process of FIG. 6, thereby receiving the desired processing results of the workpieces 5a and 5b from the input device 14, determining processing conditions corresponding to the desired processing results using the processing model, and controlling the processing device 2 to process the workpieces 5a and 5b under the determined processing conditions.

[0027] The control device 1 may be a general-purpose computer or a dedicated device.

[0028] Referring back to FIG. 1, the processing device 2 includes a laser light source device 21 and an optical system 22.

[0029] The laser light source device 21 outputs laser light by collimating a beam generated by laser oscillation. The laser light source device 21 includes, for example, a single-mode fiber laser that continuously oscillates at a wavelength of 1070 nm.

[0030] The optical system 22 includes one or more lenses and one or more mirrors, and irradiates the workpieces 5a and 5b with the laser light output from the laser light source device 21. The optical system 22 may include, for example, a mirror that reflects 90% or more of light with a wavelength of 1070 nm. The optical system 22 may also include, for example, a total reflection mirror, a galvanometer mirror, a lens, and an fθ lens. In this case, the galvanometer mirror and the fθ lens are disposed between the total reflection mirror and the lens, and the galvanometer mirror and the fθ lens are controlled to scan the laser light over the workpieces 5a and 5b.

[0031] The processing device 2 irradiates the stacked workpieces 5a and 5b with laser light, which heats and partially melts the workpieces 5a and 5b, forming a molten pool. After that, when the melted parts of the workpieces 5a and 5b cool, the workpieces 5a and 5b are mechanically connected to each other.

[0032] The inspection device 3 includes a surface shape measuring device 31 , a strength measuring device 32 , and a cross-sectional shape measuring device 33 .

[0033] The surface profile measuring device 31 measures the surface profile of the processed workpiece 5c in three dimensions. The surface profile measuring device 31 may be equipped with, for example, a camera that visually measures the three-dimensional position of each point on the workpiece 5c. When the processing device 2 is a laser welding device, the welded portion of the workpiece 5c generally has an uneven surface compared to its surrounding areas. For example, the welded portion of the workpiece 5c has a width of approximately 130 to 330 μm and a maximum height of approximately 400 μm from the surface of the workpiece 5c. The surface profile measuring device 31 is configured to have sufficient accuracy to measure the surface profile of the workpiece 5c including such welded portions.

[0034] The strength measuring device 32 measures the strength, for example, tensile strength, of the processed workpiece 5c. For example, the strength measuring device 32 grips a portion of the processed workpiece 5c corresponding to the workpiece 5a and a portion of the processed workpiece 5c corresponding to the workpiece 5b, and pulls them in opposite directions to measure the tensile strength at which the workpiece 5c breaks. The tensile strength varies, for example, in the range of 0 to 320 N. A tensile strength of 0 N represents a state in which no welding has been performed at all.

[0035] Cross-sectional shape measuring device 33 measures the cross-sectional shape of machined workpiece 5c. Machined workpiece 5c is cut into two parts 5c1 and 5c2, and the cross sections of cut workpieces 5c1 and 5c2 are polished. Cross-sectional shape measuring device 33 may include a camera that photographs one cross section of cut workpieces 5c1 and 5c2.

[0036] The inspection device 3 notifies the control device 1 of the surface shape, strength (tensile strength), and cross-sectional shape of the machined workpiece 5c as the machining results.

[0037] The conveyance device 4 may include a processing stage on which the workpieces 5a and 5b are fixed by a jig. The processing stage is movable, for example, in three mutually orthogonal axial directions, i.e., the x-axis direction, the y-axis direction, and the z-axis direction, within a predetermined distance, for example, within a range of 200 mm, 200 mm, and 50 mm, respectively. By moving the processing stage, the workpieces 5a and 5b are equivalently scanned with the laser light.

[0038] [Operation of the first embodiment] FIG. 4 is a flowchart showing the machine learning process executed by the processor 11 of FIG.

[0039] In step S1, the processor 11 sets predetermined processing conditions.

[0040] In step S2, the processor 11 controls the machining device 2 so as to machine the workpieces 5a and 5b under the set machining conditions.

[0041] In step S3, the processor 11 uses the inspection device 3 to inspect the machined workpiece 5c and obtains the machining results corresponding to the set machining conditions.

[0042] In step S4, processor 11 determines whether or not a predetermined number of attempts in steps S2 and S3 have been completed, and if YES, proceeds to step S6, and if NO, proceeds to step S5.

[0043] In step S5, the processor 11 sets other processing conditions and repeats steps S2 to S4.

[0044] By executing steps S1 to S5, the processor 11 can obtain a plurality of machining results corresponding to a plurality of machining conditions.

[0045] In step S6, the processor 11 acquires from the storage device 13 a plurality of simulation results showing the simulated behavior of the workpieces 5a, 5b during the process of machining the workpieces 5a, 5b under a plurality of machining conditions set when executing steps S1 to S5.

[0046] In step S7, the processor 11 trains the machining model based on a plurality of machining conditions, a plurality of machining results, and a plurality of in-machining behaviors including a plurality of simulation results so that when one machining result is input, it outputs a corresponding in-machining behavior and corresponding machining conditions.

[0047] FIG. 5 is a schematic diagram showing a machining model generated by executing the machine learning process of FIG. 4. The machining results, machining conditions, and in-machining behavior are provided as known training data for the machining model. The machining conditions include the machining position, machining power, and machining profile. The in-machining behavior includes the simulation results. The machining results include the surface shape, tensile strength, and cross-sectional shape of the workpiece 5c. The machining model is trained so that when one machining result is input, it outputs the corresponding in-machining behavior and corresponding machining conditions. Therefore, the machining results are explanatory variables of the machining model, and the machining conditions and in-machining behavior are target variables of the machining model.

[0048] The machine learning method of Patent Document 1 learns the state of the workpiece before and after irradiating the workpiece with laser light, but does not learn the state of the workpiece during irradiation. This is because, generally, the workpiece is hot during laser welding, and the processing time is extremely short, making it difficult to measure the state of the workpiece during laser welding. On the other hand, the processing system 100 according to the embodiment acquires simulation results showing the simulated behavior of the workpieces 5a, 5b during the welding process as the in-process behavior of the workpieces 5a, 5b. Therefore, the processing system 100 according to the embodiment trains a processing model based on processing conditions, processing results, and in-process behavior including the simulation results, thereby achieving higher accuracy of machine learning than when the processing model is trained based only on the processing conditions and processing results. By performing machine learning based on more information than conventional methods, the processing model according to the embodiment has improved generalization performance.

[0049] According to the machining system 100 of the embodiment, by training the machining model based on the machining conditions, machining results, and behavior during machining, it is possible to train the machining model so as to determine the machining conditions corresponding to the desired machining results with higher accuracy than conventionally.

[0050] FIG. 6 is a flowchart showing the processing executed by the processor 11 of FIG.

[0051] In step S11, the processor 11 sets predetermined evaluation conditions. There may be multiple processing conditions that correspond to a desired processing result. The evaluation conditions are conditions for determining the most appropriate processing conditions, preferably the optimal processing conditions, from among the multiple processing conditions.

[0052] The evaluation conditions include at least one parameter selected from the group consisting of one or more first parameters indicating machining conditions, one or more second parameters indicating behavior during machining, and one or more third parameters indicating machining results. The parameters indicating machining conditions include, for example, machining power. In this case, the evaluation conditions are set to be more appropriate as the machining power decreases. The parameters indicating behavior during machining include, for example, the depth of the molten pool. In this case, the evaluation conditions are set to be more appropriate as the depth of the molten pool decreases. The parameters indicating machining results include, for example, the width or irregularity of the molten pool, the minimum required tensile strength, etc. In this case, the evaluation conditions are set to be more appropriate as the width or irregularity of the molten pool decreases and the minimum required tensile strength decreases.

[0053] In step S12, the processor 11 acquires the desired machining results of the workpieces 5a and 5b from the input device 14. The desired machining results may be different from the machining results acquired in step S3 of FIG.

[0054] In step S13, the processor 11 uses the machining model to determine machining conditions corresponding to the desired machining result.

[0055] In step S14, processor 11 determines whether or not there are a plurality of machining conditions corresponding to the desired machining conditions, and if YES, proceeds to step S15, and if NO, proceeds to step S16.

[0056] In step S15, the processor 11 determines one of the plurality of machining conditions that is more appropriate, preferably the optimum machining condition, based on the evaluation condition.

[0057] In step S16, the processor 11 controls the machining device 2 so as to machine the workpieces 5a and 5b under the determined machining conditions.

[0058] In step S17, the processor 11 uses the inspection device 3 to inspect the machined workpiece 5c and obtains the machining results corresponding to the determined machining conditions.

[0059] In step S18, the processor 11 retrains the machining model based on one machining condition determined in step S13 or S15 and the machining result acquired in step S17.

[0060] According to the machining system 100 of the embodiment, by using a machining model trained based on machining conditions, machining results, and behavior during machining, it is possible to determine machining conditions corresponding to the desired machining results without requiring a long search.

[0061] According to the machining system 100 of the embodiment, by retraining the machining model based on the machining conditions corresponding to the desired machining result and the corresponding actual machining result (step S18 in FIG. 6), it is possible to determine the machining conditions corresponding to the desired machining result with higher accuracy.

[0062] Even if the simulation results do not reproduce with high accuracy the actual behavior of the workpieces 5a and 5b during the welding process of the workpieces 5a and 5b, the accuracy of machine learning is improved by training the processing model based on the behavior during processing, including the simulation results.

[0063] According to the machining system 100 of the embodiment, machining conditions corresponding to desired machining results can be determined with higher accuracy than ever before.

[0064] [Advantages of the first embodiment] The machine learning device according to the first embodiment includes a communication device 16 connected to a machining device 2 that machines at least one workpiece under variable machining conditions and an inspection device 3 that inspects a workpiece 5c machined by the machining device 2, a storage device 13 that stores simulation results indicating simulated behaviors of the workpieces 5a, 5b during the machining process, and a processor 11. The processor 11 controls the machining device 2 to machine at least one workpiece 5a, 5b under a plurality of machining conditions. The processor 11 inspects each of the workpieces 5c machined under the plurality of machining conditions using the inspection device 3 to obtain a plurality of machining results for the workpieces 5a, 5b. The processor 11 trains a machining model based on the plurality of machining conditions, the plurality of machining results, and a plurality of in-machining behaviors that indicate the behaviors of the workpieces 5a, 5b during the machining process, including simulation results, so as to output a corresponding in-machining behavior and corresponding machining conditions when one machining result is input.

[0065] This configuration makes it possible to train the machining model so as to determine machining conditions corresponding to desired machining results with higher accuracy than conventionally possible.

[0066] According to the machine learning device according to the first embodiment, the machining conditions may include a machining position, a machining power, and a machining profile.

[0067] With this configuration, the processing device 2 can process the workpieces 5a and 5b under variable processing conditions.

[0068] According to the machine learning device of the first embodiment, the inspection device 3 may include a surface shape measuring device 31 that three-dimensionally measures the surface shape of the machined workpiece 5c. In this case, the machining result includes the surface shape of the machined workpiece 5c.

[0069] With this configuration, the surface shape of the machined workpiece 5c can be acquired as the machining result.

[0070] According to the machine learning device of the first embodiment, the inspection device 3 may include a strength measuring device 32 that measures the strength of the machined workpiece 5c. In this case, the processing result includes the strength of the machined workpiece 5c.

[0071] With this configuration, the strength of the machined workpiece 5c can be obtained as the machining result.

[0072] According to the machine learning device of the first embodiment, the inspection device 3 may include a cross-sectional shape measuring device 33 that measures the cross-sectional shape of the machined workpiece 5c. In this case, the machining result includes the cross-sectional shape of the machined workpiece 5c.

[0073] With this configuration, the cross-sectional shape of the machined workpiece 5c can be obtained as the machining result.

[0074] According to the machine learning device of the first embodiment, the processing device 2 may be a laser welding device that welds the workpieces 5a and 5b with a laser. In this case, the simulation result indicates the simulated behavior of the workpieces 5a and 5b in the process of welding the workpieces 5a and 5b.

[0075] This configuration allows the workpieces 5a and 5b to be laser welded under processing conditions that correspond to the desired processing results.

[0076] The machining system 100 according to the first embodiment includes a machining device 2 that machines at least one workpiece 5a, 5b under variable machining conditions, and a control device that controls the machining device 2. The control device 1 includes a communication device 16 connected to the machining device 2, a storage device 13 that stores a machining model trained by the above-described machine learning method, an input device 14 that acquires desired machining results for the workpieces 5a, 5b, and a processor 11. The processor 11 receives the desired machining results for the workpieces 5a, 5b from the input device 14, determines machining conditions corresponding to the desired machining results using the machining model, and controls the machining device 2 to machine the workpieces 5a, 5b under the determined machining conditions.

[0077] This configuration makes it possible to determine the machining conditions corresponding to the desired machining results with higher accuracy than ever before.

[0078] According to the processing system 100 of the first embodiment, when there are multiple processing conditions corresponding to the desired processing result, the processor 11 may determine one processing condition corresponding to the desired processing result based on evaluation conditions including at least one parameter selected from one or more first parameters indicating the processing condition, one or more second parameters indicating the behavior during processing, and one or more third parameters indicating the processing result.

[0079] This configuration makes it possible to determine appropriate machining conditions from among a plurality of machining conditions that correspond to the desired machining results.

[0080] The machine learning method according to the first embodiment includes using a machining device 2 to machine at least one workpiece under a plurality of machining conditions. The method further includes using an inspection device 3 to inspect each of the workpieces machined under the plurality of machining conditions by the machining device 2 to obtain a plurality of machining results of the workpiece. The method further includes obtaining simulation results that indicate simulated behavior of the workpiece in the process of machining the workpiece. The method further includes training a machining model based on the plurality of machining conditions, the plurality of machining results, and a plurality of in-machining behaviors that indicate behavior of the workpiece in the process of machining the workpiece, the plurality of in-machining behaviors including the simulation results, so as to output a corresponding in-machining behavior and corresponding machining conditions when one machining result is input.

[0081] This configuration makes it possible to train the machining model so as to determine machining conditions corresponding to desired machining results with higher accuracy than conventionally possible.

[0082] The machining method according to the first embodiment includes obtaining a desired machining result of a workpiece. The method further includes determining machining conditions corresponding to the desired machining result using a machining model trained by the above-described machine learning method. The method further includes controlling a machining device 2 to machine the workpiece under the determined machining conditions.

[0083] This configuration makes it possible to determine the machining conditions corresponding to the desired machining results with higher accuracy than ever before.

[0084] [Second embodiment] [Configuration of the second embodiment] 7 is a block diagram showing the configuration of a processing system 100A according to a second embodiment. The processing system 100A includes a control device 1A and a processing device 2A instead of the control device 1 and the processing device 2 in FIG. 1, and further includes a welding process monitor 6.

[0085] The control device 1A has the same configuration as the control device 1 in Fig. 1. However, the control device 1A is further connected to a welding process monitor 6, and instead of the machine learning process in Fig. 3 and the processing process in Fig. 5, the control device 1A executes a machine learning process to be described later with reference to Fig. 9 and a processing process to be described later with reference to Fig. 10.

[0086] The processing device 2A includes an optical system 22A instead of the optical system 22 in Fig. 1. The optical system 22A transmits the returning light reflected from the portions of the workpieces 5a and 5b irradiated with the laser light to the welding process monitor 6. The optical system 22A includes a lens that receives the returning light, a folding mirror that changes the optical axis of the returning light by 90 degrees, and the like.

[0087] The welding process monitor 6 measures the behavior of the workpieces 5a, 5b during the machining process and obtains the measurement results. For example, the welding process monitor 6 separates the light returned from the workpieces 5a, 5b into visible light having a wavelength between 400 nm and 700 nm, reflected light having a wavelength of 1070 nm, and thermal radiation light having a wavelength between 1080 nm and 2000 nm. In this case, the welding process monitor 6 obtains the signal intensities of the separated visible light, thermal radiation, and reflected light as measurement results. The welding process monitor 6 converts the visible light, reflected light, and thermal radiation into electrical signals having voltage values ​​corresponding to the detected light intensities and sends them to the control device 1A.

[0088] 8 is a diagram showing waveforms of visible light, thermal radiation light, and reflected light detected by welding process monitor 6 of FIG. 7, and waveforms of laser light generated by laser light source device 21. In FIG. 8, the horizontal axis of each graph represents time, which is represented, for example, by the number of signal samplings. In FIG. 8, the vertical axis of each graph represents signal strength of visible light, thermal radiation light, and reflected light, and power of laser light, which is represented, for example, by a voltage value converted from the optical signal. Furthermore, Td is a period corresponding to one pulse of laser light, and indicates the period during which the power of the laser light reaches a steady value.

[0089] The signal intensity of the visible light fluctuates when foreign matter is present on the workpieces 5a and 5b and abnormal combustion of the foreign matter occurs, and also fluctuates when fumes are generated from the welded area. The signal intensity of the reflected light fluctuates depending on the surface shape of the molten pool, fluctuating when the workpieces 5a and 5b change from solid to liquid or from liquid to solid. The signal intensity of the thermal radiation light fluctuates depending on the temperature of the molten pool. The signal intensities of the visible light, thermal radiation light, and reflected light are further affected by the processing conditions, the state of the workpieces 5a and 5b, etc.

[0090] The behavior of the workpieces 5a, 5b during the machining process can be determined from the signal intensities of the visible light, the thermal radiation light, and the reflected light. Therefore, the control device 1A refers to the measurement results acquired by the welding process monitor 6 as the behavior during machining.

[0091] [Operation of the second embodiment] Fig. 9 is a flowchart showing the machine learning process executed by the control device 1A of Fig. 7. The process of Fig. 9 includes step S7A instead of step S7 of Fig. 4, and further includes step S21.

[0092] In step S21, the processor 11 of the control device 1A uses the welding process monitor 6 to measure the behavior of the workpieces 5a, 5b in the process of machining the workpieces 5a, 5b while machining the workpieces 5a, 5b (step S2) and acquires the measurement results.

[0093] By executing steps S1 to S5 and step S21 in FIG. 9, the processor 11 can obtain a plurality of processing results and a plurality of measurement results corresponding to a plurality of processing conditions.

[0094] In step S7A, the processor 11 trains the machining model based on a plurality of machining conditions, a plurality of machining results, and a plurality of in-machining behaviors including a plurality of simulation results and a plurality of measurement results, so that when one machining result is input, it outputs a corresponding in-machining behavior and corresponding machining conditions.

[0095] Fig. 10 is a flowchart showing the processing executed by the control device 1A of Fig. 7. The processing of Fig. 10 includes step S18A instead of step S18 of Fig. 6, and further includes step S31.

[0096] In step S31, while processing the workpieces 5a and 5b (step S16), the processor 11 uses the welding process monitor 6 to measure the behavior of the workpieces 5a and 5b in the process of processing the workpieces 5a and 5b, and obtains the measurement results.

[0097] In step S18A, the processor 11 retrains the machining model based on one machining condition determined in step S13 or S15, the machining behavior acquired in step S31, and the machining results acquired in step S17.

[0098] When measuring the behavior of the workpieces 5a, 5b during the machining process, the evaluation conditions in step S11 may include one or more parameters indicating the measurement results. In this case, the processor 11 may execute steps S14 and S15 after step S31. Specifically, the processor 11 controls the machining device 2 to machine the workpieces 5a, 5b under multiple machining conditions corresponding to desired machining results, and while machining the workpieces 5a, 5b, obtains multiple measurement results corresponding to the multiple machining conditions using the welding process monitor 6. Then, the processor 11 determines an appropriate machining condition from the multiple machining conditions, the corresponding measurement result, and the corresponding machining result based on the evaluation conditions including the measurement results.

[0099] According to the machining system 100A of this embodiment, by training the machining model based on the behavior during machining, including the measurement results in addition to the simulation results, the machining model can be trained to determine the machining conditions corresponding to the desired machining results with higher accuracy than in the first embodiment.

[0100] [Advantages of the second embodiment] According to the machining system 100A of the second embodiment, the communication device 16 may be further connected to a measurement device that measures the behavior of the workpieces 5a, 5b in the process of machining the workpieces 5a, 5b and acquires the measurement results, such as a welding process monitor 6. In this case, the behavior during machining further includes the measurement results.

[0101] With this configuration, the measurement results during the machining of the workpieces 5a and 5b can be acquired as the behavior of the workpieces 5a and 5b during machining.

[0102] [Other embodiments] As described above, the embodiments have been described as examples of the technology disclosed in this application. For this purpose, the accompanying drawings and detailed description have been provided. However, the technology in this disclosure is not limited to these, and can be applied to embodiments in which modifications, substitutions, additions, omissions, etc. are made as appropriate. Furthermore, the components described in the above embodiments can be combined to create new embodiments.

[0103] Therefore, the components shown in the accompanying drawings and detailed description may include not only components essential for solving the problem, but also components that are not essential for solving the problem in order to illustrate the above technology. Therefore, the fact that these non-essential components are shown in the accompanying drawings and / or detailed description should not be interpreted as immediately identifying these non-essential components as essential.

[0104] Furthermore, since the above-described embodiments are intended to illustrate the technology of the present disclosure, various modifications, substitutions, additions, omissions, etc. may be made within the scope of the claims or their equivalents.

[0105] The desired processing results are not limited to being obtained from the input device 14 such as a keyboard and / or a pointing device, but may be read from the storage device 13 or obtained from an external device via the communication device 16. In this case, the storage device 13 or the communication device 16 functions as an input device for obtaining the desired processing results of the workpieces 5a and 5b.

[0106] The behavior of the workpieces 5a and 5b during processing may include the temperature, humidity, surface roughness, etc. of the workpieces 5a and 5b in addition to or instead of the cross-sectional images of the welded portions of the workpieces 5a and 5b.

[0107] The machining device 2 is not limited to machining two workpieces 5a and 5b, but may machine one workpiece, or may machine three or more workpieces.

[0108] The processing device 2 is not limited to a laser welding device, and may include, for example, a furnace, a soldering device, a press device, and the like.

[0109] The cross-sectional shape of the workpiece 5c is not limited to being photographed by a camera, and depending on the material of the workpiece 5c, it may be obtained, for example, by an X-ray computed tomography device. In this case, the cross-sectional shape of the workpiece 5c can be obtained without cutting and polishing the cross-section of the workpiece 5c. Furthermore, if the position of the deepest part of the molten pool cannot be identified by cutting and polishing the cross-section, the cross-sectional shape of the workpiece 5c may be calculated by a welding simulation.

[0110] The cross-sectional shape measuring device 33 may be located outside the range transported by the transport device 4 (ie, outside the processing system).

[0111] In the described embodiment, the machine learning process (FIGS. 4 and 9) and the machining process (FIGS. 6 and 10) are performed by the same control device, but the machine learning process and the machining process may be performed by separate control devices. In this case, the first control device performs the machine learning process to train a machining model. The trained machining model is stored in a storage device of the second control device, and the second control device performs the machining process using this machining model.

[0112] [Summary of the embodiment] A machine learning device according to a first aspect of the present disclosure includes: a first communication device connected to a processing device that processes at least one workpiece under variable processing conditions and an inspection device that inspects the workpiece processed by the processing device; a first storage device that stores a simulation result indicating a simulated behavior of the workpiece during the process of machining the workpiece; a first processor; The first processor Controlling the machining device so as to machine at least one workpiece under each of a plurality of machining conditions; The inspection device inspects each of the workpieces machined under the plurality of machining conditions to obtain a plurality of machining results of the workpieces; A machining model is trained based on the plurality of machining conditions, the plurality of machining results, and a plurality of in-machining behaviors that indicate the behavior of the workpiece during the process of machining the workpiece, the plurality of in-machining behaviors including the simulation results, so that when one machining result is input, a corresponding in-machining behavior and corresponding machining conditions are output.

[0113] According to a machine learning device according to a second aspect of the present disclosure, in the machine learning device according to the first aspect, The processing conditions include a processing position, a processing power, and a processing profile.

[0114] According to a third aspect of the present disclosure, in the machine learning device according to the first or second aspect, the inspection device includes a first measuring device that three-dimensionally measures the surface shape of the machined workpiece, The machining result includes the surface shape of the machined workpiece.

[0115] According to a fourth aspect of the present disclosure, in the machine learning device according to one of the first to third aspects, The inspection device includes a second measuring device that measures the strength of the processed workpiece, The machining result includes the strength of the machined workpiece.

[0116] According to a fifth aspect of the present disclosure, in the machine learning device according to one of the first to fourth aspects, the inspection device includes a third measuring device that measures a cross-sectional shape of the processed workpiece, The machining result includes a cross-sectional shape of the machined workpiece.

[0117] According to a sixth aspect of the present disclosure, in the machine learning device according to one of the first to fifth aspects, the first communication device is further connected to a fourth measurement device that measures the behavior of the workpiece in the process of machining the workpiece and acquires measurement results; The in-process behavior further includes the measurement results.

[0118] According to a seventh aspect of the present disclosure, in the machine learning device according to one of the first to sixth aspects, the processing device is a laser welding device that welds the workpiece with a laser, The simulation results indicate the simulated behavior of the workpiece during the process of welding the workpiece.

[0119] A processing system according to an eighth aspect of the present disclosure includes: a machining device that machines at least one workpiece under variable machining conditions; A processing system including a control device that controls the processing device, The control device a second communication device connected to the processing device; a second storage device that stores a processing model trained by the machine learning device according to one of the first to seventh aspects; an input device for acquiring a desired processing result of the workpiece; a second processor; The second processor receiving a desired processing result of the workpiece from the input device; Using the machining model, determining machining conditions corresponding to the desired machining result; The machining device is controlled so as to machine the workpiece under the determined machining conditions.

[0120] According to the processing system according to the ninth aspect of the present disclosure, in the processing system according to the eighth aspect, When there are multiple machining conditions corresponding to the desired machining result, the second processor determines one machining condition corresponding to the desired machining result based on evaluation conditions including at least one parameter selected from one or more first parameters indicating the machining conditions, one or more second parameters indicating the behavior during machining, and one or more third parameters indicating the machining result.

[0121] A machine learning method according to a tenth aspect of the present disclosure includes: Using the processing device, processing at least one workpiece under a plurality of processing conditions; Using an inspection device, inspect each of the workpieces machined under the plurality of machining conditions by the machining device to obtain a plurality of machining results of the workpieces; obtaining simulation results indicative of simulated behavior of the workpiece during machining of the workpiece; The method includes training a machining model so that when one machining result is input, it outputs a corresponding machining behavior and corresponding machining conditions based on the plurality of machining conditions, the plurality of machining results, and a plurality of machining behaviors that indicate the behavior of the workpiece during the process of machining the workpiece, the plurality of machining behaviors including the simulation results.

[0122] A processing method according to an eleventh aspect of the present disclosure includes: Obtaining a desired machining result of the workpiece; determining machining conditions corresponding to the desired machining result using a machining model trained by the machine learning method according to the tenth aspect; and controlling a machining device so as to machine the workpiece under the determined machining conditions. [Industrial Applicability]

[0123] According to the machine learning method of the present disclosure, it is possible to train a machining model so as to derive appropriate machining conditions corresponding to a desired machining result with high accuracy in a short time. The machine learning method of the present disclosure is applicable to any machining that requires a long time to search for machining conditions. [Explanation of symbols]

[0124] 1,1A control device 2,2A processing equipment 3 Inspection equipment 4. Conveyor equipment 5a, 5b Work 5c Machined workpiece 5c1, 5c2 Cut workpiece 6 Welding process monitor 10 Bus 11 processors 12 Memory 13 Storage device 14 Input Devices 15 Display device 16. Communications equipment 21 Laser light source device 22,22A optical system 31 Surface profile measuring device 32 Strength measuring device 33 Cross-sectional shape measuring device 100,100A processing system

Claims

1. a first communication device connected to a processing device that processes at least one workpiece under variable processing conditions and an inspection device that inspects the workpiece processed by the processing device; a first storage device that stores a simulation result indicating a simulated behavior of the workpiece during the process of machining the workpiece; a first processor; The first processor Controlling the machining device so as to machine at least one workpiece under each of a plurality of machining conditions; The inspection device inspects each of the workpieces machined under the plurality of machining conditions to obtain a plurality of machining results of the workpieces; training a machining model based on the plurality of machining conditions, the plurality of machining results, and a plurality of in-machining behaviors indicating behaviors of the workpiece in the process of machining the workpiece, the plurality of in-machining behaviors including the simulation results, so as to output a corresponding in-machining behavior and corresponding machining conditions when one machining result is input; Machine learning device.

2. The processing conditions include a processing position, a processing power, and a processing profile. The machine learning device according to claim 1 .

3. the inspection device includes a first measuring device that three-dimensionally measures the surface shape of the machined workpiece, The machining result includes a surface shape of the machined workpiece. The machine learning device according to claim 1 .

4. the inspection device includes a second measuring device that measures the strength of the processed workpiece; The processing result includes the strength of the processed workpiece. The machine learning device according to claim 1 .

5. the inspection device includes a third measuring device that measures a cross-sectional shape of the machined workpiece, The machining result includes a cross-sectional shape of the machined workpiece. The machine learning device according to claim 1 .

6. the first communication device is further connected to a fourth measurement device that measures a behavior of the workpiece in a process of machining the workpiece and acquires a measurement result; The processing behavior further includes the measurement results. The machine learning device according to claim 1 .

7. the processing device is a laser welding device that welds the workpiece with a laser, the simulation results indicate a simulated behavior of the workpiece during the process of welding the workpiece; The machine learning device according to claim 1 .

8. a machining device that machines at least one workpiece under variable machining conditions; A processing system including a control device that controls the processing device, The control device a second communication device connected to the processing device; a second storage device that stores a processing model trained by the machine learning device according to any one of claims 1 to 7; an input device for acquiring a desired processing result of the workpiece; a second processor; The second processor receiving a desired processing result of the workpiece from the input device; Using the machining model, determining machining conditions corresponding to the desired machining result; controlling the machining device so as to machine the workpiece under the determined machining conditions; Processing system.

9. when there are a plurality of machining conditions corresponding to the desired machining result, the second processor determines one machining condition corresponding to the desired machining result based on evaluation conditions including at least one parameter selected from one or more first parameters indicating the machining condition, one or more second parameters indicating the behavior during machining, and one or more third parameters indicating the machining result; The processing system according to claim 8.

10. Using the processing device, processing at least one workpiece under a plurality of processing conditions; Using an inspection device, inspect each of the workpieces machined under the plurality of machining conditions by the machining device to obtain a plurality of machining results of the workpieces; obtaining simulation results indicative of simulated behavior of the workpiece during machining of the workpiece; training a machining model based on the plurality of machining conditions, the plurality of machining results, and a plurality of in-machining behaviors indicating behaviors of the workpiece in the process of machining the workpiece, the plurality of in-machining behaviors including the simulation results, so as to output a corresponding in-machining behavior and corresponding machining conditions when one machining result is input; Machine learning methods.

11. Obtaining a desired machining result of the workpiece; determining machining conditions corresponding to the desired machining result using a machining model trained by the machine learning method according to claim 10; controlling a machining device to machine the workpiece under the determined machining conditions; Processing method.

Images