Method for predicting laser cladding deformation and stress of non-rigid substrate based on machine learning

Abstract

The invention discloses a method for predicting laser cladding deformation and stress of a non-rigid substrate based on machine learning, and belongs to the technical field of optimization of a laser additive manufacturing process. The method comprises the following steps: constructing a thermal-mechanical coupling finite element model, simulating a laser cladding process, and establishing a mapping database of process parameters, a deformation field and a stress field; standardizing the data, dividing the data into a training set and a verification set, training a plurality of machine learning algorithms, and preferentially selecting an agent model; and the reliability and the precision of the model are ensured by verifying and optimizing the model in a closed-loop manner through actual experiments. The core innovation of the invention lies in that the controllable deformation of the non-rigid substrate is actively used as a stress release mechanism to inhibit the generation of cracks. The method further supports reverse process design with the target deformation and the target stress value as input, and optimization of laser cladding process parameters is achieved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of laser additive manufacturing and surface engineering, in particular to a method for predicting and optimizing the deformation behavior and residual stress distribution of a non-rigid substrate during laser cladding. Specifically, the present application combines thermal-mechanical coupling finite element simulation, machine learning and experimental verification to achieve rapid and accurate design and optimization of laser cladding process parameters. BACKGROUND

[0002] As an advanced surface modification and remanufacturing technology, laser cladding is widely used in aerospace, automotive manufacturing and other fields. However, during the cladding process, due to the rapid thermal interaction between the laser and the material, significant thermal stress and non-uniform plastic deformation occur, leading to macroscopic deformation such as warping and twisting of non-rigid substrates (such as thin-walled or complex structures), and the formation of harmful residual stresses inside the coating and substrate, which can even cause cracks, severely restricting the application of this technology in precision components.

[0003] Currently, the prediction of stress and deformation during laser cladding mainly relies on thermal-mechanical coupling finite element analysis. Although this method can reveal the process evolution from a physical mechanism, it is complex to model, consumes a lot of computing resources, and takes a long time to simulate each time, making it difficult to be used for large-scale process parameter screening and optimization. Essentially, it still belongs to the "trial and error" mode, which is inefficient.

[0004] In recent years, machine learning technology has provided a new approach to solving complex nonlinear engineering problems. In existing technologies, there are methods that use machine learning to predict the geometric morphology of laser cladding coatings, dilution rates or stress values at specific points. However, these methods are mostly limited to rigid substrates or single output indicator prediction. For the complex stress redistribution behavior of non-rigid substrates during laser cladding due to large deformation, existing technologies lack effective prediction means. In addition, moderate deformation of non-rigid substrates can effectively release some thermal stress and reduce the risk of cracking, but how to quantify and actively utilize this "deformation-stress release" effect to achieve a balance between "controllable deformation" and "stress suppression" is still a technical gap.

[0005] Therefore, there is an urgent need for a method that can quickly and accurately predict the overall deformation field and stress field of a non-rigid substrate after laser cladding, to guide process design, actively utilize substrate deformation to release stress and suppress cracks, and improve process reliability and yield while ensuring bonding quality. SUMMARY

[0006] (1) Technical problems to be solved

[0007] In order to overcome the industry problem that the high residual stress in the traditional laser cladding technology leads to the cracking of the cladding layer and the difficulty in cooperative prediction and control of the macroscopic deformation of the non-rigid substrate, the present application provides a method for predicting the laser cladding deformation and stress of a non-rigid substrate based on machine learning.

[0008] (II) Technical solution

[0009] In order to achieve the above purpose, the present application adopts the following technical solution:

[0010] 1. A method for predicting the laser cladding deformation and stress of a non-rigid substrate based on machine learning

[0011] The core of the method of the present application is to actively use the controllable deformation of the non-rigid substrate as a stress release mechanism to suppress the cracking of the cladding layer, and includes the following steps:

[0012] S1, constructing a thermal-mechanical coupling finite element simulation model for a non-rigid substrate:

[0013] Based on the process parameters of laser cladding, the thermal physical property parameters and geometric size parameters of the substrate and the coating material, a three-dimensional transient thermal-mechanical coupling finite element model is established. The model simulates the input of laser energy by a Gaussian body heat source, and considers the nonlinear change of material performance with temperature, and is used to calculate the temperature field, deformation field and residual stress field in the laser cladding process.

[0014] S2, establishing a process parameter-deformation / stress database:

[0015] Parameterized scanning is performed using the simulation model to generate a simulation data set. The input parameters include process parameters such as laser power, scanning speed, powder feeding rate, material parameters such as density, specific heat capacity, thermal conductivity, thermal expansion coefficient, elastic modulus, Poisson's ratio, and geometric size parameters of the substrate and the coating. The output parameters are the deformation field and residual stress field of the substrate. In order to cover a comprehensive parameter space with high efficiency, the orthogonal experimental design method is used to determine the multiple input parameter combinations of the parameterized scanning simulation.

[0016] S3, data preprocessing and division:

[0017] The simulation data set is subjected to data standardization or normalization processing to eliminate the dimensional influence of different parameters and improve the convergence and efficiency of subsequent model training. Subsequently, the preprocessed data is divided into a training set and a validation set.

[0018] S4, training of machine learning prediction model:

[0019] Choose at least one machine learning algorithm, such as linear regression, decision tree regression, random forest regression, support vector machine regression, or artificial neural network, and use the training set for supervised learning training to establish a nonlinear mapping model from the input parameters to the output parameters, i.e., a machine learning prediction model.

[0020] S5. Model Evaluation and Selection:

[0021] Through performance indicators (such as the coefficient of determination R) 2 The root mean square error (RMSE) and mean absolute error (MAE) are used to evaluate and compare multiple trained models, and the model with the highest prediction accuracy is selected as the surrogate model.

[0022] S6. Experimental Verification and Closed-Loop Optimization:

[0023] Actual laser cladding experiments were conducted to obtain measured input parameters and corresponding deformation and stress output parameters. The measured input parameters were then input into the optimal surrogate model, and its prediction results were compared with the measured results. If the prediction accuracy did not meet the preset requirements (e.g., mean absolute error (MAE) higher than 5%), the experimental data was added to the original training set, and the model training and optimization steps were re-executed (returning to S4), forming a closed-loop feedback optimization mechanism of "simulation-machine learning-experiment," ultimately obtaining an experimentally verified prediction model with high accuracy and strong generalization ability.

[0024] 2. A reverse design system for laser cladding process parameters

[0025] This invention also provides a reverse design system for laser cladding process parameters. This system uses the experimentally verified prediction model described above to achieve rapid and accurate design of process parameters. The system's operation flow includes:

[0026] (1) Use the target deformation amount (such as the maximum warping amount) and / or the target residual stress value set by the user as the optimization target;

[0027] (2) The laser process parameters are iteratively searched by optimization algorithms (such as genetic algorithms, particle swarm optimization, etc.) so that the difference between the deformation field and residual stress field predicted by the prediction model for the searched parameters and the optimization target is minimized.

[0028] (3) Finally, the optimal combination of laser process parameters that can meet the preset performance requirements is output.

[0029] (III) Beneficial Effects

[0030] Compared with existing technologies, this invention achieves rapid and accurate prediction of deformation and stress in laser cladding of non-rigid substrates through a machine learning proxy model, which greatly improves efficiency and reduces costs; the closed-loop optimization mechanism based on simulation and experimental verification effectively ensures the prediction accuracy and generalization ability of the model; and the reverse design system supports process parameter optimization oriented towards target performance, providing theoretical guidance for the optimization of laser cladding process parameters. Attached Figure Description

[0031] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0032] Figure 1 This is a flowchart of a method for predicting laser cladding deformation and stress of non-rigid substrates based on machine learning, according to an embodiment of the present invention.

[0033] Figure 2 A flowchart for reverse engineering a genetic algorithm. Detailed Implementation

[0034] The present invention will be further described below with reference to a specific embodiment and the accompanying drawings, but the scope of protection of the present invention is not limited to the specific details described in this embodiment.

[0035] It should be noted that this embodiment uses a relatively small simulation dataset to clearly and concisely illustrate the technical solution and core method flow of the present invention. In practical applications of the present invention, hundreds or even thousands of sets of simulation data can be generated by increasing the orthogonal array factor levels or adopting a full factorial design, depending on the required accuracy, to further improve the prediction accuracy and robustness of the surrogate model.

[0036] 1. A method for predicting laser cladding deformation and stress of non-rigid substrates based on machine learning, comprising:

[0037] S1. Construct a thermo-mechanical coupled finite element simulation model:

[0038] In this embodiment, a thermo-mechanical coupled finite element simulation model is constructed using the commercial software ANSYS.

[0039] Heat transfer model: using the three-dimensional transient temperature field heat source control partial differential equation:

[0040]

[0041] Where ρ is the material density, C p The specific heat capacity at constant pressure is given by T, temperature, and time, in Kelvin.x K y K z Let be the thermal conductivity in the x, y, and z directions, respectively, and Q be the heat power per unit volume.

[0042] Heat source model: Using the Gaussian moving heat source expression:

[0043]

[0044] Where q(r) is the heat flux density, η is the material's absorption rate of the laser, P is the laser power, r is the laser spot radius, and R is the distance from any point to the center of the heat source.

[0045] Material model: The substrate material is titanium alloy (Ti6Al4V), and the coating material is titanium (Ti). Its thermophysical properties (density, specific heat capacity, thermal conductivity) and mechanical properties (elastic modulus, yield stress, coefficient of thermal expansion) are all set as temperature-dependent functions, and the data are from the material library.

[0046] Geometric model: Substrate size: 100mm×50mm×2mm, cladding layer single-scan, size is 40mm×2mm×1mm.

[0047] S2. Establish a process parameter - deformation / stress database:

[0048] Input parameter range: laser power (300, 400, 500W), scanning rate (6, 8, 10mm / s), powder feeding rate (10, 15, 20g / min). To verify the feasibility of the method with high efficiency and to construct a preliminary surrogate model, this embodiment uses L9(3 3 A total of 9 sets of finite element simulation calculations were performed using orthogonal arrays.

[0049] Output parameters: The deflection curve of the center line of the upper surface of the substrate after the simulation is completed, as well as the residual stress distribution on the path, are extracted as output to characterize the overall deformation field and residual stress field of the substrate.

[0050] S3, Data Standardization / Normalization Processing:

[0051] Z-Score normalization is applied to all input and output data:

[0052]

[0053] Where x is the original data, μ is the mean, and σ is the variance. norm This is the standardized data.

[0054] The nine sets of data were randomly divided into a training set (7 sets) and a validation set (2 sets) in a 7:2 ratio.

[0055] S4. Select a machine learning model for training:

[0056] A variety of machine learning algorithms are selected, including linear regression, decision tree regression, random forest regression, support vector machine regression, and artificial neural network algorithms, and the machine learning model is trained using a training set.

[0057] S5. Model Evaluation and Selection:

[0058] Various machine learning algorithms are evaluated and selected, with the optimal algorithm based on comprehensive evaluation serving as the surrogate model. Evaluation metrics include the coefficient of determination (R²). 2 The root mean square error (RMSE) and mean absolute error (MAE) are calculated using the following formulas:

[0059]

[0060] Where n is the total number of samples in the validation set, y i Let i be the true value of the i-th sample. Let be the predicted value for the i-th sample. This is the arithmetic mean of the true values ​​of all samples.

[0061] S6. Experimental Verification and Closed-Loop Optimization:

[0062] Laser cladding experiments were conducted on non-rigid substrates. The substrate material was titanium alloy (Ti6Al4V), with dimensions of 100mm × 50mm × 2mm, and the coating material was titanium (Ti). A set of parameter combinations (laser power 400W, scanning speed 8mm / s, powder feed rate 15g / min) not present in the training and validation sets was selected for experimental verification. The deformation field of the substrate after cladding was measured using a 3D scanner, and the residual stress was measured using X-ray diffraction.

[0063] The validation set is input into the surrogate model to obtain prediction results, which are then compared with the experimental results to verify the model's accuracy. If the mean absolute error (MAE) is lower than a preset threshold (e.g., 5%), the model is considered reliable; otherwise, the experimental data is added to the original training set, and the model training and optimization steps are repeated.

[0064] 2. Examples of Reverse Engineering System Applications

[0065] To demonstrate the application of this invention in process optimization, the reverse engineering process using the verified predictive model described above is briefly described below, such as... Figure 2 As shown. The target deformation of the substrate after cladding is set as δ. target The target residual stress is σ targetAn optimization algorithm (genetic algorithm) was used as the optimizer, and the previously validated machine learning algorithm was used as the predictor. During optimization, the population size was set to 50, and the iterations were 100. The optimization objective was to minimize the weighted mean square error between the predicted and target values. The calculation formula is as follows:

[0066]

[0067] in, These are laser power, scanning speed, and powder feeding rate, respectively; δ pred and σ pred The deformation and stress predicted by the surrogate model; δ target and σ target To optimize the objective value; w1 and w2 are weighting coefficients, in this example w1 = w2 = 0.5; the normalization factor is δ. norm =||δ target ||,σ norm =||σ target At the same time, the search range of process parameters must satisfy the constraint: P∈[P min ,P max ], v∈[v min ,v max ],

[0068] Through iterative search of optimization algorithms, this reverse engineering system can automatically output one or more sets of laser process parameter combinations (such as laser power, scanning speed, and powder feeding rate) that approximate the optimization target, thereby achieving active and precise control over the deformation and stress of non-rigid substrates.

[0069] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for predicting laser cladding deformation and stress of non-rigid substrates based on machine learning, characterized in that, The method comprises the following steps: S1, constructing a thermal-mechanical coupling finite element simulation model for a non-rigid substrate, the model being based on laser cladding process parameters, thermal physical property parameters of the substrate and coating material, and geometric size parameters; S2, performing parameterized scanning simulation using the thermal-mechanical coupling finite element simulation model to generate a simulation data set, so as to establish a process parameter-deformation / stress database, the database comprising multiple groups of input parameters and corresponding substrate deformation field and residual stress field output parameters; S3, preprocessing the simulation data set, including data standardization or normalization processing, to eliminate dimensional influence; S4, dividing the preprocessed data into a training set and a validation set, and selecting at least one machine learning algorithm, training using the training set to establish a machine learning prediction model from the input parameters to the output parameters; S5, based on the validation set, evaluating and comparing the trained machine learning prediction model through performance indicators, and selecting an optimal model as a surrogate model; S6, performing laser cladding experiments, obtaining measured input parameters and corresponding measured output parameters of the substrate deformation field and residual stress field, verifying and closed-loop iterative optimizing the optimal surrogate model based on the measured data until the model accuracy meets the preset requirements, thereby obtaining an experimentally verified prediction model; Wherein, the deformation of the non-rigid substrate is actively controlled to release the residual stress, thereby inhibiting the cracking of the cladding layer.

2. The method of claim 1, wherein, In step S1, the laser process parameters include laser power, scanning speed, and powder feeding rate; the thermal physical property parameters of the substrate and coating material include density, specific heat capacity, thermal conductivity, thermal expansion coefficient, elastic modulus, and Poisson's ratio; the geometric size parameters of the substrate include the length, width, and thickness of the substrate; and the geometric size parameters of the coating material include the width, thickness, and relative substrate position of the cladding layer.

3. The method of claim 1, wherein, In step S2, the orthogonal experimental design method is used to determine the combination of multiple groups of input parameters for parameterized scanning simulation.

4. The method of claim 1, wherein, The machine learning algorithm in step S4 is selected from at least one of the following algorithms: linear regression, decision tree regression, random forest regression, support vector machine regression, and artificial neural network.

5. The method of claim 1, wherein, The closed-loop iterative optimization in step S6 is: when the model prediction accuracy does not meet the preset requirements, adding experimental data to the training set, and re-executing steps S4 and S5 to optimize the model.

6. A reverse design system for laser cladding process parameters for non-rigid substrates for implementing the method according to any one of claims 1 to 5, characterized in that, The system comprises: (1) an experimentally verified prediction model module for receiving laser process parameters and predicting corresponding substrate deformation field and residual stress field; (2) an optimization target setting module for setting target deformation and / or target residual stress value; (3) an optimization algorithm module configured to iteratively search for laser process parameters so that the difference between the deformation field and residual stress field predicted by the prediction model module for the searched parameters and the optimization target is minimized; (4) a parameter output module for outputting the final laser process parameter combination.

7. The system of claim 6, wherein, The optimization algorithm used by the optimization algorithm module is a genetic algorithm or a multi-objective particle swarm optimization algorithm.

Images