Laser cladding parameter adaptive control method and system based on AI digital twinning

Abstract

The invention discloses a laser cladding parameter self-adaptive control method and system based on AI digital twinning, and aims to solve the problems that in the prior art, laser cladding process parameter control depends on artificial experience adjustment and is difficult to adapt to material characteristic fluctuation, complex path heat accumulation and the like, and the final forming quality is affected. The invention provides a laser cladding parameter autonomous optimization method fusing digital twinning and Bayesian optimization, and aims to realize adaptive control of laser cladding parameters in a cladding process through Gaussian process modeling and real-time data iteration. According to the method, a digital twin system comprising a physical layer, a digital twin layer and a communication layer is constructed. Real-time data interaction and control instruction issuing are carried out between the physical system and the digital twin system through industrial communication protocols such as OPC UA, and closed-loop control is formed. The method can remarkably improve the quality and consistency of the formed part, reduces manual intervention, shortens the process development period, and has wide application prospects.

Description

Technical Field

[0001] The present invention relates to the field of laser additive manufacturing technology, and more particularly to a method and system for controlling parameters during the laser cladding process. More specifically, the present invention relates to a method and system for adaptively controlling laser cladding parameters based on artificial intelligence and digital twin technology. The method aims to improve the stability of the laser cladding process and the quality of formed parts, and is suitable for precision manufacturing scenarios such as 3D printing, surface strengthening, and damage repair of metal parts. Background Art

[0002] Laser cladding technology, as an advanced additive manufacturing and surface modification technology, uses a high-energy laser beam to melt the alloy powder or wire fed synchronously, causing it to rapidly solidify on the surface of the substrate to form a metallurgically bonded coating. It is widely used in the repair, strengthening and direct manufacturing of key components in the fields of aerospace, energy and power, mold manufacturing, petrochemicals, etc. In the laser cladding process, cladding parameters (such as laser power, cladding speed, wire / powder feeding speed, etc.) are the most core process parameters that determine the size of the molten pool, temperature distribution, melting depth, dilution rate, and the final coating structure and performance. Inappropriate laser cladding parameters can lead to a variety of defects such as lack of fusion, pores, cracks, overburning, coarse structure, uneven performance, etc., which seriously affect product quality and service life. At present, the parameter control method in the laser cladding process has the following main deficiencies:

[0003] Open-loop control or trial-and-error: In many production scenarios, laser cladding parameter settings still rely on operator experience and extensive process testing. This approach is inefficient and costly, and it can be difficult to ensure consistent cladding quality. This often requires lengthy debugging cycles, especially when working with new materials, new structures, or complex working conditions.

[0004] Traditional PID (Proportional-Integral-Derivative) closed-loop control: While the introduction of PID closed-loop control based on a single or a few feedback signals, such as the melt pool temperature, can improve the stability of cladding quality to a certain extent, the tuning of the PID controller parameters (Kp, Ki, Kd) is inherently challenging and often requires repeated manual adjustments. More importantly, laser cladding is a highly nonlinear, multivariable, dynamic process, influenced by a variety of factors, including temperature-dependent changes in the material's thermophysical properties, fluctuations in the powder feed rate, heat accumulation due to scanning path complexity, uneven substrate preheating, and protective atmosphere disturbances. The linear characteristics of traditional PID controllers make it difficult to effectively handle these complex dynamic changes and strongly nonlinear disturbances. Consequently, they have poor adaptability and often fail to achieve optimal control results.

[0005] Model Predictive Control (MPC): While advanced control strategies like MPC can theoretically provide superior control performance, their effectiveness is highly dependent on accurate physical process models. The laser cladding process involves complex multi-physics coupling phenomena, including laser-matter interaction, heat conduction, convection, radiation, melting and solidification phase transitions, melt pool fluid dynamics, powder particle flight and melting behavior within the laser beam, and plasma generation and decay. Establishing such a high-fidelity physical model is extremely difficult and computationally intensive, making it difficult to meet the real-time requirements of industrial applications. Furthermore, the model's accuracy is susceptible to unmodeled dynamics and parameter uncertainty.

[0006] At the same time, digital twin technology, as a bridge connecting the physical and cyber worlds, has shown tremendous potential in modern manufacturing. However, current applications of digital twins in laser cladding are limited to condition monitoring, process visualization, offline process planning, or predictive maintenance. These applications lack deep integration with real-time control loops and the ability to empower the system to make autonomous optimization decisions.

[0007] Bayesian optimization, as an efficient global optimization algorithm, is particularly well-suited for "black-box" optimization problems where objective function evaluation is expensive (such as in physical experiments or complex simulations), there is noise, or gradient information is difficult to obtain. It constructs a probabilistic surrogate model of the objective function (typically a Gaussian process regression model) and uses an acquisition function to balance exploration (sampling in areas of high uncertainty) and exploitation (sampling in areas of high expected return), thereby rapidly approaching the optimal solution with a small number of sample points.

[0008] Therefore, given the shortcomings of existing laser cladding parameter control methods in adapting to dynamic changes, achieving optimal control, and reducing manual intervention, a new intelligent control paradigm is urgently needed that can perceive the cladding process status in real time, autonomously learn and optimize laser cladding parameters, and adapt to complex working conditions. Combining digital twin technology with artificial intelligence methods such as Bayesian optimization provides a new approach to addressing this challenge. Summary of the Invention

[0009] Technical issues to be solved:

[0010] The main purpose of the present invention is to overcome the problems existing in existing laser cladding parameter control methods, such as poor adaptability to dynamic process changes, excessive reliance on manual experience, difficulty in achieving optimal process control, and the complexity and poor real-time performance of model predictive control methods. Specifically, the present invention aims to provide a laser cladding parameter adaptive control method and system based on AI digital twins. This method and system can perceive the state of the cladding process in real time, autonomously learn and optimize the laser cladding parameters in a data-driven manner, thereby improving the stability of the laser cladding process, enhancing the quality of the cladding layer, reducing manual intervention, and shortening the development cycle of new materials and new processes.

[0011] Technical solution:

[0012] To achieve the above objectives, the present invention provides a method for adaptive control of laser cladding parameters based on AI digital twins. The core of this method is to build a digital twin system that integrates physical entities, virtual models, and intelligent optimization algorithms, and to achieve closed-loop adaptive parameter control through real-time data interaction. The method includes the following main steps:

[0013] S1: Construct a digital twin system for the laser cladding process. This digital twin system is logically divided into a physical layer, a digital twin layer, and a communication layer. The physical layer is the actual execution unit of the laser cladding process and primarily includes laser cladding equipment (such as high-power lasers, multi-axis motion mechanisms, synchronous powder / wire feeding systems, protective atmosphere devices, cooling systems, etc.) and a melt pool online monitoring module (such as a coaxial infrared thermometer or multi-band pyrometer for collecting melt pool temperature, a high-speed CMOS camera or OTC laser melt pool monitor for capturing melt pool morphology, size, spatter, and plasma state, and a laser displacement sensor or structured light scanner for measuring the height or width of the cladding layer). The digital twin layer is the virtual mapping and intelligent decision-making core of the physical layer and primarily includes a high-fidelity virtual melt pool dynamics model and a Bayesian optimization engine. The communication layer is responsible for establishing a stable, high-speed, bidirectional data transmission channel between the physical layer and the digital twin layer, for example, using industrial Ethernet communication protocols such as OPC UA, EtherCAT, Profinet, or message queue technologies such as MQTT and Kafka, or wireless communication technologies such as 5G.

[0014] S2: The physical layer's laser cladding equipment performs the laser cladding operation according to preset initial parameters or parameters optimized in the previous round. During this process, the physical layer's melt pool online monitoring module synchronously collects process parameters (such as actual laser power, scanning speed, and powder feed rate) and melt pool status information (such as melt pool surface temperature distribution, melt pool geometry, melt pool dynamic behavior characteristics, plasma spectrum characteristics, and cladding layer surface morphology) in real time.

[0015] S3: The communication layer transmits various process parameters and molten pool status information collected by the physical layer to the digital twin layer in real time or quasi-real time after necessary preprocessing (such as filtering, noise reduction, and feature extraction).

[0016] S4: The Bayesian optimization engine of the digital twin layer receives real-time data from the physical layer and, combined with the simulation prediction data of the virtual melt pool dynamics model, models the complex nonlinear relationship between laser cladding parameters and cladding quality. Specifically, Gaussian process regression (GPR) is used as a proxy model. GPR can provide the predicted mean and uncertainty (variance) of the cladding quality under any untested laser cladding parameters based on existing (laser cladding parameters, corresponding to cladding quality evaluation) data points. The cladding quality can be a comprehensive evaluation of one or more key performance indicators, such as the average and stability (such as variance or fluctuation range) of the melt pool temperature, the geometric accuracy (such as width, height, flatness) of the cladding layer, the surface roughness, and the internal defect rate (such as porosity, crack sensitivity), etc. These indicators can be directly calculated or indirectly inferred from real-time monitoring data.

[0017] S5: Based on the currently updated Gaussian process regression surrogate model, the Bayesian optimization engine uses a preset acquisition function to calculate and recommend the next set of optimal or most promising laser cladding parameters. Common acquisition functions include expected improvement (EI), probability improvement (PI), or upper confidence limit (UCB). These acquisition functions effectively balance exploring the unknown parameter space (i.e., exploring regions of cladding parameters with high uncertainty to discover potentially better solutions) and leveraging known information (i.e., conducting a detailed search near the optimal parameter region predicted by the current model). The calculated recommended laser cladding parameters are then transmitted via the communication layer to the laser cladding equipment controller at the physical layer for adjustment of its actual process parameters.

[0018] S6: The system repeats steps S2 to S5, forming a closed-loop iterative optimization process. In each iteration, new experimental data (actually applied laser cladding parameters and the resulting cladding quality) is used to further update and optimize the Gaussian process regression proxy model, making its description of the real physical process increasingly accurate. This guides the continuous optimization of laser cladding parameters toward the goal of improving cladding quality until the preset cladding quality requirement is achieved, the convergence condition is satisfied, or the maximum number of iterations is reached, ultimately achieving adaptive control of laser cladding parameters.

[0019] Furthermore, the present invention also provides a laser cladding parameter adaptive control system based on AI digital twin, which is used to implement the above method and is characterized by comprising:

[0020] Physical layer: This layer includes the laser cladding equipment that performs the actual laser cladding task, and the molten pool online monitoring module that is configured on or near the laser cladding equipment and is used to collect various process parameters and molten pool status information in real time during the cladding process.

[0021] Digital Twin Layer: This layer serves as the "brain" of the system, encompassing a virtual melt pool dynamics model used to simulate and predict melt pool behavior during the laser cladding process (e.g., a multi-physics coupling model based on finite element analysis, computational fluid dynamics, etc., capable of predicting the melt pool temperature field, flow field, and solidification structure based on input process parameters). This layer also includes a core Bayesian optimization engine. This Bayesian optimization engine receives real-time monitoring data from the physical layer and simulation data from the virtual melt pool dynamics model. Using this data, it uses Gaussian process regression to establish and continuously update a proxy model between laser cladding parameters and cladding quality. Based on this proxy model, it uses an acquisition function to calculate and output a recommended next set of laser cladding parameters.

[0022] Communication layer: This layer serves as a bridge between the physical layer and the digital twin layer, responsible for real-time, reliable data exchange between the two (including uploading monitoring data and issuing control commands). For example, industrial Ethernet protocols such as OPC UA, EtherCAT, and Profinet can be used, or combined with message queue technologies such as MQTT and Kafka to ensure low latency and high throughput for data transmission.

[0023] Control module: This module receives the recommended laser cladding parameters output by the Bayesian optimization engine of the digital twin layer, and accurately adjusts and controls the parameters of the laser cladding equipment in the physical layer based on the parameters.

[0024] In some preferred embodiments, the melt pool online monitoring module may include, but is not limited to: a coaxial infrared thermometer or multi-wavelength pyrometer for non-contact measurement of the melt pool surface temperature; a high-speed camera (which can be used with a specific wavelength filter or illumination light source) for capturing the dynamic morphology, size, intensity, and form of the melt pool's spatter; an OTC laser melt pool monitor, laser displacement sensor, line laser scanner, or structured light 3D profile measurement system for online measurement of the cladding layer's height, width, or surface profile; and a spectrum analyzer for analyzing plasma composition and state. The multimodal data acquired by these sensors can be fused and processed for a more comprehensive and accurate assessment of cladding quality.

[0025] In some preferred embodiments, the virtual melt pool dynamics model can not only be used to provide additional training data to accelerate the Bayesian optimization process (especially when the experimental cost is high or the initial data is sparse), but can also receive measured data from the physical layer online, and calibrate and correct its own model parameters through parameter identification, state estimation and other methods, thereby improving the model's fidelity and prediction accuracy, and achieving continuous synchronization and co-evolution of the physical model and the actual process.

[0026] In some preferred embodiments, the Bayesian optimization engine can also consider process constraints (such as the upper and lower limits of laser power, the molten pool temperature cannot exceed the vaporization point of the material, etc.) and safety constraints (such as avoiding excessive thermal stress leading to cracking, etc.) when performing optimization, by introducing penalty terms in the acquisition function or using a constrained optimization algorithm.

[0027] Beneficial effects:

[0028] Compared with the existing technology, the laser cladding parameter adaptive control method and system based on AI digital twin proposed in this invention has the following significant beneficial effects:

[0029] Highly adaptable and robust: By building a digital twin system and introducing a real-time data-driven Bayesian optimization mechanism, the present invention can perceive various dynamic changes in the laser cladding process in real time, such as material batch differences, uneven substrate preheating, heat accumulation effects caused by complex geometric paths, and slight fluctuations in powder feed rate. It can then adaptively adjust the laser cladding parameters, effectively suppressing the impact of these disturbances on cladding quality, maintaining process stability and consistent final product quality. This significantly outperforms traditional PID control in its ability to handle dynamic disturbances.

[0030] Significantly improve optimization efficiency and reduce costs: The Bayesian optimization algorithm, renowned for its sample efficiency, leverages a surrogate model established through Gaussian process regression to guide subsequent parameter selection. This allows for rapid convergence to near-optimal laser cladding parameter settings within a relatively small number of physical experiments or costly simulation iterations. This significantly reduces the time and associated experimental costs required to explore parameter windows for new materials and processes, accelerating product development and process optimization.

[0031] Effectively improve the quality of the cladding layer: By comprehensively considering and closed-loop optimizing multi-dimensional quality indicators such as the molten pool temperature, size, morphology, stability, and the geometric accuracy, surface quality, and internal defects of the final cladding layer, the present invention can more accurately control the thermal behavior and solidification process of the molten pool, thereby effectively suppressing the generation of common metallurgical defects such as pores, cracks, unfusion, and slag inclusions, and improving the density, uniformity, mechanical properties, and service reliability of the cladding layer.

[0032] Reducing reliance on human experience and enabling intelligent decision-making: This invention transforms the laser cladding parameter optimization process from relying on operator experience and tedious trial and error to a data-driven, autonomous learning and intelligent decision-making process based on artificial intelligence algorithms. This not only reduces reliance on highly skilled operators, making process parameter setting more scientific and standardized, but also lays the foundation for achieving less- or even unmanned intelligent manufacturing in the laser cladding process.

[0033] Enhanced Digital Twin Capabilities and Process Understanding: This invention goes beyond simply using digital twins for monitoring or simple control, integrating them deeply into an intelligent optimization closed loop. The high-fidelity virtual melt pool model not only aids the optimization process, but its continuous synchronization and mutual calibration with the physical entity also helps deepen understanding of the complex physical mechanisms of laser cladding, accumulate process knowledge, and provide a powerful digital platform for more advanced process planning, fault diagnosis, and predictive maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0035] Figure 1 It is a schematic diagram of the overall architecture of a laser cladding parameter adaptive control system based on AI digital twin proposed according to one embodiment of the present invention.

[0036] Figure 2 4 is a flow chart of a method for adaptively controlling laser cladding parameters according to an embodiment of the present invention.

[0037] Figure 3 Schematic diagram of the working principle of laser cladding parameter optimization using a Bayesian optimization engine according to an embodiment of the present invention. DETAILED DESCRIPTION

[0038] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0039] Example 1

[0040] This embodiment provides a method and system for adaptive control of laser cladding parameters based on AI digital twins, and illustrates this by taking the laser cladding of thin-walled structures using 316L stainless steel powder on a Q235 steel substrate as an example.

[0041] Reference Figure 1 The AI ​​digital twin-based laser cladding parameter adaptive control system 100 of this embodiment primarily includes a physical layer 10, a digital twin layer 20, a communication layer 30, and a control module 40. The physical layer 10 and the digital twin layer 20 exchange data in real time via the communication layer 30, and the control module 40 (which can be integrated into the digital twin layer 20 or function as a standalone unit) adjusts the parameters of the laser cladding equipment 11 in the physical layer 10 based on the optimization results of the digital twin layer 20.

[0042] The physical layer 10 includes:

[0043] Laser cladding equipment 11: This example uses a YLS-2000 fiber laser (rated power 2000W, wavelength 1070nm), equipped with a collimating and focusing unit and a scanning galvanometer. The laser spot diameter is set to 2mm. The laser is integrated with a six-axis industrial robot (such as the KUKA KR 30) to achieve precise movement along complex paths. The powder feeding system uses a GTV PF2 / 2 dual-drum synchronous powder feeder, using argon as the feed gas at a feed rate of 15g / min. The substrate is a 100mm x 100mm x 10mm Q235 steel plate, which is sandblasted and cleaned with alcohol before cladding. The entire cladding process is performed in an argon-filled protective atmosphere chamber to prevent oxidation.

[0044] Melt Pool Monitoring Module 12: A coaxial infrared thermometer or pyrometer, such as the IMPAC IGA 6 Advanced pyrometer, has a measurement range of 300°C to 2500°C and monitors the temperature in the center of the melt pool in real time. A high-speed camera, such as the Photron FASTCAM Mini AX200, equipped with a narrowband filter (corresponding to the laser wavelength) is used to capture the dynamic morphology, size (melt width and length), and spatter intensity and form of the melt pool. The image acquisition frame rate is set to 1000 fps. A laser displacement sensor, such as the KEYENCE LK-G5000 series, is used to measure the height and width of a single cladding layer or the consistency of layer height after multi-layer deposition.

[0045] The digital twin layer 20 includes:

[0046] Virtual Melt Pool Dynamics Model 21: Built using COMSOL Multiphysics software and solved using the finite element method (FEM), this model couples heat conduction, fluid flow (accounting for the Marangoni effect, buoyancy, and electromagnetic forces), solidification and melting phase transitions, and energy absorption by the laser, powder, and melt pool (using a combination of a volumetric heat source model and a surface heat flow model). Model inputs include laser power, scanning speed, powder feed rate, material thermophysical properties (temperature-dependent), and ambient temperature. Model outputs include the melt pool's temperature field, velocity field, geometry, and solidification microstructure characteristics (such as cooling rate and temperature gradient). The model is pre-calibrated using a small amount of calibration experimental data to improve prediction accuracy.

[0047] Bayesian optimization engine 22: implemented in Python and can utilize open source libraries such as GPyOpt and BoTorch. Core functions include: Gaussian process regression (GPR) modeling unit: receives real-time monitoring data from the physical layer 10 (such as the average temperature of the melt pool, temperature standard deviation, melt width, estimated melt depth, defect indication characteristics, etc.) and corresponding laser cladding parameter values, or receives simulation data from the virtual melt pool dynamics model 21, and establishes a proxy model Q = GPR(P) between the laser cladding parameters P and the comprehensive cladding quality score Q. In this embodiment, the GPR kernel function uses the Matern 5 / 2 kernel to take into account the smoothness of the function and its sensitivity to local changes. Cladding quality assessment unit: calculates the comprehensive cladding quality score Q based on real-time monitoring data or simulation data. For example, Q can be defined as: Q = w1 * (1 - |T_avg - T_target| / T_target) + w2 * (1 - T_std / T_target) + w3 * (1 - |W_actual - W_target| / W_target) - w4 * N_defect. Here, T_avg is the average melt pool temperature, T_target is the target melt pool temperature, T_std is the standard deviation of the melt pool temperature, W_actual is the actual weld width, W_target is the target weld width, and N_defect is a defect indicator (such as the number of incomplete fusion or pores based on image analysis). w1, w2, w3, and w4 are weighting coefficients for each factor, which are set according to specific application requirements. The acquisition function calculation unit uses the expected improvement (EI) acquisition function to recommend the next most promising laser cladding parameter, P_next, for testing based on the mean and variance predicted by the GPR model. The EI acquisition function formula is: EI(P) = (μ(P) -Q_best - ξ) * Φ((μ(P) - Q_best - ξ) / σ(P)) + σ(P) * φ((μ(P) - Q_best - ξ) / σ(P)). Where μ(P) and σ(P) are the mean and standard deviation of the quality score predicted by the GPR model for parameter P, Q_best is the best quality score currently observed, ξ is a small positive constant (for example, 0.01) used to balance exploration and exploitation, and Φ(·) and φ(·) are the cumulative distribution function and probability density function of the standard normal distribution, respectively.

[0048] Real-time twin synchronization engine 23 (optional): responsible for synchronizing the real-time state data of the physical layer 10 with the state of the virtual melt pool dynamics model 21, and can calibrate the model parameters online according to the deviation.

[0049] Data processing and storage unit 24: responsible for cleaning, filtering, feature extraction, and fusion of the collected raw data, and storing historical data and optimization process data in the database.

[0050] Visualization and Interaction Unit 25: Provides a user interface to display the operating status of the physical system, virtual model simulation results, and the Bayesian optimization process (such as the agent model surface, acquisition function curve, historical sampling points, quality score evolution trend, etc.), and allows users to set optimization goals, constraints, etc.

[0051] Communication layer 30: In this embodiment, the OPC UA protocol is used to implement data exchange between the sensors and actuators of the physical layer 10 and the modules of the digital twin layer 20. OPC UA offers advantages such as platform independence, security, and strong information modeling capabilities, making it suitable for industrial automation applications.

[0052] The control module 40 receives the recommended parameter P_next output by the Bayesian optimization engine 22 and adjusts the actual parameters in the laser cladding equipment 11 through the cladding equipment control interface (such as an analog voltage signal or a digital communication interface).

[0053] Reference Figure 2 The specific process of the laser cladding parameter adaptive control method of this embodiment is as follows:

[0054] Step S201: Initialization. Set the search range for laser cladding parameters (e.g., laser power 800W to 1500W, scanning speed 6-10mm / s, powder feed rate 10-15g / min, etc.). Perform a small initial sampling, such as using Latin Hypercube Sampling (LHS), to select N_init (e.g., 10-20) initial parameter points within the parameter range. For each initial parameter point, perform a single-pass or short-distance multi-pass cladding experiment at the physical layer 10. The corresponding molten pool state information is collected by the molten pool monitoring module 12, and an initial data pair (cladding parameter, cladding quality score) is calculated. This initial data is used for the initial training of the GPR model.

[0055] Step S202: Physical layer execution and data acquisition. The laser cladding equipment 11 performs the cladding operation based on the currently set parameters (initial sampling points in the initial stage, and parameter points recommended by Bayesian optimization in the subsequent stage). The molten pool monitoring module 12 collects data such as the molten pool temperature and morphology in real time.

[0056] Step S203: Data Transmission and Preprocessing. The collected raw data is transmitted via the communication layer 30 to the data processing and storage unit 24 of the digital twin layer 20. Data cleaning, filtering, and feature extraction are performed (for example, calculation of average melt pool temperature, temperature standard deviation, weld width, weld depth indication, spatter quantity, plasma intensity, etc.).

[0057] Step S204: Cladding quality assessment: Calculate the comprehensive cladding quality score Q under the current parameters based on the pre-processed feature data.

[0058] Step S205: GPR proxy model update. The new (cladding parameter P, cladding quality score Q) data points are added to the existing dataset. The GPR modeling unit in the Bayesian optimization engine 22 uses the updated dataset to retrain the GPR model, obtaining an updated cladding parameter-cladding quality proxy model.

[0059] Step S206: Acquisition function calculation and parameter recommendation. Figure 3 , the Bayesian optimization engine 22 is based on the updated GPR model ( Figure 3 (a) is the GPR prediction mean, (b) is the GPR prediction standard deviation), using the EI acquisition function ( Figure 3 (c) shows the EI function curve. Within the set cladding parameter search space, the parameter point P_next that maximizes the EI value is searched. This point is the recommended cladding parameter for the next round of experiments.

[0060] Step S207: Parameter adjustment: The control module 40 sends the recommended cladding parameters P_next to the laser cladding equipment 11 to adjust the cladding parameters.

[0061] Step S208: Determine the termination condition. Check whether the preset maximum number of iterations (e.g., 50-100) has been reached, whether the cladding quality score has converged (e.g., the improvement after M consecutive iterations is less than a threshold ε), or whether a cladding quality that meets the requirements has been found. If the termination condition is met, the optimization process ends and the currently optimal laser cladding parameters and their corresponding cladding quality are output. Otherwise, the process returns to step S202 and continues with the next round of iterative optimization.

[0062] Through the above iterative process, the system can automatically explore and converge to the laser cladding parameters that optimize the cladding quality, effectively adapting to the dynamic changes under different working conditions.

[0063] Experimental Results and Comparisons: To verify the beneficial effects of the present invention, comparative experiments can be conducted. For example, under the same cladding task (e.g., single-pass cladding, thin-walled workpiece accumulation), laser cladding can be performed using fixed empirical parameters, traditional PID control (based on melt pool average temperature feedback), and the adaptive control method based on AI digital twins proposed in this invention. After the experiments, the cladding layer's macromorphology (e.g., width consistency, height uniformity, surface flatness), microstructure, hardness, and internal defects (e.g., pores and cracks, which can be detected through metallographic analysis or nondestructive testing) can be characterized and compared.

[0064] Expected Results: Compared with fixed parameters and traditional PID control, the cladding layers produced using this method demonstrate significant improvements in dimensional accuracy, surface quality, microstructure uniformity, and defect control. For example, during the buildup of thin-walled parts, this method adaptively adjusts laser cladding parameters based on the increasing number of layers and changes in heat accumulation, effectively avoiding problems such as overheating and collapse of the bottom layer or lack of fusion of the top layer, resulting in more regular parts with superior internal quality. Furthermore, the total number of experiments required to achieve stable, high-quality cladding (including iterations during the optimization process) is far fewer than with traditional trial-and-error methods.

[0065] Those skilled in the art will appreciate that the above embodiments are merely illustrative, and that various modifications and variations are possible without departing from the spirit and scope of the present invention. For example, the molten pool monitoring module can employ a wider variety of sensors for multimodal information fusion; the virtual molten pool dynamics model can employ other numerical calculation methods or a machine learning-based proxy model; the Bayesian optimization engine can employ different kernel functions, acquisition functions, or incorporate other optimization algorithms; and the communication protocol and hardware platform can also be selected based on actual needs. These modifications and variations are intended to fall within the scope of protection of the present invention.

Claims

1. A laser cladding parameter adaptive control method based on AI digital twin, characterized in that: The following steps are involved: S1: Constructing a digital twin system for the laser cladding process, the digital twin system includes a physical layer, a digital twin layer, and a communication layer; the physical layer includes the laser cladding equipment and a molten pool monitoring module; the digital twin layer includes a virtual molten pool dynamics model and a Bayesian optimization engine; S2: The laser cladding equipment of the physical layer performs a cladding operation, and the molten pool monitoring module collects process parameters and molten pool status information in real time during the cladding process; S3: The communication layer transmits the collected process parameters and molten pool status information to the digital twin layer in real time; S4: The Bayesian optimization engine of the digital twin layer uses Gaussian process regression to establish a proxy model between laser cladding parameters and cladding quality based on the received real-time information and / or the simulation data of the virtual molten pool dynamics model; S5: The Bayesian optimization engine calculates a next set of recommended laser cladding parameters through an acquisition function, and adjusts parameters of the laser cladding equipment based on the recommended laser cladding parameters; S6: Repeat steps S2 to S5 to achieve adaptive control of laser cladding parameters.

2. The method according to claim 1, characterized in that The molten pool monitoring module includes part or one of a coaxial infrared thermometer, a high-speed camera, a laser displacement sensor, and an OTC laser molten pool monitor, and is used to collect information on the molten pool temperature, molten pool morphology, and cladding layer thickness.

3. The method according to claim 1, characterized in that The virtual molten pool dynamics model is established based on the finite element method or the finite difference method, and is used to simulate the temperature field, flow field and solidification behavior of the molten pool under different laser cladding parameters.

4. The method according to claim 1, wherein In the Bayesian optimization engine, the evaluation index of the cladding quality includes one or more of molten pool stability, layer thickness uniformity, penetration depth, molten pool size, and defect rate.

5. The method according to claim 1, wherein The acquisition function is one of expected improvement (EI), probability improvement (PI) or upper confidence limit (UCB).

6. The system according to claim 1, wherein: The communication layer uses one or more of OPC UA, MQTT, DDS, and 5G for data interaction.

Images