Control device, machining system, control method, and program

Abstract

This control device comprises: an acquisition unit that acquires control information of a machine tool that forms a three-dimensional structure having at least one type of periodicity on the surface of a workpiece, and an analysis result generated by analyzing the workpiece obtained as a result of operating the machine tool on the basis of the control information; a machine learning unit that trains a machine learning model on the basis of a data set including the control information and the analysis result; a prediction unit that predicts the control information by inputting a desired analysis result to the machine learning model; and a control unit that operates the machine tool on the basis of the prediction result of the control information.

Description

Control device, processing system, control method, and program

[0001] This disclosure relates to a control device, a processing system, a control method, and a program.

[0002] For example, the microsurface structures that plants, insects, and animals have acquired through evolution to adapt to the Earth's environment can exhibit new physical functions such as hydrophilicity / hydrophobicity, slipperiness, stain resistance, anti-reflective properties, structural color, high / low thermal conductivity, low friction / flow resistance, and low noise. Such surface microstructures in the nano / micrometer range are called functional textures. The technology to artificially reproduce functional textures is called biomimetics. Biomimetics is expected to be used in measures to improve the added value of products.

[0003] Patent No. 6599069

[0004] However, it is difficult to artificially and perfectly reproduce three-dimensional microsurface structures. Conventional techniques rely solely on trial and error to determine how to machine functional textures, and this method has not been established as a design or machining technique. If it were possible to automatically determine the control information for a machine tool simply by specifying the desired physical function, this would be an extremely promising manufacturing technology.

[0005] One aspect of this disclosure provides a technology for precisely machining a workpiece.

[0006] A control device according to one aspect of the present disclosure includes: an acquisition unit that acquires control information for a machine tool that forms a three-dimensional structure having at least one type of periodicity on the surface of a workpiece, and analysis results obtained by analyzing a workpiece as a result of operating the machine tool based on the control information; a machine learning unit that trains a machine learning model based on a dataset including the control information and the analysis results; a prediction unit that predicts control information by inputting desired analysis results into the machine learning model; and a control unit that operates the machine tool based on the prediction results of the control information.

[0007] According to one aspect of this disclosure, a workpiece can be machined with high precision.

[0008] Figure 1 is a block diagram showing an example of the overall configuration of a processing system. Figure 2 is a schematic diagram showing an example of a laser processing apparatus. Figure 3 is a block diagram showing an example of a computer. Figure 4 is a block diagram showing an example of the functional configuration of a control device according to the first embodiment. Figure 5 is a diagram showing an example of a prediction model according to the first embodiment. Figure 6 is a flowchart showing an example of a processing method according to the first embodiment. Figure 7 is a block diagram showing an example of the functional configuration of a control device according to the second embodiment. Figure 8 is a flowchart showing an example of periodicity analysis processing according to the second embodiment. Figure 9 is a diagram showing an example of shape data. Figure 10 is a diagram showing a first example of periodicity data. Figure 11 is a diagram showing a second example of periodicity data. Figure 12 is a diagram showing an example of a laser processing algorithm. Figure 13 is a diagram showing an example of the cross-sectional shape of a periodic structure. Figure 14 is a diagram showing an example of the contact angle measurement result. Figure 15 is a diagram showing an example of prediction accuracy. Figure 16 is a diagram showing an example of the contribution of parameters.

[0009] Hereinafter, embodiments of this disclosure will be described with reference to the accompanying drawings. In this specification and drawings, components having substantially the same functional configuration are denoted by the same reference numerals, and redundant descriptions will be omitted.

[0010] [First Embodiment] The first embodiment of this disclosure is an example of a processing system for processing a workpiece. The processing system according to this embodiment is an example, and this disclosure is not limited thereto.

[0011] <Overall Configuration> The overall configuration of the processing system according to the first embodiment of this disclosure will be described with reference to Figure 1. Figure 1 is a block diagram showing an example of the overall configuration of the processing system.

[0012] As shown in Figure 1, the machining system 1000 includes a machine tool 10, an analysis device 20, and a control device 30. The control device 30 may be electrically connected to the machine tool 10 and the analysis device 20, respectively. The machine tool 10, the analysis device 20, and the control device 30 may be connected to each other via a communication network so that they can communicate data with one another. The communication network may include, for example, a LAN (Local Area Network), a VPN (Virtual Private Network), or the Internet.

[0013] The machine tool 10 is an example of equipment for processing a workpiece 9. The machine tool 10 may process the surface of the workpiece 9. The machine tool 10 may form a three-dimensional structure having at least one type of periodicity (hereinafter also referred to as the "periodic structure") on the surface of the workpiece 9. The periodic structure may also include non-periodicity. For example, the periodic structure may have periodicity in one predetermined axis and not periodicity in another axis (in other words, it may have non-periodicity). The machine tool 10 may also be, for example, a laser processing device capable of short-pulse laser processing (hereinafter also referred to as the "short-pulse laser processing device").

[0014] The analysis device 20 is an example of equipment for analyzing a workpiece 9. The analysis device 20 may analyze a workpiece 9 processed by a machine tool 10. The analysis device 20 may measure the shape of the periodic structure formed on the surface of the workpiece 9. The analysis device 20 may measure physical quantities that indicate the physical properties of the surface of the workpiece 9. The analysis device 20 may output the analysis results of the workpiece 9. The analysis device 20 may be, for example, a laser microscope or a fully automatic contact angle meter.

[0015] The control device 30 is an example of an information processing device that controls the machine tool 10. The control device 30 may be a computer such as a personal computer, workstation, server, smartphone, or tablet terminal. The control device 30 may be built into the machine tool 10. The control device 30 may be installed in a remote location different from where the machine tool 10 is installed.

[0016] The control device 30 may accept input of machining conditions from the user of the machining system 1000. The control device 30 may determine control information for the machine tool 10 based on the machining conditions. The control device 30 may control the operation of the machine tool 10 based on the control information for the machine tool 10.

[0017] The control device 30 may predict control information for the machine tool 10 based on machine learning. The control device 30 may have a trained prediction model. The prediction model is a machine learning model that predicts control information for the machine tool 10 based on machining conditions including the analysis results to be predicted. The analysis results to be predicted may be analysis results desired by the user of the machining system 1000. The control device 30 may train the prediction model based on a dataset including control information for the machine tool 10 and analysis results for the workpiece 9.

[0018] The control device 30 may repeatedly acquire control information of the machine tool 10 and analysis results of the workpiece 9, and train the predictive model. Each time the workpiece 9 is processed, the control device 30 may accumulate control information of the machine tool 10 and analysis results of the workpiece 9, and when a predetermined amount of data has been accumulated, it may update the predictive model. Updating the predictive model may include, for example, additional training of the predictive model or reconstruction of the predictive model.

[0019] <Laser Processing Device> The machine tool 10 is realized, for example, by a laser processing device. Figure 2 is a schematic diagram showing an example of a laser processing device.

[0020] As shown in Figure 2, the laser processing apparatus 50 may mainly include a laser light generation unit 51, a scanning unit 52, and a focusing lens 53. The laser light source in the laser light generation unit 51 is not particularly limited and can be appropriately selected according to the purpose. Specific examples of laser light sources include excimer lasers, Nd:YAG lasers, and Nd:YVO lasers. ４ Examples include lasers and semiconductor lasers. Furthermore, the oscillation wavelength of the laser light can be between 100 nm and 1100 nm.

[0021] The laser beam LB generated by the laser beam generation unit 51 can be scanned or swept over the surface of the workpiece 9 using the scanning unit 52. The scanning unit 52 may be, for example, a galvanoscanner using a galvanoscanner mirror, a polygon scanner using a polygon mirror, etc. The scanning unit 52 can also scan the laser beam LB along one direction (for example, the x-direction or y-direction on the surface) along the surface of the workpiece 9, more specifically, so that spaced-apart linear grooves are formed. Alternatively, the scanning unit 52 can also scan the laser beam LB along two mutually orthogonal directions (for example, the x-direction and y-direction on the surface), more specifically, so that spaced-apart linear grooves along one direction intersect with spaced-apart grooves perpendicular to that direction.

[0022] By scanning the workpiece 9 with a short-pulse laser beam LB, recesses (grooves) can be formed on the surface of the workpiece 9. The scanning of the laser beam LB required to obtain the recesses may be done once or multiple times.

[0023] It is preferable to use a focusing lens 53 with a focal length of f10 or more and f2000 or less, and with specifications such as telecentricity. Although not shown in the illustration, optimal irradiation conditions can be obtained by using additional components as appropriate, such as adjusting the spot diameter of the laser beam LB using a beam expander.

[0024] The irradiation conditions of the laser beam LB can be adjusted so that the ratio of the depth d of the recess to the spacing τ of the uneven structure (d / τ) is at least 0.5. The specific conditions may vary depending on the intended use of the workpiece 9, the appropriate shape of the uneven structure for that use, the material of the workpiece 9, etc. As an example, the irradiation conditions of the laser beam LB may be a pulse width of 100 fs to 500 ns, a repetition frequency of 1 kHz to 10 MHz, and a fluence of 10 mJ / mm². ２ 1J / mm or more ２ The following is possible:

[0025] Furthermore, when irradiating with laser beam (LB), masks or other means may be used as appropriate to avoid irradiating areas where the laser beam is not desired.

[0026] <Computer> The control device 30 is implemented by a computer, for example. Figure 3 is a block diagram showing an example of the computer's hardware configuration.

[0027] As shown in Figure 3, the computer 500 includes a CPU (Central Processing Unit) 501, ROM (Read Only Memory) 502, RAM (Random Access Memory) 503, HDD (Hard Disk Drive) 504, input device 505, display device 506, communication interface 507, and external interface 508. The CPU 501, ROM 502, and RAM 503 form what is known as a computer. Each piece of hardware in the computer 500 is interconnected via a bus line 509. The input device 505 and display device 506 may also be used by connecting them to the external interface 508.

[0028] The CPU 501 is a computing device that reads programs and data from a storage device such as ROM 502 or HDD 504 onto RAM 503 and executes processing to realize the overall control and functions of the computer 500. The computer 500 may have a GPU (Graphics Processing Unit) in addition to or instead of the CPU 501.

[0029] ROM 502 is an example of a non-volatile semiconductor memory (storage device) that can retain programs and data even when the power is turned off. ROM 502 functions as the main memory, storing various programs and data necessary for the CPU 501 to execute the various programs installed on HDD 504. Specifically, ROM 502 stores boot programs such as BIOS (Basic Input Output System) and EFI (Extensible Firmware Interface) that are executed when the computer 500 starts up, as well as data such as OS (Operating System) settings and network settings.

[0030] RAM 503 is an example of volatile semiconductor memory (storage device) whose programs and data are erased when the power is turned off. RAM 503 can be, for example, DRAM (Dynamic Random Access Memory) or SRAM (Static Random Access Memory). RAM 503 provides a workspace that is expanded when various programs installed on HDD 504 are executed by CPU 501.

[0031] HDD 504 is an example of a non-volatile storage device that stores programs and data. The programs and data stored in HDD 504 include the OS, which is the basic software that controls the entire computer 500, and applications that provide various functions on the OS. Note that the computer 500 may use a storage device that uses flash memory as its storage medium (for example, an SSD: Solid State Drive) instead of HDD 504.

[0032] The input device 505 includes a touch panel used by the user to input various signals, operation keys and buttons, a keyboard and mouse, and a microphone for inputting sound data such as voice.

[0033] The display device 506 consists of a display such as a liquid crystal or organic EL (Electro-Luminescence) that displays a screen, and a speaker that outputs sound data such as audio.

[0034] Communication I / F 507 is an interface that connects to a communication network and allows the computer 500 to perform data communication.

[0035] External I / F 508 is an interface to external devices. External devices include drive devices 510, etc.

[0036] The drive device 510 is a device for setting the recording medium 511. The recording medium 511 here includes media that record information optically, electrically, or magnetically, such as CD-ROMs, flexible disks, magneto-optical disks, etc. The recording medium 511 may also include semiconductor memories that record information electrically, such as ROMs, flash memories, etc. Thereby, the computer 500 can read and / or write to the recording medium 511 via the external I / F 508.

[0037] In addition, various programs installed in the HDD 504 are installed, for example, when the distributed recording medium 511 is set in the drive device 510 connected to the external I / F 508 and the various programs recorded on the recording medium 511 are read by the drive device 510. Alternatively, various programs installed in the HDD 504 may be installed by being downloaded from a communication network via the communication I / F 507.

[0038] <Function Configuration> The functional configuration of the control device 30 will be described while referring to FIG. 4. FIG. 4 is a block diagram showing an example of the functional configuration of the control device.

[0039] As shown in FIG. 4, the control device 30 includes a data storage unit 301, a model storage unit 302, an input unit 310, a prediction unit 320, a control unit 330, an acquisition unit 340, and a machine learning unit 350. The control device 30 functions as the data storage unit 301, the model storage unit 302, the input unit 310, the prediction unit 320, the control unit 330, the acquisition unit 340, and the machine learning unit 350 when a pre-installed control program is executed.

[0040] For example, the data storage unit 301 and the model storage unit 302 are realized by the RAM 503 or the HDD 504 shown in FIG. 3. For example, the input unit 310, the prediction unit 320, the control unit 330, the acquisition unit 340, and the machine learning unit 350 are realized by the processing in which the program expanded from the HDD 504 shown in FIG. 3 onto the RAM 503 causes the CPU 501 to execute.

[0041] The data storage unit 301 stores a dataset used for training the predictive model. The dataset may include material information of the workpiece 9, control information of the machine tool 10, and analysis results of the workpiece 9. The analysis results of the workpiece 9 may include at least one of shape data or physical property data.

[0042] The material information of the workpiece 9 is information about the substance that makes up the workpiece 9. The material information of the workpiece 9 may also include the material and form of the workpiece 9. The material of the workpiece 9 may include, for example, metals such as steel, aluminum, and copper; inorganic materials such as glass and ceramics; and organic materials such as polyethylene, polypropylene, and fluororesin. The form of the workpiece 9 may include, for example, a plate, a cylinder, a film, a sphere, etc.

[0043] The control information for the machine tool 10 is information used to control the operation of the machine tool 10. The control information for the machine tool 10 may also include adjustment parameters for the machine tool 10. If the machine tool 10 is a laser processing device, the adjustment parameters may include, for example, at least one of the laser irradiation conditions or the three-dimensional shape. The laser irradiation conditions may include, for example, the spot diameter, fluence (energy per unit area), number of shots, laser profile (Gaussian shape or hat-top shape, etc.). The three-dimensional shape may include, for example, a pillar, waffle, taper, reverse taper, etc.

[0044] The shape data is data obtained by analyzing the shape of the periodic structure formed on the surface of the workpiece 9. The shape data may also be data representing the three-dimensional structure of the surface of the workpiece 9 in x, y, and z coordinates. The shape data may also be data obtained by measuring the surface of the workpiece 9 with a laser microscope.

[0045] The physical properties data is data obtained by analyzing the physical properties of the surface of the workpiece 9. Examples of physical properties may include at least one of the following: hydrophilicity, hydrophilicity, slipperiness, stain resistance, anti-reflectivity, structural color, high thermal conductivity, low thermal conductivity, low friction, flow resistance, or low noise. For example, physical properties data related to hydrophilicity may include data obtained by measuring at least one of the equilibrium contact angle, apparent contact angle, hysteresis angle, or slip angle.

[0046] The model memory unit 302 stores the prediction model. The prediction model is a machine learning model that takes machining conditions including the analysis results of the target to be predicted as input and outputs control information to operate the machine tool 10 so that the said analysis results can be obtained. The prediction model may be a machine learning model based on deep learning. The prediction model may be a machine learning model based on an ensemble method. The prediction model may include, as an example, a neural network, a deep neural network, a decision tree, a random forest, or a gradient boosting tree.

[0047] The input unit 310 receives input of processing conditions for the workpiece 9. The input unit 310 may also receive processing conditions entered by a user of the processing system 1000. The processing conditions may include material information of the workpiece 9 and analysis results desired by the user.

[0048] The prediction unit 320 predicts the control information of the machine tool 10. The prediction unit 320 may predict the control information of the machine tool 10 based on a prediction model read from the model storage unit 302. The prediction unit 320 may also predict the control information of the machine tool 10 by inputting the machining conditions received by the input unit 310 into the prediction model.

[0049] The control unit 330 controls the operation of the machine tool 10. The control unit 330 may operate the machine tool 10 based on the prediction result of the control information for the machine tool 10 predicted by the prediction unit 320. The control unit 330 may transmit a control signal to the machine tool 10 instructing it to operate based on the adjustment parameters included in the prediction result. The control unit 330 may set the adjustment parameters included in the prediction result in the machine tool 10 and control it so that it operates based on the adjustment parameters when the machine tool 10 operates thereafter. The control unit 330 may also indirectly control the machine tool 10 by presenting the control information of the machine tool 10 to the user of the machining system 1000 and having the user operate the machine tool 10 according to the control information.

[0050] The acquisition unit 340 acquires control information of the machine tool 10 and analysis results of the workpiece 9. The acquisition unit 340 may also acquire control information input by the user of the machining system 1000. The acquisition unit 340 may also acquire control information of the machine tool 10 predicted by the prediction unit 320. The acquisition unit 340 may also acquire control information used by the control unit 330 to control the machine tool 10. The acquisition unit 340 may also acquire analysis results (specifically, shape data or physical property data) output by the analysis device 20.

[0051] The acquisition unit 340 stores the control information of the machine tool 10 and the analysis results of the workpiece 9 in the data storage unit 301. The acquisition unit 340 may also associate the control information of the machine tool 10 with the analysis results obtained by analyzing the workpiece 9 as a result of operating the machine tool 10 based on the control information, and add them to the dataset stored in the data storage unit 301.

[0052] The machine learning unit 350 trains a predictive model. The machine learning unit 350 may train the predictive model based on a dataset read from the data storage unit 301. The machine learning unit 350 may update the trained predictive model based on a dataset read from the data storage unit 301. The machine learning unit 350 may repeatedly update the trained predictive model.

[0053] Figure 5 shows an example of a prediction model. As shown in Figure 5, the prediction model is a machine learning model that takes the machining conditions of the workpiece 9 as input information and the control information of the machine tool 10 as output information. Figure 5 shows an example in which the prediction model is a deep learning model based on the ensemble method.

[0054] The processing conditions may include material information of the workpiece 9 and the analysis results of the workpiece 9. The material information may include the material and form of the workpiece 9. The analysis results of the workpiece 9 may include at least one of shape data or physical property data. The control information of the machine tool 10 may include at least one of the laser irradiation conditions or the three-dimensional shape.

[0055] <Processing Procedure> The processing method of the workpiece performed by the processing system 1000 will be explained with reference to Figure 6. Figure 6 is a flowchart showing an example of a processing method.

[0056] In step S1, the input unit 310 of the control device 30 receives input of processing conditions for the workpiece 9. The processing conditions for the workpiece 9 include analysis results desired by the user of the processing system 1000. The analysis results only need to include at least one of shape data or physical property data. The input unit 310 sends the received processing conditions to the prediction unit 320.

[0057] In step S2, the prediction unit 320 of the control device 30 receives machining conditions from the input unit 310. The prediction unit 320 reads a prediction model from the model storage unit 302. The prediction unit 320 inputs the machining conditions into the prediction model. Based on the input machining conditions, the prediction model predicts the control information of the machine tool 10 and outputs the prediction result. The prediction unit 320 obtains the control information of the machine tool 10 from the prediction result output by the prediction model. The prediction unit 320 sends the control information of the machine tool 10 to the control unit 330.

[0058] In step S3, the control unit 330 of the control device 30 receives control information for the machine tool 10 from the prediction unit 320. Based on the control information for the machine tool 10, the control unit 330 performs control to operate the machine tool 10. The machine tool 10 operates based on the control information. As a result, a periodic structure is formed on the surface of the workpiece 9.

[0059] In step S4, the analysis device 20 analyzes the workpiece 9 processed in step S3. The analysis device 20 may measure the periodic structure formed on the surface of the workpiece 9. The analysis device 20 may also measure physical quantities that indicate the physical properties of the surface of the workpiece 9. The analysis device 20 outputs the analysis results (shape data or physical property data) of the workpiece 9 to the control device 30.

[0060] In step S5, the acquisition unit 340 of the control device 30 acquires control information of the machine tool 10 from the control unit 330. The acquisition unit 340 acquires the analysis results of the workpiece 9 from the analysis device 20. The acquisition unit 340 associates the control information of the machine tool 10 with the analysis results of the workpiece 9 and adds them to the data set stored in the data storage unit 301.

[0061] In step S6, the machine learning unit 350 of the control device 30 reads a dataset from the data storage unit 301. The machine learning unit 350 may read only datasets that have been added since the last time step S6 was executed. The machine learning unit 350 reads a prediction model from the model storage unit 302. The machine learning unit 350 updates the prediction model based on the read dataset. The machine learning unit 350 stores the updated prediction model in the model storage unit 302.

[0062] Step S6 may be executed at a predetermined timing. For example, step S6 may be executed when a predetermined number of data points have been added to the dataset since the last execution of step S6 (i.e., since the prediction model was updated). Alternatively, for example, step S6 may be executed when a user of the processing system 1000 performs an operation to instruct an update.

[0063] The processing system 1000 repeatedly executes the processes from step S1 to step S6. Predictive models tend to improve in prediction accuracy as the amount of data they are trained on increases. Therefore, the processing system 1000 can accumulate more and more datasets the longer it continues to operate, and as a result, it can build a highly accurate predictive model.

[0064] [Second Embodiment] In the first embodiment, the analysis results of the workpiece 9 were either shape data, which measured the shape of the periodic structure formed on the surface of the workpiece 9, or physical property data, which measured physical quantities that indicate the physical properties of the surface of the workpiece 9. In the second embodiment, the analysis results of the workpiece 9 are used, which are data obtained by analyzing the periodicity of the periodic structure formed on the surface of the workpiece 9.

[0065] In machine learning, the quality and quantity of data significantly impact the estimation accuracy of the model. This embodiment proposes a method for improving data quality.

[0066] <Functional Configuration> The functional configuration of the control device 30 will be explained with reference to Figure 7. Figure 7 is a block diagram showing an example of the functional configuration of the control device according to the second embodiment.

[0067] As shown in Figure 7, the control device 30 comprises a data storage unit 301, a model storage unit 302, an input unit 310, a prediction unit 320, a control unit 330, an acquisition unit 340, a machine learning unit 350, and an analysis unit 360. In other words, the control device 30 according to this embodiment differs from the first embodiment in that it further comprises an analysis unit 360.

[0068] The analysis unit 360 analyzes the workpiece 9. The analysis unit 360 may analyze the workpiece 9 processed by the machine tool 10. The analysis unit 360 may analyze the periodic structure formed on the surface of the workpiece 9. The analysis unit 360 may numerically analyze the periodicity of the periodic structure formed on the surface of the workpiece 9. The analysis unit 360 may numerically analyze the periodicity of the periodic structure formed on the surface of the workpiece 9 based on the shape data analyzed by the analysis device 20.

[0069] The data storage unit 301 stores a dataset used for training the prediction model. The dataset may include material information of the workpiece 9, control information of the machine tool 10, and analysis results of the workpiece 9. In this embodiment, the analysis results of the workpiece 9 may include periodic data.

[0070] The periodicity data is data showing the results of numerical analysis of the periodicity of the periodic structure formed on the surface of the workpiece 9. The numerical analysis may include, as an example, at least one of the following: Fourier analysis, wave number space method, autocorrelation function, group theory, Penrose tile analysis, quasi-periodic structure analysis, quasicrystal analysis, topological methods, or pattern analysis. The periodicity data may also be generated by searching for periodicity from the shape data of the workpiece 9. The periodicity data may also be data obtained by the analysis unit 360 analyzing the shape data.

[0071] Fourier analysis is a technique for transforming periodic signals and geometric structures into the frequency domain. Periodic patterns appear as characteristic peaks in the spectrum after the Fourier transform, allowing for quantitative analysis of periodicity. Fourier analysis is best suited for identifying the geometric periodicity of shapes. For example, Fourier analysis can be applied to periodic pattern analysis of images and surfaces, image processing, and analysis of the crystal structure of materials.

[0072] The wave number space method (or k-space method) is a technique that describes space as a collection of waves, and it is a method that transforms the wave equation in physical space into wave number space (Fourier space). In systems with a spatial periodic structure, analysis in wave number space is effective in clearly capturing the periodic structure. For example, the wave number space method is used in solid-state physics and quantum mechanics, and can be applied to the analysis of band structure. It is also sometimes used to create pulse sequence charts, which are blueprints for MRI imaging techniques.

[0073] The autocorrelation function is sometimes used to calculate time averages. It is used to evaluate how much self-similarity a function maintains over time and space. If a function has a periodic structure, the autocorrelation function will show peaks with corresponding periods. For example, the autocorrelation function can be applied to detecting periodic structures and repeating patterns in crystals.

[0074] Group theory is a mathematical method for analyzing symmetrical structures. Since many structures, such as crystals and hydrogen atoms, can be represented by point groups, group theory is important in crystallography and physics. For example, group theory is used to classify the symmetries of crystals and molecules, and is well-suited for analyzing the symmetries of periodic structures.

[0075] Quasiperiodic structures and quasicrystals are structures that possess spatial order despite not being periodic. Penrose tiling is a typical example. Methods for mathematically analyzing quasiperiodic structures and quasicrystals are being studied. This includes theories of aperiodic tiling and quasicrystals. A tiling is aperiodic if it does not have periodic regions of arbitrary size. Penrose tilings do not have translational symmetry, but they can have mirror symmetry and quintuple rotational symmetry.

[0076] Topological methods are techniques for capturing the topological features of spatial periodic structures. Homology and homotopy theory are used in topological methods to capture the topological features of spatial periodic structures. This allows for the analysis of properties related to the continuity and deformation of periodic structures.

[0077] Pattern analysis (or template matching) is a technique used in the fields of image processing and data analysis. Pattern analysis is useful for searching for specific patterns or structures within a dataset. For example, pattern analysis can be applied to set a pattern with a spatial periodic structure as a template and detect regions that match it.

[0078] <Processing Procedure> The periodicity analysis process performed by the analysis unit 360 will be explained with reference to Figure 8. Figure 8 is a flowchart showing an example of the periodicity analysis process.

[0079] In step S11, the analysis unit 360 acquires three-dimensional shape data from the dataset stored in the data storage unit 301. The three-dimensional shape data included in the dataset is generated by the analysis device 20. In this case, the analysis device 20 may be, for example, a laser microscope.

[0080] Figure 8 shows an example of shape data. Figure 8 shows the height distribution in the z direction in a two-dimensional region in the x-y direction of the three-dimensional shape data in a Cartesian coordinate system. In this embodiment, it is assumed that a periodic structure is formed on the surface of the workpiece 9, with recesses arranged at 120 μm intervals in the x direction and convex portions arranged at approximately 60 μm intervals in the y direction.

[0081] In step S12, the analysis unit 360 performs a two-dimensional Fourier transform on the three-dimensional shape data acquired in step S11 in the x, y, and z directions. In step S13, the analysis unit 360 acquires an amplitude spectrum based on the results of the Fourier transform on the three-dimensional shape data. In this way, the analysis unit 360 converts the three-dimensional shape data in a Cartesian coordinate system into wavelength data based on mathematical methods.

[0082] Figure 9 shows the first example of periodic data. Figure 9 shows the spectrum distribution generated by performing a Fourier transform on the x-coordinate data of the three-dimensional shape data in a Cartesian coordinate system.

[0083] Figure 10 shows a second example of periodic data. Figure 10 shows the wavelength spectral distribution obtained by taking the reciprocal of the spectral distribution shown in Figure 9. The wavelength spectral distribution shows the various periodic data that the surface of the workpiece 9 has in the x direction.

[0084] Referring to Figure 10, it can be seen that the wavelength spectral distribution has a short wavelength region of less than 1 μm, a medium wavelength region of 1 μm to less than 150 μm, and a long wavelength region of 150 μm or more. The short wavelength region corresponds to the surface roughness component that the base material had before processing. The long wavelength region corresponds to the waviness component that the base material had before processing. The medium wavelength region corresponds to the original unevenness processing component formed by the machine tool 10.

[0085] The advantage of exploring and representing periodicity such as wavelength spectra from 3D shape data is that only the original surface texture components formed by laser processing can be extracted and used in machine learning. For example, with the Fourier transform, the data can be divided into low-wavelength, medium-wavelength, and long-wavelength regions. By excluding the data in the low-wavelength and long-wavelength regions, only the noise-free medium-wavelength region (i.e., the original surface texture components formed by laser processing) can be extracted.

[0086] As described above, the machining system 1000 can automatically determine control information for the machine tool 10 that can obtain the desired periodic structure or physical characteristics based on a machine learning model. Conventionally, the control information for the machine tool 10 was determined through trial and error using a cut-and-try method. According to this embodiment, it is possible to efficiently machine or produce workpieces that can obtain the desired periodic structure or physical characteristics.

[0087] Furthermore, since data generated from the same source lacks independence from the original data, multicollinearity can occur, potentially leading to a decrease in prediction accuracy. When training a prediction model, it may be possible to use physical property data or periodicity data instead of shape data for training. This can help avoid multicollinearity and suppress a decrease in prediction accuracy.

[0088] For example, in methods that statistically determine the relationship between inputs and outputs, such as regression analysis, if there are combinations of variables with high correlation coefficients among the input variables (also called independent variables or explanatory variables), the estimation accuracy of the regression equation may become extremely poor, or the regression coefficients and odds ratios may take abnormal values. This phenomenon, where the analysis results become unstable and consequently reduce the accuracy of predictions, is called multicolinearity.

[0089] This can be caused by strongly correlated input variables pulling the output variable in opposite directions. If outliers exist and there are combinations of input variables with strong correlations, one of the input variables must be excluded and the analysis performed again. Choosing an algorithm that is less susceptible to the effects of multicollinearity is also effective. For example, using a gradient boosting tree, which is less affected by the correlation between input variables, is also effective.

[0090] [Third Embodiment] In liquid-repellent texture processing using ultrashort pulse lasers, which artificially mimics the uneven surface shape of micro- or nano-regions on biological surfaces to impart liquid-repellent properties, the Cassie-Baxter theory has been established to describe liquid-repellent properties. However, due to multiple nonlinearities in the laser's energy absorption, heat conduction, ablation characteristics, etc., it is not possible to design a surface shape that reproduces the desired function. In this embodiment, an estimation model is constructed using machine learning to express the relationship between processing conditions, including surface shape, and liquid-repellent properties.

[0091] A polyethylene terephthalate (PET) plate (specifically, PET-6820 manufactured by Takiron Co., Ltd.) with a thickness of 4 mm was used as the base material to be processed. PET is a general-purpose polymer material, and in laser processing, the nonlinearity of the resin's thermal conductivity is a factor that hinders the realization of quantitative design.

[0092] In this embodiment, three adjustment parameters related to laser processing were used. The first parameter is the energy per unit area (mJ / mm²). ２ The first parameter is the fluence. The second parameter is the overlap ratio (OR), which indicates the overlap rate of the spots due to laser oscillation. The third parameter is the tooth width f of the periodic structure that is formed. １ and groove width f ２ The ratio of groove width to tooth width (f ２ / f １ (Ratio). Table 1 shows the setting ranges for these parameters and other laser processing conditions.

[0093]

[0094] Based on different combinations of the above parameters, a total of 250 processing conditions were set. FIG. 12 is a diagram showing an example of a laser processing algorithm. In this embodiment, a pillar structure was fabricated using the algorithm shown in FIG. 12. FIG. 13 is a diagram showing an example of the cross-sectional shape of a periodic structure. The geometric parameters of the cross-sectional shape are the tooth width f １ and the groove width f ２ (that is, the pitch τ = f １ + f ２ ), and the depth d.

[0095] The plates processed under each condition were stored for 3 days in an incubator maintained at 50°C for aging, and then five apparent contact angle measurements were performed. As a result, (fluence, 1.3 - 10.4 mJ / mm ２ : 10 conditions) × (OR, 83 - 98%: 5 conditions) × (f ２ / f １ ratio, 0.3 - 2.5: 5 conditions) × (apparent contact angle: n = 5), a total of 1,250 data were obtained.

[0096] FIG. 14 is a diagram showing an example of the measurement results of the contact angle. In FIG. 14, the measurement results of the contact angle for each parameter are shown. The obtained data were used for learning and evaluation with the teacher data: verification data = 9:1.

[0097] A predictive model was created using laser processing parameters as explanatory variables and the apparent contact angle as the dependent variable. Two machine learning algorithms were used: LightGBM (Light Gradient Boosting Machine), a supervised learning algorithm, and a neural network (NN). The predictive accuracy of the machine learning models constructed with each algorithm was compared. Hyperparameters were adjusted in each algorithm to find the optimal model. In LightGBM, hyperparameters such as the number and depth of leaves in the decision tree were adjusted. In NN, hyperparameters such as the number of layers in the hidden layer, the number of neurons, and the learning rate were adjusted. In this way, more complex nonlinear relationships were modeled.

[0098] Figure 15 shows an example of prediction accuracy. As shown in Figure 15, the coefficient of determination (R) of the prediction accuracy of the contact angle relative to PET is shown. ２ The result was 0.94. Furthermore, the root mean squared error (RMSE) for the apparent contact angle was around 3.7°. High-precision results were obtained that fell within the range of measurement error. Although the difference is slight, the prediction model trained with NN had less error compared to LightGBM, and therefore, for the data set used in this study, it resulted in a highly accurate prediction model.

[0099] Figure 16 shows an example of the contribution of parameters. In Figure 16, SHAP (SHapley Additive exPlanations) was used to analyze the contribution of each parameter to the prediction of the machine learning model. SHAP allows for quantitative evaluation of the influence of each input parameter on the prediction. As shown in Figure 16, when the influence of fluence, overlap rate, and groove width / tooth width ratio on liquid repellency was visualized, it was identified that the influence of the parameters was in the order of fluence > overlap rate > groove width / tooth width ratio. In other words, it became clear that fluence had the greatest influence in the data set used in this study. It was also confirmed that the influence of overlap rate and groove width / tooth width ratio was relatively small.

[0100] The evaluation of the contributions showed that optimizing fluence is particularly important for improving the liquid repellency of PET. According to the Cassie-Baxter theory of liquid repellency, the apparent contact angle is improved by air entering the grooves of the periodic structure. In short-pulse laser processing, ablation is the dominant removal process, and deeper grooves are formed as the fluence increases. The prediction results in this study also confirmed the tendency for deeper grooves to be formed and the apparent contact angle to increase with higher fluence, which was qualitatively and theoretically consistent.

[0101] The above demonstrates that a model can be constructed to predict liquid repellency from laser processing conditions using machine learning. Effective input variables for the machine learning model include pulse width, wavelength, processing frequency, laser beam profile, spot diameter, workpiece material, workpiece shape, and surface roughness.

[0102] <Effects of the Embodiment> The control device 30 according to one embodiment of the present disclosure trains a machine learning model based on a dataset including control information of a machine tool that forms a three-dimensional structure having at least one type of periodicity on the surface of a workpiece, and analysis results obtained by analyzing the workpiece as a result of operating the machine tool based on the control information, and predicts the control information by inputting the desired analysis results into the machine learning model, and operates the machine tool based on the predicted results of the control information. In one aspect, according to this embodiment, the workpiece can be processed with high precision.

[0103] [Supplement] Each function of the embodiments described above can be realized by one or more processing circuits. Hereinafter, "processing circuit" in this specification includes processors programmed to execute each function by software, such as processors implemented by electronic circuits, as well as devices such as ASICs (Application Specific Integrated Circuits), DSPs (Digital Signal Processors), FPGAs (Field Programmable Gate Arrays), and conventional circuit modules designed to execute each function described above.

[0104] Although embodiments of the present invention have been described in detail above, the present invention is not limited to these embodiments, and various modifications or changes are possible within the scope of the gist of the present invention as described in the claims.

[0105] This application claims priority to Japanese Patent Application No. 2024-182429, filed with the Japan Patent Office on 18 October 2024, which is incorporated herein by reference to its entire contents.

[0106] 9: Workpiece 10: Machine tool 20: Analysis device 30: Control device 50: Laser processing device 51: Laser light generation unit 52: Scanning unit 53: Focusing lens 301: Data storage unit 302: Model storage unit 310: Input unit 320: Prediction unit 330: Control unit 340: Acquisition unit 350: Machine learning unit 360: Analysis unit 1000: Processing system

Claims

1. A control device comprising: an acquisition unit configured to acquire control information of a machine tool that forms a three-dimensional structure having at least one type of periodicity on the surface of a workpiece, and analysis results obtained by analyzing the workpiece as a result of operating the machine tool based on the control information; a machine learning unit configured to train a machine learning model based on a dataset including the control information and the analysis results; a prediction unit configured to predict the control information by inputting a desired analysis result into the machine learning model; and a control unit configured to operate the machine tool based on the prediction result of the control information.

2. The control device according to claim 1, wherein the analysis results include data obtained by numerical analysis of the periodicity of the three-dimensional structure formed on the surface of the workpiece.

3. The control device according to claim 2, wherein the numerical analysis includes at least one of the following: Fourier transform, wave number space method, autocorrelation function, group theory, Penrose tile analysis, quasi-periodic structure analysis, quasicrystal analysis, topological methods, or pattern analysis.

4. The control device according to claim 1, wherein the analysis results include data analyzing the physical properties of the surface of the workpiece.

5. The control device according to claim 4, wherein the physical properties include at least one of: hydrophilicity, repellency, slipperiness, stain resistance, non-reflectiveness, structural color, high thermal conductivity, low thermal conductivity, low friction, flow resistance, or low noise.

6. A control device according to any one of claims 1 to 5, which repeatedly performs the actions of acquiring the control information and the analysis results and updating the machine learning model.

7. The control device according to any one of claims 1 to 5, wherein the machine tool is a short-pulse laser processing device, and the control information includes at least one of spot diameter, fluence, number of shots, or laser profile.

8. A machining system comprising: a machine tool for forming a three-dimensional structure having at least one type of periodicity on the surface of a workpiece; an analysis device for analyzing the workpiece; and a control device for controlling the machine tool, wherein the control device comprises: an acquisition unit configured to acquire control information of the machine tool and analysis results obtained by analyzing the workpiece as a result of operating the machine tool based on the control information; a machine learning unit configured to train a machine learning model based on a dataset including the control information and the analysis results; a prediction unit configured to predict the control information by inputting a desired analysis result into the machine learning model; and a control unit configured to operate the machine tool based on the prediction result of the control information.

9. A control method comprising: a procedure for a computer to obtain control information for a machine tool that forms a three-dimensional structure having at least one type of periodicity on the surface of a workpiece, and an analysis result obtained by analyzing the workpiece as a result of operating the machine tool based on the control information; a procedure for training a machine learning model based on a dataset including the control information and the analysis result; a procedure for predicting the control information by inputting a desired analysis result into the machine learning model; and a procedure for operating the machine tool based on the predicted result of the control information.

10. A program for causing a computer to execute the following steps: a procedure for obtaining control information of a machine tool that forms a three-dimensional structure having at least one type of periodicity on the surface of a workpiece, and an analysis result obtained by analyzing the workpiece as a result of operating the machine tool based on the control information; a procedure for training a machine learning model based on a dataset including the control information and the analysis result; a procedure for predicting the control information by inputting a desired analysis result into the machine learning model; and a procedure for operating the machine tool based on the predicted result of the control information.

Images