Laser printing defect adjusting method and system

Abstract

The invention discloses a laser printing defect adjusting method and system, and relates to the technical field of laser manufacturing, the laser printing defect adjusting method is applied to the laser printing defect adjusting system, and the system comprises an ultrafast laser micro-shaping device and an ultrasonic-assisted micro-shaping device. And establishing a physical entity model to obtain a mapping relationship between the microscopic data and the macroscopic data. Applying the mapping relation to the initial twinborn model to obtain a target twinborn model, and performing model training on the target twinborn model based on the historical defect data and the current defect data to obtain a prediction model; and inputting the physical data of the to-be-tested workpiece into the prediction model to obtain an output result of the prediction model. And on the basis of the output result, micro-shaping is conducted on the workpiece to be detected through the ultrafast laser micro-shaping device or the ultrasonic-assisted micro-shaping device, the shaped workpiece to be detected is obtained, and real-time defect adjustment can be conducted in the manufacturing process through the method.

Description

Technical Field

[0001] This application relates to the field of laser manufacturing technology, and in particular to a method and system for adjusting defects in laser printing. Background Technology

[0002] With the rapid popularization of laser additive manufacturing technology in high-end manufacturing fields such as aerospace, medical devices, and precision molds, its application scenarios have expanded from prototype manufacturing to the direct production of key functional components. Especially in fields with stringent requirements for material performance, such as aero-engine blades and artificial joint implants, the quality of the cladding layer directly determines the product's service life and safety performance, which places higher demands on the quality control of laser additive manufacturing processes.

[0003] Currently, quality control methods in traditional laser additive manufacturing processes mainly rely on process parameter optimization and offline inspection. Defect detection and process adjustment are independent of each other, and the inspection results cannot be directly used for real-time optimization of process parameters. Consequently, it is difficult to achieve real-time control of the manufacturing process and cannot predict the impact of defects such as micropores and microcracks on the overall performance of components.

[0004] Therefore, how to perform real-time defect adjustment during the manufacturing process is a problem that urgently needs to be solved. Summary of the Invention

[0005] The main purpose of this application is to provide a laser printing defect adjustment method and system, which aims to solve the technical problem that the detection results cannot be directly used for real-time optimization of process parameters.

[0006] To achieve the above objectives, this application proposes a laser printing defect adjustment method, which is applied to a laser printing defect adjustment system. The laser printing defect adjustment system includes an ultrafast laser micro-shaping device and an ultrasonic-assisted micro-shaping device. The laser printing defect adjustment method includes:

[0007] Acquire physical and defect data of the workpiece under test;

[0008] Based on the physical data and defect data of the workpiece to be tested, a physical entity model is established to obtain the mapping relationship between microscopic data and macroscopic data;

[0009] The mapping relationship between the microscopic and macroscopic data is applied to the initial twin model to obtain the target twin model. The initial twin model includes a roughness calculation model and a residual stress calculation model.

[0010] The target twin model is trained based on historical defect data and current defect data to obtain a prediction model. The prediction model is used to predict defects in the workpiece under test. The prediction model includes a machine learning model and a time series analysis model.

[0011] The physical data of the workpiece to be tested is input into the prediction model to obtain the output results of the prediction model, which include process adjustment parameters.

[0012] Based on the process adjustment parameters, the workpiece to be tested is micro-shaped using the ultrafast laser micro-shaping device or the ultrasonic-assisted micro-shaping device to obtain the shaped workpiece.

[0013] In one embodiment, the output of the prediction model includes predicted defect data. Following the step of performing micro-reshaping on the workpiece to be tested using the ultrafast laser micro-reshaping device or the ultrasound-assisted micro-reshaping device based on the output of the prediction model to obtain the reshaped workpiece, the method further includes:

[0014] Based on the shaped workpiece under test, the current defect data is reacquired to obtain updated twin data;

[0015] Calculate the relative error between the predicted defect data and the current defect data;

[0016] If the relative error is greater than the preset error range, the target twin model is retrained based on the updated twin data until the relative error is less than or equal to the preset error range, and the current prediction model is obtained.

[0017] The model parameters of the current prediction model are stored in a component library, which is used to store physical data, defect data, processing parameters and their corresponding relationships.

[0018] In one embodiment, the laser printing defect adjustment system further includes a holographic camera, an X-ray stress detection device, and an industrial control computer. The defect data includes roughness detection data, hole distribution data, and residual stress data of the cladding layer. The step of acquiring the physical data and defect data of the workpiece to be tested includes:

[0019] The holographic camera acquires physical data and optical information of the cladding layer, the optical information of which includes refraction and scattering information.

[0020] The refraction and scattering information are processed by an industrial control computer to obtain roughness detection data and hole distribution data;

[0021] Residual stress data of the cladding layer were obtained using an X-ray stress detection device.

[0022] In one embodiment, the microscopic data includes atomic interaction potential functions, and the macroscopic data includes the bulk elastic modulus of the cladding layer and the bonding energy at the interface between the cladding layers. The step of establishing a physical entity model based on the physical data and defect data of the workpiece under test to obtain the mapping relationship between the microscopic and macroscopic data includes:

[0023] Based on the physical and defect data of the workpiece to be tested, the atomic interaction potential function is calculated using a molecular dynamics model.

[0024] The bulk elastic modulus of the cladding layer is calculated based on the atomic interaction potential function.

[0025] Based on the bulk elastic modulus of the cladding layer, the bonding energy at the interface between the cladding layers is calculated using the scratch method.

[0026] In one embodiment, the step of calculating the bonding energy of the cladding layer-to-cladding layer interface based on the bulk elastic modulus of the cladding layer using the scratch method includes:

[0027] The equivalent elastic modulus is determined based on Poisson's ratio and the bulk elastic modulus of the cladding layer;

[0028] The bonding energy between the cladding layers is calculated based on the critical load of the cladding layer, the scratch width, and the equivalent elastic modulus.

[0029] In one embodiment, the microscopic data further includes sampling locations, atomic masses, atomic velocities, atomic volumes, interatomic spacing, and interatomic interactions; the macroscopic data includes the number of sampling points and the number of atoms. The step of applying the mapping relationship between the microscopic and macroscopic data to the initial twin model to obtain the target twin model includes:

[0030] The relative height and average height of the sampling points are determined based on the sampling points.

[0031] The roughness of the cladding layer is calculated by combining the number of sampling points and the relative and average height of the sampling point locations;

[0032] The residual stress of a single atom is calculated based on the atomic mass, atomic velocity, atomic volume, interatomic spacing, and interatomic interaction forces.

[0033] Based on the number of atoms, the average stress and average biaxial stress of the single-atom layer are calculated according to the residual stress of the individual atom, and the residual stress of the cladding layer is obtained.

[0034] In one embodiment, the step of training the target twin model based on historical defect data and current defect data to obtain a prediction model, wherein the prediction model is used to predict defects in the workpiece under test, and the prediction model includes a machine learning model and a time series analysis model, comprising:

[0035] Based on the target twin model, multiple sets of simulated defect data are generated according to historical defect data and current defect data to obtain a sample dataset, which includes point cloud data of the cladding layer surface and time series data of process parameters.

[0036] The point cloud data of the cladding layer surface is input into a machine learning model for feature extraction to obtain the probability distribution of defect types.

[0037] The process parameter timing data is input into a timing analysis model for probability prediction to obtain the probability value of defect occurrence.

[0038] In one embodiment, the step of inputting the point cloud data of the cladding layer surface into a machine learning model for feature extraction to obtain the probability distribution of defect types includes:

[0039] The point cloud data of the cladding layer surface is input into a convolutional neural network for feature extraction to obtain key features, including edge contours and texture patterns.

[0040] The key features are input into a graph neural network to obtain a microstructure defect topology graph, which consists of multiple nodes and multiple edges. The nodes are used to represent atomic group properties, and the edges are used to represent the weights of interatomic forces.

[0041] The node states in the microstructure defect topology graph are updated by a message passing mechanism to obtain the defect type probability distribution.

[0042] In one embodiment, the step of inputting the process parameter timing data into a timing analysis model for probability prediction to obtain a defect occurrence probability value includes:

[0043] The time-series data of the process parameters are segmented by a sliding time window to extract the residual stress sequence and surface roughness sequence of the cladding layer.

[0044] The residual stress sequence and surface roughness sequence of the cladding layer are input into a long short-term memory network, and the probability value of defect occurrence is calculated through a gating mechanism.

[0045] Furthermore, to achieve the above objectives, this application also proposes a laser printing defect adjustment system, which includes:

[0046] The laser printing defect adjustment system includes an industrial control computer, a holographic camera, an X-ray stress detection device, a workpiece to be tested, an ultrafast laser micro-shaping device, and an ultrasonic-assisted micro-shaping device.

[0047] The holographic camera is connected to the industrial control computer via a spectrometer to acquire physical data and optical information of the cladding layer;

[0048] The X-ray stress detection device includes an X-ray generator, an X-ray detector, and an X-ray measurement recorder. The X-ray generator is used to emit X-rays to the surface area of ​​the cladding layer. The X-ray detector is used to detect the surface area of ​​the cladding layer and obtain information on the changes in the positions of the X-ray diffraction peaks. The X-ray measurement recorder is used to store the information on the changes in the positions of the X-ray diffraction peaks.

[0049] The ultrafast laser micro-plastic surgery device includes an ultrafast laser generator, an ultrafast laser optical path system, and an ultrafast laser probe. The ultrafast laser generator is used to generate a laser pulse sequence, the ultrafast laser optical path system is used to guide the laser to the target area on the surface of the cladding layer, and the ultrafast laser probe is used to locate the laser spot.

[0050] The ultrasonic-assisted micro-shaping device includes an ultrasonic vibrating head, a transducer, an ultrasonic transmitter, and an ultrasonic amplitude rod. The ultrasonic vibrating head is fixedly connected to the substrate of the workpiece to be tested. The ultrasonic transmitter is used to generate an electrical signal. The transducer is used to convert the electrical signal into mechanical vibration. The ultrasonic amplitude rod is used to transmit ultrasonic waves to the target area on the surface of the cladding layer.

[0051] The industrial control computer is used to receive and process the physical data, the optical information of the cladding layer, and the information on the changes in the position of the X-ray diffraction peaks.

[0052] The industrial control computer is also used to control the ultrafast laser micro-shaping device or the ultrasonic-assisted micro-shaping device to perform micro-shaping on the workpiece to be tested according to the process adjustment parameters as described above.

[0053] One or more technical solutions proposed in this application have at least the following technical effects:

[0054] The laser printing defect adjustment method is applied to the laser printing defect adjustment system, which includes an ultrafast laser micro-shaping device and an ultrasonic-assisted micro-shaping device. It acquires physical data and defect data of the workpiece under test, providing accurate data input for subsequent model building and prediction.

[0055] Based on the physical and defect data of the workpiece under test, a physical solid model is established to obtain the mapping relationship between microscopic and macroscopic data. By establishing a physical solid model that simultaneously includes a micromolecular dynamics model (microscopic) and a macroscopic finite element model (macroscopic), the correlation between atomic-level defects (such as micropores and microcracks) and macroscopic mechanical properties is directly established, solving the problem in traditional methods where microscopic defects are difficult to predict macroscopic effects.

[0056] By applying the mapping relationship between microscopic and macroscopic data to the initial twin model, the target twin model is obtained. The initial twin model includes a roughness calculation model and a residual stress calculation model, realizing the precise coupling between the physical entity and the virtual model at the atomic level and macroscopic mechanical properties, which significantly improves the characterization accuracy of defect data.

[0057] The target twin model is trained based on historical and current defect data to obtain a prediction model. This prediction model is used to predict defects in the workpiece under test. The prediction model includes a machine learning model and a temporal analysis model. The machine learning model is used to locate spatial defects, and the temporal analysis model is used to predict temporal anomalies. The physical data of the workpiece under test is input into the prediction model to obtain its output, which includes process adjustment parameters. By dynamically responding to defect changes during the manufacturing process, process adjustment parameters are directly output, eliminating the disconnect between detection results and process adjustments, and achieving a closed loop from defect detection to parameter optimization.

[0058] Based on process adjustment parameters, the workpiece to be tested is micro-shaped using an ultrafast laser micro-shaping device or an ultrasonic-assisted micro-shaping device to obtain the shaped workpiece. Micro-shaping includes at least one of ultrafast laser processing or ultrasonic processing. The optimal processing method is matched for different defect types, and real-time defect adjustment can be performed during the manufacturing process, which also greatly improves the adjustment accuracy. Attached Figure Description

[0059] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0060] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0061] Figure 1 This is a schematic diagram of the laser printing defect adjustment system of this application;

[0062] Figure 2This is a flowchart illustrating the first embodiment of the laser printing defect adjustment method of this application;

[0063] Figure 3 This is a flowchart illustrating the second embodiment of the laser printing defect adjustment method of this application.

[0064] Explanation of icon numbers:

[0065] 1. Industrial control computer; 2. Holographic camera; 301. X-ray generator; 302. X-ray detector; 303. X-ray measurement recorder; 4. Workpiece to be measured; 501. Ultrafast laser generator; 502. Ultrafast laser optical path system; 503. Ultrafast laser probe; 601. Ultrasonic vibrating head; 602. Transducer; 603. Ultrasonic transmitter; 604. Ultrasonic amplitude rod; 7. Spectrometer; 701. Microprocessor; 702. Filter; 703. Amplifier; 704. Analog-to-digital converter; 8. Current source; 9. Laser generator; 10. Laser 3D (or 4D) printing optical path scanning system; 11. Protective gas intake system; 12. Feeding system; 13. Laser processing head; 14. Molten pool; 15. Pulse generator; 16. Stage; 17. Cladding layer; 18. Pressure sensor; 19. Temperature sensor.

[0066] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0067] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0068] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0069] Traditional laser additive manufacturing processes rely primarily on process parameter optimization and offline inspection for quality control. This lack of correlation analysis between microscopic defect evolution and macroscopic performance degradation makes it impossible to predict the impact of defects such as micropores and microcracks on the overall component performance. Furthermore, traditional defect detection and process adjustment are independent, meaning detection results cannot be directly used for real-time process parameter optimization. The lack of effective data fusion and analysis also leads to insufficient defect prediction accuracy. Therefore, how to perform real-time defect adjustment during manufacturing is a pressing issue that needs to be addressed.

[0070] This application provides a laser printing defect adjustment method, which is applied to a laser printing defect adjustment system. To make the following embodiments clearer, the laser printing defect adjustment system will be described first.

[0071] Based on this, embodiments of this application provide a laser printing defect adjustment system, referring to... Figure 1 , Figure 1 This is a schematic diagram of the laser printing defect adjustment system of this application.

[0072] In this embodiment, the laser printing defect adjustment system includes an industrial control computer 1, a holographic camera 2, an X-ray stress detection device, a workpiece to be tested 4, an ultrafast laser micro-shaping device, and an ultrasonic-assisted micro-shaping device. The holographic camera 2 is connected to the industrial control computer 1 via a spectrometer 7 to acquire physical data and optical information of the cladding layer. The X-ray stress detection device includes an X-ray generator 301, an X-ray detector 302, and an X-ray measurement recorder 303. The X-ray generator 301 emits X-rays onto the surface area of ​​the cladding layer 17, the X-ray detector 302 detects the surface area of ​​the cladding layer 17 and obtains information on the changes in the position of the X-ray diffraction peaks, and the X-ray measurement recorder 303 stores the information on the changes in the position of the X-ray diffraction peaks. The ultrafast laser micro-shaping device includes an ultrafast laser generator 501, an ultrafast laser optical path system 502, and an ultrafast laser probe 503. The generator 501 is used to generate a laser pulse sequence, the ultrafast laser optical path system 502 is used to guide the laser to the target area on the surface of the cladding layer 17, and the ultrafast laser probe 503 is used to locate the laser spot; the ultrasonic-assisted micro-shaping device includes an ultrasonic vibration head 601, a transducer 602, an ultrasonic transmitter 603, and an ultrasonic amplitude rod 604. The ultrasonic vibration head 601 is fixedly connected to the substrate of the workpiece 4 to be tested. The ultrasonic transmitter 603 is used to generate an electrical signal, the transducer 602 is used to convert the electrical signal into mechanical vibration, and the ultrasonic amplitude rod 604 is used to transmit ultrasonic waves to the target area on the surface of the cladding layer 17; the industrial control computer 1 is used to receive and process physical data, optical information of the cladding layer, and information on changes in the position of X-ray diffraction peaks; the industrial control computer 1 is also used to control the ultrafast laser micro-shaping device or the ultrasonic-assisted micro-shaping device to perform micro-shaping on the workpiece 4 to be tested according to the process adjustment parameters obtained in the laser printing defect adjustment method.

[0073] It should be noted that the target area on the surface of the cladding layer 17 can be understood as the area on the surface of the cladding layer 17 that needs to be shaped. For example, X-rays are used to irradiate the surface area of ​​the cladding layer 17 to test which location on the surface area of ​​the cladding layer 17 has the greatest residual stress, and this location is defined as the target area on the surface of the cladding layer 17. During micro-shaping, the target area on the surface of the cladding layer 17 is micro-shaped using laser or ultrasound. The laser printing defect adjustment system also includes a digital signal feedback module and a current source 8. The digital signal feedback module includes a microprocessor 701, a filter 702, an amplifier 703, and an analog-to-digital converter 704, which are used to monitor, detect, and provide feedback on the strength of the electrical signal of the ultrasonic transmitter 603 in real time and convert it into a corresponding digital signal. The strength of the electrical signal of the ultrasonic transmitter 603 is controlled by the current source 8.

[0074] Specifically, the laser printing defect adjustment system also includes a laser generator 9, a laser 3D (or 4D) printing optical path scanning system 10, a protective gas intake system 11, a feeding system 12, and a laser processing head 13. For example, the industrial control computer 1 sends a laser command to the feeding system 12, feeding a thin layer of powder into the laser processing head 13. Simultaneously, the industrial control computer 1 sends a laser command to the laser generator 9, controlling the laser to pass through the laser 3D (or 4D) printing optical path scanning system 10 to the laser processing head 13, emitting a high-energy laser to "sinter" the thin layer of powder and metal powder onto the bottom layer of the workpiece 4 to be tested, gradually forming a cladding layer 17. At the same time, inert gas is introduced through the protective gas intake system 11.

[0075] Specifically, the laser printing defect adjustment system also includes a molten pool 14 and a pulse generator 15. The pulse generator 15 is positioned near the molten pool 14 in front of the cladding layer 17, and emits a stress wave into the detection area. Because the cladding layer 17 has a certain roughness and porosity, discontinuous areas on the surface will generate a series of wave signals after refraction and diffuse scattering. The holographic camera 2 also integrates a stress wave sensor, which receives the wave signals and sends them to the industrial control computer 1. The laser printing defect adjustment system also includes a stage 16, the cladding layer 17, a pressure sensor 18, and a temperature sensor 19. The stage 16 is used to place the workpiece 4 to be tested. The pressure sensor 18 and the temperature sensor 19 are positioned around the workpiece 4 to acquire physical data, such as ambient temperature and pressure data.

[0076] In this embodiment, the laser printing defect adjustment system suppresses crack initiation and reduces or eliminates porosity between the surface or interface of the cladding layer by using an ultrafast laser micro-shaping device and an ultrasonic-assisted micro-shaping device, thereby improving the hardness and wear resistance of the laser-printed workpiece. By altering the microstructure of the cladding layer, especially by refining the grains and suppressing the columnar crystal orientation of epitaxial growth, its tensile properties and yield strength are improved. The industrial control computer controls the ultrafast laser and ultrasonic vibration device to perform ultrafast laser micro-shaping and ultrasonic-assisted processing on the workpiece according to process adjustment parameters, resulting in better flatness of the cladding layer interface and surface. The ultrafast laser also provides a secondary polishing effect on the rough surface of the cladding layer, thus obtaining a laser-clad printed workpiece with a relatively flat cladding layer and no pores.

[0077] Based on this, embodiments of this application provide a laser printing defect adjustment method, referring to... Figure 2 , Figure 2 This is a flowchart illustrating the second embodiment of the laser printing defect adjustment method of this application.

[0078] In this embodiment, the laser printing defect adjustment method is applied to the laser printing defect adjustment system in the above embodiment. The laser printing defect adjustment system includes an ultrafast laser micro-shaping device and an ultrasonic-assisted micro-shaping device. The laser printing defect adjustment method includes steps S10 to S60:

[0079] Step S10: Obtain the physical data and defect data of the workpiece to be tested.

[0080] It should be noted that physical data includes geometric properties, material properties, and mechanical properties, while defect data includes roughness inspection data and residual stress data of the cladding layer. The workpiece under test can be understood as a laser 3D (or 4D) printed workpiece undergoing quality inspection, prediction, or shaping. Geometric properties can be spatial information such as the shape, size, angle, and contour of the workpiece (e.g., length, width, height, diameter, surface shape, etc.). Material properties can be the essential characteristics of the materials constituting the workpiece (e.g., chemical composition, crystal structure, density, specific heat capacity, thermal conductivity, electrical conductivity, etc.). Mechanical properties can be the behavioral characteristics of the workpiece under stress (e.g., elastic modulus, yield strength, tensile strength, hardness, toughness, fatigue strength, etc.). Roughness inspection data can be used to quantify the degree of microscopic unevenness on the workpiece surface, and residual stress data of the cladding layer can be understood as the data of residual, unreleased internal stress within the cladding layer formed on the workpiece surface during additive manufacturing (e.g., laser cladding) due to rapid heating and cooling.

[0081] Step S20: Based on the physical data and defect data of the workpiece to be tested, establish a physical entity model to obtain the mapping relationship between microscopic data and macroscopic data.

[0082] It should be noted that physical entity models include micromolecular dynamics models and macroscopic finite element models. Microscopic data is used to describe interatomic interactions, while macroscopic data is used to describe mechanical properties. A physical entity model can be understood as a computational model based on physical laws (such as Newtonian mechanics, thermodynamics, and material constitutive relations) used to simulate the physical behavior of a workpiece. Micromolecular dynamics models simulate material behavior at the atomic or molecular scale. The material of the workpiece under test can be considered as a large number of interacting atoms or molecules, and their positions, velocities, and interaction forces can be calculated based on potential energy functions to study microscopic mechanisms (such as dislocation motion, grain boundary behavior, and atomic diffusion). Microscopic data refers to this atomic-scale information. Macroscopic finite element models simulate the overall mechanical and thermal behavior of the workpiece at the macroscopic scale (under the assumption of a continuous medium). The workpiece can be discretized into a finite number of small elements (mesh), and physical equations (such as equilibrium equations and constitutive equations) can be applied to each element to solve for the response of the entire structure (such as stress, strain, displacement, and temperature distribution). Macroscopic data refers to information on the mechanical properties of a whole or local area (such as overall stress field, strain field, deformation, modes, etc.). The mapping relationship between microscopic and macroscopic data can be understood as a bridge established through physical models, allowing atomic-scale information obtained from microscopic simulations (such as local stress and material parameter variations) to be reasonably upscaled or averaged to the macroscopic scale for use in correcting or driving the calculations of the macroscopic model, or vice versa. This overcomes the limitations of single-scale models.

[0083] Step S30: Apply the mapping relationship between microscopic data and macroscopic data to the initial twin model to obtain the target twin model. The initial twin model includes a roughness calculation model and a residual stress calculation model.

[0084] It should be noted that the initial twin model is a virtual mapping of the physical entity, reflecting its state, behavior, and performance in real-time or near real-time. For example, the initial twin model can be built based on general knowledge or preliminary data; it is understood that the initial twin model has not yet been fully calibrated for the workpiece under test. The roughness calculation model can be used to predict or calculate the surface roughness of the workpiece, and the residual stress calculation model can be used to predict or calculate the residual stress of the workpiece under test (especially the cladding layer). The target twin model is a more accurate digital twin model enhanced by physical mechanisms, obtained by applying the mapping relationship between microscopic and macroscopic data to the initial twin model. It combines the mechanistic aspects of physical modeling with the dynamic updating capability of the twin model.

[0085] Step S40: Train the target twin model based on historical defect data and current defect data to obtain a prediction model. The prediction model is used to predict defects in the workpiece under test. The prediction model includes a machine learning model and a time series analysis model.

[0086] It should be noted that historical defect data can refer to accumulated, known defect data from the past, while current defect data can be understood as the latest, real-time detected actual defect data. Machine learning models are used to locate spatial defects, while time-series analysis models are used to predict temporal anomalies. For example, a machine learning model can predict which area on the surface of a workpiece is prone to cracks, porosity, or excessive roughness. A time-series analysis model can predict when a key parameter (such as vibration, temperature, or residual stress relaxation) of the workpiece will exceed the normal range during processing or service.

[0087] Step S50: Input the physical data of the workpiece to be tested into the prediction model to obtain the output results of the prediction model, which include process adjustment parameters.

[0088] It should be noted that process adjustment parameters can be specific parameter recommendations for adjusting subsequent manufacturing or repair processes in order to improve or eliminate predicted defects. Examples include adjusting laser power, scanning speed, scanning strategy, spot size, etc.; adjusting ultrasonic power, frequency, amplitude, and action time, etc.; and adjusting heat treatment temperature, time, processing path, and speed, etc.

[0089] Step S60: Based on the process adjustment parameters, the workpiece to be tested is micro-shaped using an ultrafast laser micro-shaping device or an ultrasonic-assisted micro-shaping device to obtain the shaped workpiece to be tested.

[0090] For example, micro-shaping includes at least one of ultrafast laser processing or ultrasonic processing. Process adjustment parameters may include adjusting laser power, scanning speed, scanning strategy, spot size, etc.; adjusting ultrasonic power, frequency, amplitude, and action time, etc.; and adjusting heat treatment temperature, time, processing path, and speed, etc. Ultrafast laser micro-shaping devices can use extremely short pulse (picosecond, femtosecond) lasers to precisely ablate, modify, or reconstruct the workpiece. Ultrasonic-assisted micro-shaping devices can superimpose high-frequency mechanical vibrations (ultrasound) during traditional processing (such as cutting, grinding, and impact). Ultrasonic vibration can reduce cutting forces, improve surface finish, and increase material removal efficiency.

[0091] In this embodiment, physical data and defect data of the workpiece under test are acquired to provide accurate data input for subsequent model building and prediction. By establishing a physical entity model that simultaneously includes a micromolecular dynamics model (microscopic) and a macroscopic finite element model (macroscopic), the correlation between atomic-level defects (such as micropores and microcracks) and macroscopic mechanical properties is directly established, solving the problem in traditional methods where microscopic defects are difficult to predict macroscopic effects. By applying the mapping relationship between microscopic and macroscopic data to the initial twin model, a target twin model is obtained, achieving precise coupling between the physical entity and the virtual model at the atomic level and macroscopic mechanical properties, significantly improving the characterization accuracy of defect data. Based on historical and current defect data, the target twin model is trained to obtain a prediction model. The physical data of the workpiece under test is then input into the prediction model to obtain its output, which includes process adjustment parameters. By dynamically responding to defect changes during the manufacturing process, process adjustment parameters are directly output, eliminating the disconnect between detection results and process adjustments, and achieving a closed loop from defect detection to parameter optimization. Based on process adjustment parameters, the workpiece under test is micro-shaped using an ultrafast laser micro-shaping device or an ultrasonic-assisted micro-shaping device. The optimal treatment method is matched for different defect types, and real-time defect adjustment can be performed during the manufacturing process, which also greatly improves the adjustment accuracy.

[0092] Reference Figure 3 , Figure 3 This is a flowchart illustrating the second embodiment of the laser printing defect adjustment method of this application, based on the above. Figure 2 The first embodiment shown presents a second embodiment of the laser printing defect adjustment method of this application.

[0093] In the second embodiment, after step S60, the method further includes:

[0094] Step S70: Based on the shaped workpiece to be tested, reacquire the current defect data to obtain updated twin data.

[0095] It should be noted that the updated twin data can be understood as the physical data and defect data re-collected after the workpiece under test is shaped, which is used to refresh the model parameters of the target twin model.

[0096] Step S80: Calculate the relative error between the predicted defect data and the current defect data.

[0097] It should be noted that relative error is used to quantify the accuracy of the prediction model.

[0098] Step S90: If the relative error is greater than the preset error range, retrain the target twin model based on the updated twin data until the relative error is less than or equal to the preset error range, and obtain the current prediction model.

[0099] It should be noted that the preset error range is a pre-defined acceptable error threshold, and the specific setting can be determined by process requirements. The current prediction model can be understood as the latest prediction model that meets the accuracy requirements after iterative training.

[0100] Step S100: Store the model parameters of the current prediction model in the component library. The component library is used to store physical data, defect data, processing parameters and their corresponding relationships.

[0101] It should be noted that the component library can be understood as a structured database used to store physical data, defect data, processing parameters, and their mapping relationships, supporting rapid data retrieval and reuse. For example, surface roughness indicators for laser cladding 3D (or 4D) printing can be defined and stored in an industrial control computer processing system. Surface roughness indicators can include the following levels: Level N1 (R... z : 0.1μm~0.4μm) indicates ultra-fine machining; grade N2 (R z : 0.4μm~0.4μm) indicates extremely fine processing; grade N3 (R z (1.5μm~6.3μm) indicates ultra-fine machining; grade N4 (R z (6.3μm~12.5μm) indicates semi-finished; grade N5 (R) z (12.5μm~25μm) indicates medium processing; grade N6 (R) z : 25μm~50μm) indicates rough machining; grade N7 (R z 50μm to 100μm indicates extremely rough machining. In the next working condition, real-time data from online monitoring and detection is transmitted to the industrial control computer, where the surface roughness index is judged and automatically assigned to the corresponding level. For example, if a large-scale surface microstructure defect occurs, such as defect level N5 (R... z (12.5μm~25μm) A large database structure (component library) has been established through the target twin model. By using physical data, defect data, processing parameters and their mapping relationship (i.e., the one-to-one correspondence between the surface defect level of the microstructure and the micro-shaping process parameters), the improvement data of the micro-shaping process parameters can be directly obtained, and micro-shaping can be performed on the workpiece to be tested.

[0102] In this embodiment, by comparing the real-time collected data after shaping with the predicted results, the model is driven to iteratively train, continuously bringing the prediction accuracy closer to the actual process requirements, forming a closed loop of detection, prediction, shaping, and verification. Simultaneously, the standardized data and model parameters accumulated in the component library provide plug-and-play process solutions for similar workpieces, significantly reducing the cost of repeated trial and error, improving the self-optimization capability and knowledge reuse efficiency of the intelligent manufacturing system, and ultimately achieving precise defect control and intensive use of production resources. By automatically optimizing the current experimental parameters, not only are human and financial resources saved, but experimental efficiency is also improved. Real-time monitoring and quality assurance are also provided, ensuring the stability and reliability of laser cladding 3D (or 4D) printing.

[0103] To make the above embodiments clearer, this application proposes various implementation methods based on the above embodiments.

[0104] In one embodiment, the defect data includes roughness detection data, void distribution data, and residual stress data of the cladding layer. Step S10 includes: acquiring physical data and optical information of the cladding layer using a holographic camera, the optical information of the cladding layer including refraction information and scattering information; processing the refraction information and scattering information using an industrial control computer to obtain roughness detection data and void distribution data; and acquiring residual stress data of the cladding layer using an X-ray stress detection device.

[0105] It should be noted that refraction information can be data on the directional deflection of light as it passes through the cladding layer, reflecting the scattering information of the material's density or composition distribution. Scattering information can be data on diffuse reflection of light on or within the cladding layer due to inhomogeneities (such as particles or pores). X-ray stress detection equipment can measure changes in lattice spacing and calculate residual stress within the cladding layer.

[0106] In this embodiment, the holographic camera simultaneously acquires surface morphology (roughness) and surface defects (holes) through non-contact optical scanning, overcoming the limitations of traditional contact probes that are prone to damaging workpieces and single-point measurement; the X-ray device accurately quantifies internal residual stress, forming a full-dimensional defect monitoring system, providing high-fidelity input for subsequent prediction and micro-shaping, and ensuring the accuracy and real-time performance of quality control from the source.

[0107] In one implementation, it should be noted that the microscopic data includes atomic interaction potential functions, and the macroscopic data includes the bulk elastic modulus of the cladding layer and the bonding energy between the cladding layers. The physical entity model includes a micromolecular dynamics model and a macroscopic finite element model. Specifically, the step of establishing a physical entity model based on the physical data and defect data of the workpiece under test to obtain the mapping relationship between the microscopic and macroscopic data includes: calculating the atomic interaction potential function using a molecular dynamics model based on the physical data and defect data of the workpiece under test; calculating the bulk elastic modulus of the cladding layer based on the atomic interaction potential function; and calculating the bonding energy between the cladding layers using the scratch method based on the bulk elastic modulus of the cladding layer.

[0108] It should be noted that the step of calculating the bonding energy of the interface between the cladding layers based on the bulk elastic modulus of the cladding layer using the scratch method includes: determining the equivalent elastic modulus based on Poisson's ratio and the bulk elastic modulus of the cladding layer; and calculating the bonding energy of the interface between the cladding layers based on the critical load of the cladding layer, the scratch width, and the equivalent elastic modulus.

[0109] Specifically, the atomic interaction potential function is expressed as formula (1).

[0110]

[0111] In formula (1), E i The sum of the potential energies of the metals is represented by α and β, where α and β represent the atomic nuclides of the atoms, respectively, and φ represents the two-body potential, which is the sum of the potential energies of the atomic nuclides α and the potential energy of the atoms with respect to the interatomic distance r. ij The function, F, represents the many-body potential, which is the potential energy generated by the interaction between the sum of the electron densities contributed by atom j at atom i and atom i. It depends on the atom type α of atom i and the electron density at i. ρ is the electron density provided by atom i to atom j, which is related to the atom type β of j and the interatomic distance r. ij The molecular dynamics potential function is also known as the atomic interaction potential function (eam potential function). In the formula for the eam potential function, the electron density ρ is related to the type of atom, so its value is fixed. However, the two-body potential φ and the many-body potential F are both related to the interatomic spacing, which in turn affects the roughness calculation model and the residual stress calculation model.

[0112] It should be noted that the minimum energy structure of the laser cladding 3D (or 4D) printed workpiece material is calculated using the eam potential function, and the bulk modulus at equilibrium stability is obtained through the minimum energy structure of the system.

[0113] The bulk elastic modulus of the cladding layer is the scalar ratio of the bulk stress to the bulk strain of the system, expressed as formula (2).

[0114]

[0115] In formula (2), B represents the bulk modulus, p is the bulk stress or pressure, and dV / V is the bulk strain. It should be noted that for a cubic unit cell: the total energy of the system is E. all =-NE, the total volume of the system is V all =-a 3 Where N is the total number of atoms in the system, E is the energy of a single atom, and a is the lattice constant of the unit cell. The bulk stress can be expressed as formula (3).

[0116]

[0117] Therefore, the bulk modulus can also be expressed as formula (4).

[0118]

[0119] In formula (4), a0 represents the equilibrium lattice constant.

[0120] The mechanical parameters of the macroscopic model also include the bonding energy between the cladding layers and the cladding layer interface, which is expressed by formula (5).

[0121]

[0122] In formula (5), G represents the bonding energy at the interface between the cladding layers, and L... c For the critical load, E * ω is the equivalent elastic modulus, and ω is the scratch width. It should be noted that the equivalent elastic modulus can be calculated using formula (6).

[0123] E * =3B(1-2v) (6)

[0124] In formula (6), v represents Poisson's ratio.

[0125] In this embodiment, a molecular dynamics model is constructed to calculate the surface roughness of the cladding layer and the residual stress between the cladding layers. This reduces the complexity of the model calculation and improves the computational efficiency of the associated twin model.

[0126] In one implementation, the microscopic data further includes sampling locations, atomic masses, atomic velocities, atomic volumes, interatomic spacing, and interatomic interactions; the macroscopic data includes the number of sampling locations and the number of atoms. The step of applying the mapping relationship between the microscopic and macroscopic data to the initial twin model to obtain the target twin model includes: determining the relative height and average height of the sampling locations based on the sampling locations; calculating the cladding layer roughness by combining the number of sampling locations and the relative and average heights of the sampling locations; calculating the residual stress of a single atom based on the atomic mass, atomic velocity, atomic volume, interatomic spacing, and interatomic interactions; and calculating the average stress and average biaxial stress of a single atomic layer based on the residual stress of a single atom, thus obtaining the residual stress of the cladding layer.

[0127] Specifically, the roughness calculation model is expressed as formula (7).

[0128]

[0129] In formula (7), R z This represents the average surface roughness, where M and N are the number of points taken from the plane, and x... j y i Here, x and y are the x and y coordinates of the point's location in the plane, respectively, and Z represents the relative height value. Z represents the average height of all points on the reference surface. For example, the relative height value Z of points on the cladding layer surface is acquired and collected using a holographic camera, where the average height of the cladding layer is equal to the thickness of a single cladding layer. Real-time data of the ordinate values ​​of all points within the surface are stored. The real-time and historical roughness data serve as a sample space for machine learning. Machine learning is then used to perform deep learning on these large sample spaces of surface arithmetic mean roughness. By minimizing a loss function, the data patterns and features of the computer's deep learning of surface arithmetic mean roughness are compared with the thickness values ​​of a single cladding layer set and referenced. The difference between them is addressed by continuously adjusting the set and reference thickness values ​​of individual cladding layers by minimizing the loss function. This continuously improves the high prediction accuracy of the arithmetic mean roughness of the microstructure surface. If the arithmetic mean roughness of the microstructure surface of the laser-clad 3D (or 4D) printed workpiece exceeds a preset threshold, the industrial control computer uses data trained in the sample space to perform ultrafast laser micro-shaping on the cladding surface. The ultrafast laser can generate instantaneous pulses that can instantly disperse and detach the protruding material clustered in the cladding layer, and also disperse the particle clusters agglomerated on the cladding surface, thus giving the interface and surface of the cladding layer of the workpiece under test better surface smoothness. The ultrafast laser can also play a secondary polishing role on the rough surface of the cladding layer. The process stops when the arithmetic mean roughness of the microstructure surface of the laser-clad 3D / 4D printed workpiece is less than the preset threshold.

[0130] The residual stress calculation model is expressed as Equation (8).

[0131]

[0132] In formula (8), σ represents atomic residual stress, and m i v represents the mass of atom i. i V represents the velocity of atom i. i r represents the volume of atom i. ij and f ij This represents the interaction distance and interaction force between atoms i and j. The tensor product of vectors is represented. The stress in a single atomic layer can be expressed by the average stress of the single atomic layer, as shown in formula (9).

[0133]

[0134] In formula (9), Let n represent the average stress of the atomic layer, m represent the stress tensor exponents, and N represent the number of atoms. The evolution of residual stress between cladding layers is characterized by the biaxial stress between cladding layers deposited by laser cladding 3D (or 4D) printing, so as to analyze the influence of residual stress on the deposition quality of the film during the deposition process. The average biaxial stress can then be derived, expressed as formula (10).

[0135]

[0136] In formula (10), Indicates the average biaxial stress. and These represent the average stresses along the X and Y axes of the atomic layers, respectively. Further, the residual stresses between the cladding layers are obtained, expressed as formula (11).

[0137]

[0138] In formula (11), σ represents the residual stress between the cladding layers.

[0139] In one implementation, step S40 includes: generating multiple sets of simulated defect data based on the target twin model, according to historical defect data and current defect data, to obtain a sample dataset, the sample dataset including cladding layer surface point cloud data and process parameter time series data; inputting the cladding layer surface point cloud data into a machine learning model for feature extraction to obtain a defect type probability distribution; inputting the process parameter time series data into a time series analysis model for probability prediction to obtain a defect occurrence probability value.

[0140] It should be noted that virtual defect datasets generated using the target twin model can be used to expand the training samples and cover more potential anomaly scenarios. For example, the point cloud data of the cladding layer surface can be a set of three-dimensional coordinates, describing the spatial location of each point on the cladding layer surface. The time-series data of process parameters can be understood as processing parameters recorded in chronological order. The probability distribution of defect types can be a probability vector of various defects (such as cracks, porosity, and lack of fusion) occurring, used to predict in real time the distribution of defects in the cladding layer structure (such as roughness, voids, and microcracks) and the degree of mechanical property damage, and to determine the feasibility of the current laser process parameters. The probability value of defect occurrence can be the probability of a defect occurring at a future time.

[0141] For example, since the digital twin model of the cladding layer deposition in laser cladding 3D (or 4D) printing can maintain a high degree of consistency with the in-situ elastic force data values ​​of the physical entity model, a sample space can be constructed using a twin model that correlates the evolution of microstructural defects in laser cladding 3D (or 4D) printing with the macroscopic interface mechanical properties of laser cladding 3D (or 4D) printing. During construction, the relative height values ​​Z of a large number of different sampling points are obtained by changing the positions of sampling points in the roughness model; the distances between a large number of different sampling points are obtained by changing the positions of atomic sampling points in the microstructural mechanical defect model of the cladding layer of the laser cladding 3D (or 4D) printed workpiece, thereby obtaining the average stress of single atomic layers of different sizes. By changing the scratch width and critical load in the bonding energy model of the cladding layer interface, different interface bonding energy values ​​G are obtained; by changing the lattice constant value in the molecular dynamics calculation model of deposition cladding, different bulk modulus values ​​B are obtained, thus obtaining a large sample space.

[0142] Specifically, the steps of inputting the surface point cloud data of the cladding layer into a machine learning model for feature extraction to obtain the probability distribution of defect types include: inputting the surface point cloud data of the cladding layer into a convolutional neural network for feature extraction to obtain key features, including edge contours and texture patterns; inputting the key features into a graph neural network to obtain a microstructure defect topology graph, which consists of multiple nodes and multiple edges, where nodes represent atomic cluster attributes and edges represent interatomic force weights; and updating the node states in the microstructure defect topology graph through a message passing mechanism to obtain the probability distribution of defect types.

[0143] For example, point cloud data (roughness R) of the cladding layer surface can be acquired using a holographic camera. z Input the material properties of the cladding layer (melting point, elastic modulus) and the mechanical indices of the macroscopic finite element model (bulk elastic modulus, interfacial bonding energy), and extract surface texture features (such as rough stripes and hole edges) through a convolutional neural network (CNN). Combine this with the microstructure topology map constructed by a graph neural network (GNN) to locate high-risk defect areas (such as areas where stress concentration and roughness abrupt changes overlap). CNN specifically includes extracting spatial features of the cladding layer surface image, such as roughness texture and hole edge gradient. It can extract features from the surface point cloud data collected by the holographic camera through multiple convolutional kernels (such as 3×3 convolutional kernels), including: the first convolutional layer extracts the edge contour (identifying hole boundaries) and the second convolutional layer extracts the texture pattern (distinguishing the spacing of rough stripes). At the same time, the feature dimension is reduced by pooling layers (such as max pooling) to retain key defect features. The surface point cloud data is converted into grayscale images (height values ​​are mapped to pixel values) through CNN, input into the CNN network, and output the probability distribution of defect types (such as the probability of roughness exceeding the standard and the probability of hole existence). GNN specifically includes modeling the topological relationship of the cladding layer microstructure and capturing the spatial correlation of defect evolution. It can abstract the cladding layer microstructure into a graph structure, including nodes representing atomic clusters or grains (attributes are atomic stress and potential energy), and edges representing inter-atomic interactions (weights are the forces calculated by the EAM potential function). Nodes and edges iteratively update the node state through a message passing mechanism, specifically as shown in formula (12).

[0144]

[0145] In formula (12), Let N(v) be the feature vector of node v in layer l, N(v) be the neighboring nodes, cv,u be the normalization coefficients, σ ​​be the activation function, and W be the function vector. l and All parameters are learned parameters. The probability of grain boundary crack propagation in the cladding layer is calculated using a GNN, and the atomic stress tensor from molecular dynamics simulations is combined to predict the location of microcrack initiation.

[0146] For example, a nonlinear mapping model of "process parameters and defect indexes" containing target twin model parameters (such as roughness and residual stress) can also be established, expressed as formula (13).

[0147] P 可行性 =σ(∑w i ·(R z ·α+σ res ·β+E bond ·γ)) (13)

[0148] In formula (13), σ is the activation function, wi is the weight coefficient, α, β, and γ are the defect index weights, and E bond For interfacial bonding energy; σ res For residual stress, R z The surface roughness is taken as an average. Furthermore, by inputting the microstructural damage parameters (such as grain boundary crack density) output by the CNN into the macroscopic finite element model, the tensile strength and yield strength attenuation ratio of the cladding layer are calculated, thereby achieving prediction of mechanical effects.

[0149] Specifically, the steps of inputting process parameter time series data into a time series analysis model for probability prediction to obtain the probability value of defect occurrence include: segmenting the process parameter time series data by a sliding time window, extracting the residual stress sequence and surface roughness sequence of the cladding layer; inputting the residual stress sequence and surface roughness sequence of the cladding layer into a long short-term memory network, and calculating the probability value of defect occurrence through a gating mechanism.

[0150] For example, time series analysis can employ a Long Short-Term Memory (LSTM) network combined with an attention mechanism to adapt to the dynamic characteristics of process parameter time series data. Process parameter time series data may include the surface average roughness R. z Residual stress σ res Process parameters (laser power P, scanning speed v) are input into a time-series analysis model to obtain defect probability predictions and early warning time points for abnormal events (such as stress mutations) in the next 3-5 layers of the cladding layer. The model can perform sliding window segmentation (window size 100-200 layers) on the real-time data stream to extract time-series features; introduce an attention mechanism to focus on high-weight features (such as stress gradient change rate) and suppress noise interference; and optimize the number of LSTM layers (e.g., 2 layers) and hidden layer nodes (e.g., 128 nodes) through cross-validation.

[0151] When 5 consecutive layers of R zWhen the value increases at a rate of 0.1 μm / layer, a "roughness deterioration" warning can be triggered. When the residual stress σ... res When the pressure fluctuates by more than 20 MPa within 10 minutes, it can be identified as a risk of "stress instability".

[0152] LSTM specifically involves controlling the flow of information through gating mechanisms (forget gate, input gate, output gate) to capture long-term dependencies in time-series data, such as the evolution trend of residual stress in the cladding layer with the number of laser-printed layers. LSTM introduces residual stress to characterize the gradient relationship of residual stress in the cladding layer of laser 3D (or 4D) printed layers with the number of layers, specifically including formulas (14) to (18).

[0153] f t =σ(W f ·[h t-1 ,x t ]+b f (14)

[0154] i t =σ(W i ·[h t-1 ,x t ]+b i (15)

[0155] c t =f t ⊙c t-1 +i t ⊙tanh(W c ·[h t-1 ,x t ]+b c (16)

[0156] o t =σ(W o ·[h t-1 ,x t ]+b o (17)

[0157] h t =o t ⊙tanh(ct) (18)

[0158] Formulas (14) to (18) represent the forgetting gate, the input gate, the cell state update, and the cell state update, respectively.

[0159] Output gate and hidden state. Specifically, f t W represents the Forgotten Gate. f The weight matrix of the forget gate, [h t-1 ,x t The symbol ] represents concatenating two vectors into a longer vector, b. fThis represents the bias term of the forget gate, σ is the sigmoid function, and i t W represents the input gate. i It is the weight matrix of the input gate, b i It is the bias term of the input gate, x t Input at time t (e.g., R) z (value, stress value), h t To be in a hidden state, W c The weight matrix b represents the cell state update. f c represents the cell state update bias term. t In cellular state, W o Let b represent the weight matrix of the output gate. o This represents the output bias term; the further input sequence is the roughness R of the most recent 50 layers. z Residual stress σ res Feasibility P, output the predicted probability of defects in the next 3 layers.

[0160] Furthermore, by weighting the feature vectors of historical anomalous events through an attention mechanism, a probability prediction model is constructed to achieve the calculation of anomalous probability. The formula for calculating anomalous probability can be expressed as formula (19).

[0161]

[0162] In formula (19), a j h is the attention weight for the j-th historical anomaly feature. j This is the output of the LSTM hidden layer.

[0163] It should be noted that the machine learning model and the time series analysis model are linked in the data flow. Specifically, the machine learning model directly reads the microstructural parameters (atomic spacing, potential energy distribution) of the target twin model and calls the real-time roughness detection data from the holographic camera (sampling frequency can be 100Hz). The time series analysis model accesses the time series data (sampling interval 5s) from the X-ray stress detection device and simultaneously obtains the process parameter logs (laser power, scanning speed adjustment records) from the industrial control computer.

[0164] The target twin model serves as the foundational training model, achieving a closed loop through iterative optimization. Specifically, the atomic stress tensor calculated by the molecular dynamics model provides the basis for the evolution of microscopic defects in the machine learning model; the stress field distribution output by the macroscopic finite element model sets the initial threshold for time-series monitoring in the machine learning model. The prediction results of the machine learning model are fed back to the target twin model to correct material property parameters (such as updating the elastic modulus); anomaly warnings from the time-series analysis model trigger rapid simulations in the digital twin model, generating process adjustment suggestions (such as reducing laser power by 10%). The machine learning model and the time-series analysis model are retrained using the micro-adjustment detection data (such as roughness improvement values), forming a closed loop of "prediction-verification-optimization".

[0165] In this embodiment, a cascaded model of convolutional neural network and graph neural network is constructed to correlate the surface morphology (point cloud) of the cladding layer with microscopic atomic forces (topological graph), accurately identifying the crack initiation location. Long Short-Term Memory (LSTM) network utilizes a sliding window to capture the dynamic coupling relationship between residual stress and roughness; the synergy of the two models significantly improves prediction accuracy. The machine learning model achieves a nonlinear mapping from defect types to process parameters, overcoming the limitations of empirical parameter tuning. The time-series analysis model combines LSTM and digital twins to achieve 4-6 layers of advance prediction of dynamic time-series anomalies, enabling real-time control of the manufacturing process.

[0166] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the laser printing defect adjustment method of this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0167] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0168] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A method for adjusting defects in laser printing, characterized in that, The laser printing defect adjustment method is applied to a laser printing defect adjustment system, which includes an ultrafast laser micro-shaping device and an ultrasonic-assisted micro-shaping device. The laser printing defect adjustment method includes: Acquire physical and defect data of the workpiece under test; Based on the physical data and defect data of the workpiece to be tested, a physical entity model is established to obtain the mapping relationship between microscopic data and macroscopic data; The mapping relationship between the microscopic and macroscopic data is applied to the initial twin model to obtain the target twin model. The initial twin model includes a roughness calculation model and a residual stress calculation model. The target twin model is trained based on historical defect data and current defect data to obtain a prediction model. The prediction model is used to predict defects in the workpiece under test. The prediction model includes a machine learning model and a time series analysis model. The physical data of the workpiece to be tested is input into the prediction model to obtain the output results of the prediction model, which include process adjustment parameters. Based on the process adjustment parameters, the workpiece to be tested is micro-shaped using the ultrafast laser micro-shaping device or the ultrasonic-assisted micro-shaping device to obtain the shaped workpiece.

2. The method as described in claim 1, characterized in that, The output of the prediction model includes predicted defect data. Following the step of performing micro-shaping on the workpiece under test using the ultrafast laser micro-shaping device or the ultrasound-assisted micro-shaping device based on the output of the prediction model to obtain the shaped workpiece, the method further includes: Based on the shaped workpiece under test, the current defect data is reacquired to obtain updated twin data; Calculate the relative error between the predicted defect data and the current defect data; If the relative error is greater than the preset error range, the target twin model is retrained based on the updated twin data until the relative error is less than or equal to the preset error range, and the current prediction model is obtained. The model parameters of the current prediction model are stored in a component library, which is used to store physical data, defect data, processing parameters and their corresponding relationships.

3. The method as described in claim 1, characterized in that, The laser printing defect adjustment system also includes a holographic camera, an X-ray stress detection device, and an industrial control computer. The defect data includes roughness detection data, hole distribution data, and residual stress data of the cladding layer. The steps of acquiring the physical data and defect data of the workpiece to be tested include: The holographic camera acquires physical data and optical information of the cladding layer, the optical information of which includes refraction and scattering information. The refraction and scattering information are processed by an industrial control computer to obtain roughness detection data and hole distribution data; Residual stress data of the cladding layer were obtained using an X-ray stress detection device.

4. The method as described in claim 1, characterized in that, The microscopic data includes atomic interaction potential functions, and the macroscopic data includes the bulk elastic modulus of the cladding layer and the bonding energy at the interface between the cladding layers. The step of establishing a physical entity model based on the physical data and defect data of the workpiece under test to obtain the mapping relationship between the microscopic and macroscopic data includes: Based on the physical and defect data of the workpiece to be tested, the atomic interaction potential function is calculated using a molecular dynamics model. The bulk elastic modulus of the cladding layer is calculated based on the atomic interaction potential function. Based on the bulk elastic modulus of the cladding layer, the bonding energy at the interface between the cladding layers is calculated using the scratch method.

5. The method as described in claim 4, characterized in that, The step of calculating the bonding energy of the interface between the cladding layers based on the bulk elastic modulus of the cladding layer using the scratch method includes: The equivalent elastic modulus is determined based on Poisson's ratio and the bulk elastic modulus of the cladding layer; The bonding energy between the cladding layers is calculated based on the critical load of the cladding layer, the scratch width, and the equivalent elastic modulus.

6. The method as described in claim 1, characterized in that, The microscopic data also includes sampling locations, atomic masses, atomic velocities, atomic volumes, interatomic spacing, and interatomic interactions. The macroscopic data includes the number of sampling points and the number of atoms. The step of applying the mapping relationship between the microscopic and macroscopic data to the initial twin model to obtain the target twin model includes: The relative height and average height of the sampling points are determined based on the sampling points. The roughness of the cladding layer is calculated by combining the number of sampling points and the relative and average height of the sampling point locations; The residual stress of a single atom is calculated based on the atomic mass, atomic velocity, atomic volume, interatomic spacing, and interatomic interaction forces. Based on the number of atoms, the average stress and average biaxial stress of the single-atom layer are calculated according to the residual stress of the individual atom, and the residual stress of the cladding layer is obtained.

7. The method according to any one of claims 1 to 6, characterized in that, The step of training the target twin model based on historical defect data and current defect data to obtain a prediction model, which is used to predict defects in the workpiece under test, includes the following steps: Based on the target twin model, multiple sets of simulated defect data are generated according to historical defect data and current defect data to obtain a sample dataset, which includes point cloud data of the cladding layer surface and time series data of process parameters. The point cloud data of the cladding layer surface is input into a machine learning model for feature extraction to obtain the probability distribution of defect types. The process parameter timing data is input into a timing analysis model for probability prediction to obtain the probability value of defect occurrence.

8. The method as described in claim 7, characterized in that, The step of inputting the point cloud data of the cladding layer surface into a machine learning model for feature extraction to obtain the probability distribution of defect types includes: The point cloud data of the cladding layer surface is input into a convolutional neural network for feature extraction to obtain key features, including edge contours and texture patterns. The key features are input into a graph neural network to obtain a microstructure defect topology graph, which consists of multiple nodes and multiple edges. The nodes are used to represent atomic group properties, and the edges are used to represent the weights of interatomic forces. The node states in the microstructure defect topology graph are updated by a message passing mechanism to obtain the defect type probability distribution.

9. The method as described in claim 7, characterized in that, The step of inputting the process parameter time series data into a time series analysis model for probability prediction to obtain the defect occurrence probability value includes: The time-series data of the process parameters are segmented by a sliding time window to extract the residual stress sequence and surface roughness sequence of the cladding layer. The residual stress sequence and surface roughness sequence of the cladding layer are input into a long short-term memory network, and the probability value of defect occurrence is calculated through a gating mechanism.

10. A laser printing defect adjustment system, characterized in that, The laser printing defect adjustment system includes an industrial control computer, a holographic camera, an X-ray stress detection device, a workpiece to be tested, an ultrafast laser micro-shaping device, and an ultrasonic-assisted micro-shaping device. The holographic camera is connected to the industrial control computer via a spectrometer to acquire physical data and optical information of the cladding layer; The X-ray stress detection device includes an X-ray generator, an X-ray detector, and an X-ray measurement recorder. The X-ray generator is used to emit X-rays to the surface area of ​​the cladding layer. The X-ray detector is used to detect the surface area of ​​the cladding layer and obtain information on the changes in the positions of the X-ray diffraction peaks. The X-ray measurement recorder is used to store the information on the changes in the positions of the X-ray diffraction peaks. The ultrafast laser micro-plastic surgery device includes an ultrafast laser generator, an ultrafast laser optical path system, and an ultrafast laser probe. The ultrafast laser generator is used to generate a laser pulse sequence, the ultrafast laser optical path system is used to guide the laser to the target area on the surface of the cladding layer, and the ultrafast laser probe is used to locate the laser spot. The ultrasonic-assisted micro-shaping device includes an ultrasonic vibrating head, a transducer, an ultrasonic transmitter, and an ultrasonic amplitude rod. The ultrasonic vibrating head is fixedly connected to the substrate of the workpiece to be tested. The ultrasonic transmitter is used to generate an electrical signal. The transducer is used to convert the electrical signal into mechanical vibration. The ultrasonic amplitude rod is used to transmit ultrasonic waves to the target area on the surface of the cladding layer. The industrial control computer is used to receive and process the physical data, the optical information of the cladding layer, and the information on the changes in the position of the X-ray diffraction peaks. The industrial control computer is also used to control the ultrafast laser micro-shaping device or the ultrasonic-assisted micro-shaping device to perform micro-shaping on the workpiece to be tested according to the process adjustment parameters as described in any one of claims 1 to 9.

Images