An ultra-high-speed laser cladding process parameter closed-loop optimization method for shaft parts

Abstract

The application relates to a kind of shaft parts ultra-high-speed laser cladding process parameter closed-loop optimization method, the method takes laser power, scanning speed and powder feeding rate as input variable, takes cladding layer hardness, dilution rate and heat affected zone size as fatigue sensitive quality index. Through two-stage orthogonal test to construct sample database, the multi-output nonlinear mapping model between process parameters and quality index is established by using genetic algorithm optimized back propagation neural network, the Pareto frontier is solved under the double constraints of surface roughness and surface energy density by combining multi-objective particle swarm optimization algorithm, and the comprehensive optimal process parameter combination is determined by the organization analysis of representative parameters, rotary bending fatigue test and stress relief annealing treatment. The closed-loop optimization path of "process window identification-agent modeling-multi-objective optimization-fatigue verification-post-processing strengthening" is established, the direct correlation between process parameters and fatigue service performance is realized, and the fatigue reliability of shaft parts remanufacturing repair is significantly improved.

Description

Technical Field

[0001] This application relates to the fields of laser additive manufacturing and remanufacturing technology, and in particular to a closed-loop optimization method for ultra-high-speed laser cladding process parameters of shaft parts. Background Technology

[0002] Shaft components, especially critical load-bearing components such as railway axles, motor shafts, crankshafts, turbine rotors, and rolling mill rolls, are subjected to complex conditions such as alternating loads, impacts, and corrosion for extended periods. Their surfaces are highly susceptible to damage such as wear, corrosion pits, scratches, and localized fatigue cracks. For these fatigue-sensitive components, the core requirement of remanufacturing repair is not only the restoration of geometric dimensions, but also how to effectively control the degree of thermal damage, dilution rate, and microstructure gradient distribution in the repair area to achieve the restoration or even improvement of fatigue performance after repair. Laser cladding technology, as a cutting-edge surface modification technology, is widely used in component repair due to its advantages such as low heat input, controllable dilution rate, and high coating bonding strength. However, traditional laser cladding suffers from excessively wide heat-affected zones due to excessive heat input, and also results in a higher dilution rate. Ultra-high-speed laser cladding technology, with its extremely high scanning speed and precise energy input characteristics, significantly shortens the residence time of the molten pool, effectively reducing the width of the heat-affected zone and lowering the dilution rate, while simultaneously improving deposition efficiency. This represents an ideal technical approach to solving these problems. However, there is a strong nonlinear coupling relationship between the core parameters involved in this technology, such as laser power, scanning speed, and powder feeding rate. How to establish a set of process parameter optimization methods for fatigue performance under limited experimental costs has become a key technical bottleneck in current engineering applications.

[0003] Currently, the optimization of laser cladding process parameters mainly includes single-factor methods, orthogonal experimental methods, response surface methodology, and some machine learning-assisted methods. Single-factor methods are simple to operate but cannot reflect the interactions between parameters; while orthogonal experimental methods can analyze the primary and secondary effects of factors with fewer experimental points, their fitting ability for strongly nonlinear and multi-parameter coupled problems in ultra-high-speed cladding processes is limited; response surface methodology often uses quadratic polynomial modeling, and its prediction accuracy is often insufficient when facing multi-input, multi-objective, and strongly nonlinear processes. While some existing technical solutions incorporating machine learning models such as neural networks have improved nonlinear prediction capabilities to some extent, they generally suffer from two shortcomings: first, the optimization objectives often remain at macroscopic forming indicators such as the aspect ratio, dilution rate, or surface hardness of the cladding layer, making it difficult to directly map the fatigue risk of the repaired component; second, the optimization process is disconnected from the final service performance verification, and the selected process parameter combinations often only satisfy "forming feasibility," but may not be "service optimal" under actual fatigue load conditions.

[0004] In summary, existing technologies for optimizing ultra-high-speed laser cladding processes for fatigue-sensitive shaft parts still suffer from the following core problems: First, parameter window identification efficiency is low, and single-stage experimental design struggles to simultaneously meet the dual requirements of "wide-range initial screening" and "narrow-range refinement," leading to high costs of ineffective experiments and difficulty in quickly locking in a stable forming window. Second, there is a significant disconnect between optimization objectives and service performance. While conventional optimization indicators are fatigue-sensitive factors, they lack synergistic consideration with heat-affected zone gradients and microstructure uniformity, and even more so, direct fatigue test closed-loop verification. Third, there is a lack of a complete closed-loop control path covering post-processing stages. Existing solutions generally do not consider the control of residual stress distribution and hardness gradient after repair, and the unstress-relieved cladding layer exhibits significant mechanical discontinuities near the heat-affected zone, easily becoming a source of fatigue crack initiation. In particular, existing optimization schemes based on orthogonal experiments, response surface methodology, or neural networks often use model prediction errors or forming quality as evaluation endpoints, lacking a closed-loop mechanism to feed fatigue test results back to the sample database and revise the prediction model, resulting in a disconnect between optimization parameters and final service reliability.

[0005] Therefore, there is an urgent need in related technologies to solve the problems of strong parameter coupling, low test efficiency, disconnect between optimization results and fatigue performance, and lack of post-processing synergistic control. Summary of the Invention

[0006] Based on this, it is necessary to provide a closed-loop optimization method for ultra-high-speed laser cladding process parameters of shaft parts, which can solve the problems of strong parameter coupling, low test efficiency, disconnect between optimization results and fatigue performance, and lack of post-processing synergistic control.

[0007] Firstly, this application provides a closed-loop optimization method for ultra-high-speed laser cladding process parameters of shaft-type parts. The method includes: Obtain the basic material parameters and fixed process conditions of the shaft parts to be repaired; The first stage of wide-area orthogonal experiment was performed within the preset first parameter range to prepare the first cladding sample set, and the formable process window was screened based on the continuity of the cladding layer, fusion state, cross-sectional defects, dilution rate, heat-affected zone size and surface roughness. A second-stage refined orthogonal experiment is performed within the formable process window, and random parameter combination samples are added to prepare a second cladding sample set. The second cladding sample set and the effective samples falling into the formable process window in the first cladding sample set together constitute a sample database. The hardness, dilution rate, and heat-affected zone size of the cladding layer of each sample in the sample database are measured as fatigue-sensitive quality indicators. Laser power, scanning speed, and powder feeding rate are used as input variables, and the fatigue-sensitive quality indicators are used as output responses to construct a multi-objective prediction model based on a backpropagation neural network optimized by a genetic algorithm. With the optimization objectives of maximizing the hardness of the cladding layer, minimizing the dilution rate, and minimizing the size of the heat-affected zone as the optimization conditions, and with surface roughness and surface energy density as the constraints, the Pareto optimal solution set is solved on the multi-objective prediction model using a multi-objective particle swarm optimization algorithm. Representative parameter schemes are selected from the Pareto optimal solution set according to the thermal input gradient and comprehensive evaluation rules, and laser cladding reproduction verification, microstructure analysis and rotational bending fatigue test are performed; when the reproduction verification result or fatigue test result does not meet the preset fatigue performance criterion, the corresponding measured data is added to the sample database and the multi-objective prediction model is updated, and multi-objective optimization is re-executed; when the preset fatigue performance criterion is met, the comprehensive optimal process parameter combination is determined; The cladding repair parts prepared using the aforementioned optimal combination of process parameters are subjected to medium-low temperature stress relief heat treatment.

[0008] Optionally, in one embodiment of this application, both the first-stage wide-range orthogonal experiment and the second-stage refined orthogonal experiment are three-factor, five-level orthogonal designs. The level ranges of the variables to be optimized in the two stages are different, and additional random parameter combination samples are added in the second-stage experiment.

[0009] Optionally, in one embodiment of this application, the hardness of the cladding layer is obtained by performing multiple micro Vickers hardness tests along the middle region of the cladding layer cross-section and taking the average value. The dilution rate is obtained by calculating the ratio of the projected area formed by the melting of the substrate in the molten pool to the total projected area of ​​the molten pool and the cladding layer. The dimensions of the heat-affected zone are obtained by measuring the width of the area where the microstructure of the substrate changes significantly along the normal direction of the fusion line, and averaging the measurement results at multiple locations.

[0010] Optionally, in one embodiment of this application, the backpropagation neural network multi-objective prediction model optimized by genetic algorithm includes an input layer, a hidden layer and an output layer, wherein the number of nodes in the input layer corresponds to the number of variables to be optimized, the number of nodes in the output layer corresponds to the number of quality evaluation indicators, and the genetic algorithm is used to optimize the initial weights and thresholds of the neural network.

[0011] Optionally, in one embodiment of this application, the multi-objective particle swarm optimization algorithm uses laser power, scanning speed and powder feeding rate as decision variables, the upper limit of surface roughness in the constraints does not exceed a preset threshold, and the surface energy density range is predetermined based on a comprehensive analysis of the formable process window.

[0012] Optionally, in one embodiment of this application, the representative parameter scheme is divided into low heat input, medium heat input and high heat input intervals from the Pareto optimal solution set according to the surface energy density from low to high, and the normalized comprehensive evaluation values ​​of cladding layer hardness, dilution rate, heat-affected zone size and surface roughness in each interval are screened. The rotational bending fatigue test is carried out under the condition of stress ratio of negative one, and the number of cycles or fatigue fracture is used as the test termination criterion.

[0013] Optionally, in one embodiment of this application, the comprehensive optimal process parameter combination is obtained by comprehensively screening based on the matching relationship between the hardness of the cladding layer and the hardness of the substrate, the gradient characteristics of the heat-affected zone, the uniformity of the microstructure, and the fatigue response performance.

[0014] Optionally, in one embodiment of this application, the temperature of the medium-low temperature stress relief heat treatment is 500–600°C, and after holding at that temperature for 1–3 hours, it is cooled in the furnace to below 200°C and then air-cooled.

[0015] Optionally, in one embodiment of this application, the criteria for determining the formable process window include: the cladding layer is continuous, there are no obvious unfusion defects or through cracks, the dilution rate is lower than a preset threshold, the size of the heat-affected zone is lower than a preset threshold, and the surface roughness is lower than a preset threshold.

[0016] Optionally, in one embodiment of this application, the fixed process conditions include defocusing amount, overlap rate, laser spot diameter, powder feeding gas flow rate, and protective gas flow rate, which remain constant throughout the entire test.

[0017] The aforementioned closed-loop optimization method for ultra-high-speed laser cladding process parameters of shaft parts, by constructing a complete technical route of "wide-domain initial screening—narrow-domain refinement—multi-objective collaborative optimization—fatigue closed-loop verification—post-processing strengthening," has the following advantages compared with existing technologies: (1) A direct correlation between process parameters and fatigue service performance has been achieved. This invention breaks through the limitation of traditional methods that only use macroscopic forming quality as the optimization endpoint. It uses the hardness of the cladding layer, the dilution rate η, and the size of the heat-affected zone D as fatigue-sensitive quality indicators, and performs closed-loop verification through rotational bending fatigue tests of representative parameters to ensure the service reliability and engineering applicability of the optimization results under fatigue load conditions.

[0018] (2) Significantly improves the optimization efficiency and accuracy under strongly nonlinear coupled process parameters. This invention adopts a two-stage orthogonal experimental design to efficiently identify stable forming windows under limited sample conditions, and uses a GA-BPNN surrogate model to establish a high-precision nonlinear mapping relationship between process parameters and multiple quality indicators, effectively solving the problem of significant parameter interaction and insufficient prediction accuracy of traditional modeling methods in ultra-high-speed laser cladding.

[0019] (3) A dual constraint mechanism is introduced to balance forming quality and heat input control. In the MOPSO multi-objective optimization process, this invention innovatively uses surface roughness and surface energy density as dual constraints to ensure that the parameter combination in the Pareto optimal solution set not only meets the intrinsic quality requirements of low dilution and low thermal influence, but also has good surface integrity and reasonable energy input level.

[0020] (4) A closed-loop strengthening path covering post-treatment was established. Based on the determination of the comprehensive optimal process parameters, this invention further introduces a 550℃ stress-relief annealing post-treatment process, which effectively releases the residual tensile stress in the transition area between the cladding layer and the heat-affected zone, significantly weakens the interface hardness gradient and mechanical discontinuity, thereby further reducing the sensitivity to fatigue crack initiation and substantially improving the fatigue life and fatigue limit of the repaired component. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating a closed-loop optimization method for ultra-high-speed laser cladding process parameters of shaft-type parts in one embodiment. Figure 2 This is a schematic diagram showing the cross-sectional quality indicators of the cladding layer and the location of microhardness tests in one embodiment. Figure 3 Here is a diagram of the multi-target prediction BPNN neural network structure in one embodiment; Figure 4 Here is a flowchart of GA-BPNN multi-objective prediction and MOPSO multi-objective optimization in one embodiment; Figure 5 This is the Pareto front solution set in one embodiment; Figure 6 This includes, in one embodiment, an elemental distribution diagram of the fatigue specimen cross section, a microstructure distribution diagram, and a fatigue fracture surface (including elemental distribution) diagram; Figure 7 This is a comparison diagram of the hardness gradient before and after stress-relief heat treatment in one embodiment. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0023] In one embodiment, such as Figure 1 As shown, a closed-loop optimization method for ultra-high-speed laser cladding process parameters of shaft parts is provided, including the following steps: S101: Obtain the basic material parameters and fixed process conditions of the shaft parts to be repaired.

[0024] In this application, the basic parameters of the material of the shaft part to be repaired are obtained, and process variables and fixed process conditions are determined based on the target material, sample geometry, and the capabilities of the ultra-high-speed laser cladding equipment. The basic parameters include the shaft part material grade, sample geometry, type of cladding material, laser cladding equipment model, and the equipment's allowable power, scanning speed, and powder feeding range. Preferably, the base material is EA4T axle steel, and the cladding material is Inconel 625 nickel-based alloy powder. The powder particle size ranges from 20 to 160 μm, with an average particle size of 75 μm. Before cladding, the powder is dried in a vacuum drying oven at 120°C for 2 hours to ensure good powder flowability and feeding stability. The ultra-high-speed laser cladding system used in this invention mainly consists of a laser cladding head, a laser system, a CNC machine tool control system, a powder feeder, a water cooling system, and a protective gas system. During parameter screening, model construction, and experimental verification, except for the process variables to be optimized, the remaining process parameters are kept constant to ensure comparability between different experimental groups. Preferably, the fixed process parameters are: defocusing amount +2mm, overlap rate 81.25%, laser spot diameter 1.6mm, powder feeding gas flow rate 9L / min, and protective gas flow rate 30L / min. Service fatigue requirements include target stress level, fatigue life threshold, and test termination cycle. These fixed process conditions remain consistent throughout all tests, model training, and optimization verification processes to reduce the interference of non-optimized variables on cladding quality indicators.

[0025] S102: Perform the first stage wide-range orthogonal experiment within the preset first parameter range, prepare the first cladding sample set, and screen the formable process window based on the continuity of the cladding layer, fusion state, cross-sectional defects, dilution rate, heat-affected zone size and surface roughness.

[0026] In this embodiment, the first-stage wide-range orthogonal experiment adopts a three-factor, five-level orthogonal experimental design, as shown in Table 1. Twenty-five groups of laser cladding samples were prepared within a wide parameter range to quickly identify the formable process window for ultra-high-speed laser cladding and eliminate parameter combinations that result in discontinuous cladding layers, insufficient fusion, or excessive heat input leading to significant thermal damage. The parameter-optimized sample design is a straight cylinder with dimensions of Φ7mm × 50mm, and the cladding layer width is set to 15mm. Before cladding, the sample surface is pretreated, and the initial surface roughness is controlled to Ra = 1.6μm to reduce the influence of substrate surface state differences on the forming quality. The prepared first-stage samples were subjected to cross-sectional sample preparation, morphological observation, and performance testing to obtain the actual process parameter combinations and cladding layer quality indicators corresponding to each group of samples, as shown in Table 2. Specifically, the process parameters include laser power P, scanning speed V, and powder feeding rate F, as shown in Table 2. Figure 2 As shown, the quality indicators of the cladding layer include the cladding layer hardness H, dilution rate η, and heat-affected zone size D. Preferably, the parameter ranges for the first stage are: P=800–1600W, V=10–30m / min, F=12.8–19.2g / min. The screening rules for the formable process window are as follows: when the cladding layer is continuous along the axial direction, the cross-section has no obvious incomplete fusion, through cracks, or large-sized holes, the cladding layer forms a stable metallurgical bond with the substrate, and the dilution rate η, heat-affected zone size D, and surface roughness Sa all do not exceed the preset threshold, the corresponding parameter combination is marked as a valid forming sample; when the sample shows discontinuous cladding, spheroidization, insufficient fusion, or abnormal expansion of the heat-affected zone, it is marked as an invalid sample or a high-heat-damage sample, and will not participate in subsequent refined modeling or will only be recorded as a boundary sample.

[0027] Table 1. Horizontal Table of Wide-Domain Orthogonal Experiments

[0028] Table 2. Results of the wide-domain orthogonal experiment

[0029] S103: Perform a second-stage refined orthogonal experiment within the formable process window, supplement random parameter combination samples, prepare a second cladding sample set, and combine the second cladding sample set with the valid samples from the first cladding sample set that fall within the formable process window to form a sample database.

[0030] In this embodiment, within the selected reasonable process window, a second-stage parameter refinement experiment is further conducted. A three-factor, five-level orthogonal experimental design is still used, as shown in Table 3, with additional random experimental samples to improve sample resolution and enhance the prediction accuracy of the subsequent machine learning proxy model within the effective parameter domain. Preferably, the parameter ranges for the second stage are: P = 1400–1800W, V = 15–25m / min, F = 12.8–19.2g / min. The second-stage data, together with the effective data from the first stage, constitute the sample database. Specifically, a unified quality evaluation and data processing are performed on each group of samples to establish a sample database containing process parameters and cladding layer quality indicators. Specifically, the cladding sample is halved along the axial direction, and its surface and cross-sectional morphology are obtained after inlaying, grinding, polishing, and etching. The surface roughness Sa, cladding layer thickness, dilution zone width, and heat-affected zone size D of the cladding layer are measured. The geometric dimensions and surface roughness of the cladding layer are measured using a laser confocal microscope, and the statistical results are shown in Table 4.

[0031] In one embodiment of this application, the hardness H of the cladding layer is measured using the Vickers microhardness method, such as... Figure 2 As shown, the middle section of the cladding layer was selected as the test location for the fifth pass, and the average of the five measurements was taken as the final hardness value. The dilution rate η was calculated based on the geometric relationship of the molten pool. The size of the heat-affected zone D was defined as the width or size of the area where the matrix structure near the cladding interface undergoes significant changes, and was obtained by averaging multiple measurement locations. The dilution rate η was calculated according to η=A_b / (A_t+A_c)×100%, where A_b is the projected area formed by the melting of the matrix in the molten pool, and A_t+A_c is the total projected area of ​​the molten pool formed by the melting areas of the cladding layer and the matrix. The area was obtained by converting the actual area from the pixel count of the microscopic image. At the same time, based on the results of the two-stage test, the area where Es is lower than the lower limit of stable forming was identified as the low heat input failure area, and the area where Es is higher than the upper limit of stable forming was identified as the high heat input thermal damage area. 2.00–2.50 J / mm² was determined as the reasonable heat input window. The surface energy density Es is calculated according to Es=P / (V·d), where P is the laser power, V is the scanning speed, and d is the laser spot diameter or effective working width; when V is in m / min, it is converted to mm / s before calculation.

[0032] Table 3. Orthogonal experimental level table for moldable process window

[0033] Table 4. Results of orthogonal experiments on the formable process window

[0034] S104: Measure the hardness, dilution rate, and heat-affected zone size of the cladding layer for each sample in the sample database as fatigue-sensitive quality indicators, and use laser power, scanning speed, and powder feeding rate as input variables, and the fatigue-sensitive quality indicators as output responses to construct a multi-objective prediction model based on a backpropagation neural network optimized by a genetic algorithm.

[0035] In this embodiment, process parameters P, V, and F are used as input features, and cladding layer hardness H, dilution rate η, and heat-affected zone size D are used as output responses. A backpropagation neural network model based on genetic algorithm optimization, namely the GA-BPNN multi-objective prediction model, is constructed to establish a nonlinear mapping relationship between process parameters and cladding layer quality indicators. At the same time, an independent prediction model for surface roughness Sa is established for subsequent constraint screening.

[0036] In one embodiment of this application, such as Figure 3 As shown, the GA-BPNN model has 3 input layer nodes, corresponding to P, V, and F respectively, and 3 output layer nodes, corresponding to H, η, and D respectively. The number of hidden layer nodes is determined through candidate search or cross-validation. The hidden layer activation function is tansig, and the output layer activation function is purelin. The preferred training algorithm is the Levenberg-Marquardt algorithm. All input and output data are normalized to [-1, 1]. The genetic algorithm is used to optimize the initial weights and thresholds of the BP neural network. Performance evaluation metrics include RMSE, MAPE, MAE, and R². The individual encoding of the genetic algorithm is composed of the input layer to hidden layer weights, hidden layer thresholds, hidden layer to output layer weights, and output layer thresholds concatenated sequentially. The mean squared error of the validation set or cross-validation set is used as the fitness function. After the genetic algorithm completes global optimization, the optimal individual is decoded into the initial weights and thresholds of the BPNN, and local training is performed using the BP algorithm. If the number of samples is small, leave-one-out cross-validation or K-fold cross-validation is used to evaluate the model's generalization ability.

[0037] S105: With the optimization objectives of maximizing the hardness of the cladding layer, minimizing the dilution rate, and minimizing the size of the heat-affected zone, and with surface roughness and surface energy density as constraints, the Pareto optimal solution set is solved on the multi-objective prediction model using a multi-objective particle swarm optimization algorithm.

[0038] In this embodiment, based on the trained GA-BPNN model, the multi-objective particle swarm optimization algorithm (MOPSO) is used for multi-objective optimization. The optimization objectives are set as: maximizing the cladding layer hardness H, minimizing the dilution rate η, and minimizing the heat-affected zone size D. Process parameters P, V, and F are used as decision variables, with surface roughness Sa and surface energy density Es as dual constraints, to obtain the Pareto optimal solution set. The MOPSO algorithm includes external file maintenance, crowding distance or grid density filtering, and infeasible solution elimination or penalty function processing. The number of particles is 50–100, and the maximum number of iterations is 100–300. Figure 4 The diagram shown is a flowchart of the GA-BPNN multi-objective prediction and MOPSO multi-objective optimization.

[0039] In one embodiment of this application, the MOPSO algorithm uses H maximization, η minimization, and D minimization as three objectives, while using Sa < 20 μm and Es = 2.00–2.50 J / mm² as constraints, filtering out solutions that do not meet the requirements of equipment capability, process feasibility, or surface quality. The MOPSO optimization range is determined based on a combination of two-stage effective forming samples and first-stage low thermal damage samples, and the range of formable parameters that meet the Sa and Es constraints is extended to P = 1200–1400 W, V = 20–25 m / min, and F = 12.8–17.6 g / min.

[0040] The accuracy of multi-objective prediction was verified through experiments. As shown in Table 5, the errors between the predicted and actual values ​​for all targets were small. Figure 5 The Pareto front solution set shown is used to select results with different energy densities, while satisfying the above constraints, resulting in a total of 8 sets of parameters.

[0041] Table 5. Comparison of Predicted and Experimental Results

[0042] S106: Select representative parameter schemes from the Pareto optimal solution set according to the thermal input gradient and comprehensive evaluation rules, and conduct laser cladding reproduction verification, microstructure analysis and rotational bending fatigue test; when the reproduction verification result or fatigue test result does not meet the preset fatigue performance criteria, supplement the corresponding measured data to the sample database and update the multi-objective prediction model, and re-execute the multi-objective optimization; when the preset fatigue performance criteria are met, determine the comprehensive optimal process parameter combination.

[0043] In this embodiment, representative process parameter schemes are selected from the obtained Pareto optimal solution set for laser cladding reproduction experiments. The hardness H, dilution rate η, heat-affected zone size D, and surface roughness Sa of the cladding layer corresponding to each representative parameter are measured and compared with model prediction results to verify the reliability of the surrogate model and multi-objective optimization results. Simultaneously, microstructure, elemental distribution, and rotational bending fatigue tests are conducted on the representative parameter samples. Parameter groups with good matching hardness to the matrix are selected for fatigue sensitivity comparison. The optimal parameter group within the current process window is determined by comprehensively considering dilution rate, hardness matching, heat-affected zone gradient characteristics, microstructure uniformity, and fatigue response. Figure 6 The diagram shows the elemental distribution of the fatigue specimen cross section, the microstructure distribution, and the fatigue fracture surface (including elemental distribution).

[0044] In one embodiment of this application, the selection of representative parameter schemes should first eliminate solutions that do not satisfy the Sa and Es constraints. Preferably, the Pareto optimal solution set is divided into low heat input, medium heat input, and high heat input intervals according to the surface energy density from low to high, and the selection is performed based on the normalized comprehensive evaluation values ​​of cladding layer hardness, dilution rate, heat-affected zone size, and surface roughness in each interval.

[0045] In one embodiment of this application, a rotary bending fatigue testing machine is preferably used in fatigue verification. The test is conducted in a room temperature air environment with a stress ratio of R=-1. The test is stopped when the specimen fractures or the cycle life reaches 1×10^7 cycles. The fatigue behavior is comprehensively evaluated by combining the fracture morphology, element distribution and microstructure uniformity.

[0046] S107: Perform low-temperature stress relief heat treatment on the cladding repair part prepared using the aforementioned optimal combination of process parameters.

[0047] In this embodiment, the cladding repair specimen corresponding to the determined optimal parameter set undergoes low-to-medium temperature stress-relieving heat treatment. The heat treatment temperature can be 500–600 °C, and the holding time can be 1–3 h; preferably, a holding time of 550 °C for 2 h is used, followed by furnace cooling to below 200 °C and air cooling. This heat treatment is used to release residual tensile stress in the transition region between the cladding layer and the heat-affected zone, reduce the hardness gradient between the cladding layer, the heat-affected zone, and the substrate, and alleviate stress concentration near the interface, thereby reducing the susceptibility of fatigue cracks to preferential initiation at the interface or heat-affected zone. Figure 7 The image shown is a comparison of the hardness gradient before and after stress-relief heat treatment. Table 6 below shows the rotational bending fatigue life test results of specimens with different parameters under the fatigue limit (355 MPa) of the base material.

[0048] Table 6. Results of Rotary Bending Fatigue Life Test

[0049] Table 6 shows that, compared with the untreated specimens of parameters 1, 4 and 6, all three specimens in the heat-treated group 6 achieved 1×10^7 cycles without fracture under a stress level of 355 MPa. This indicates that after further stress-relief annealing at 550 ℃ based on the optimal process parameters, the crack initiation sensitivity of the repaired specimens was significantly reduced, the fatigue life was significantly improved, and the fatigue limit was also significantly improved.

[0050] This embodiment illustrates that by adopting a closed-loop optimization path of "two-stage orthogonal experiment + GA-BPNN surrogate modeling + MOPSO multi-objective optimization + representative parameter fatigue verification + 550 ℃ stress-relief annealing", it is possible to achieve rapid decision-making, low dilution and low thermal impact control of ultra-high speed laser cladding process parameters based on limited experimental samples. Furthermore, by weakening the hardness gradient and improving fatigue service performance through subsequent heat treatment, it provides a technical route with engineering feasibility and licensing support for the remanufacturing and repair of shaft parts.

[0051] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0052] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0053] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0054] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A closed-loop optimization method for ultra-high-speed laser cladding process parameters of shaft parts, characterized in that, The method includes: Obtain the basic material parameters and fixed process conditions of the shaft parts to be repaired; The first stage of wide-area orthogonal experiment was performed within the preset first parameter range to prepare the first cladding sample set, and the formable process window was screened based on the continuity of the cladding layer, fusion state, cross-sectional defects, dilution rate, heat-affected zone size and surface roughness. A second-stage refined orthogonal experiment is performed within the formable process window, and random parameter combination samples are added to prepare a second cladding sample set. The second cladding sample set and the effective samples falling into the formable process window in the first cladding sample set together constitute a sample database. The hardness, dilution rate, and heat-affected zone size of the cladding layer of each sample in the sample database are measured as fatigue-sensitive quality indicators. Laser power, scanning speed, and powder feeding rate are used as input variables, and the fatigue-sensitive quality indicators are used as output responses to construct a multi-objective prediction model based on a backpropagation neural network optimized by a genetic algorithm. With the optimization objectives of maximizing the hardness of the cladding layer, minimizing the dilution rate, and minimizing the size of the heat-affected zone as the optimization conditions, and with surface roughness and surface energy density as the constraints, the Pareto optimal solution set is solved on the multi-objective prediction model using a multi-objective particle swarm optimization algorithm. Representative parameter schemes are selected from the Pareto optimal solution set according to the thermal input gradient and comprehensive evaluation rules, and laser cladding reproduction verification, microstructure analysis and rotational bending fatigue test are performed; when the reproduction verification result or fatigue test result does not meet the preset fatigue performance criterion, the corresponding measured data is added to the sample database and the multi-objective prediction model is updated, and multi-objective optimization is re-executed; when the preset fatigue performance criterion is met, the comprehensive optimal process parameter combination is determined; The cladding repair parts prepared using the aforementioned optimal combination of process parameters are subjected to medium-low temperature stress relief heat treatment.

2. The closed-loop optimization method for ultra-high-speed laser cladding process parameters of shaft parts according to claim 1, characterized in that, Both the first-stage wide-range orthogonal experiment and the second-stage refined orthogonal experiment are three-factor, five-level orthogonal designs. The level ranges of the variables to be optimized are different in the two stages, and additional random parameter combination samples are added in the second-stage experiment.

3. The closed-loop optimization method for ultra-high-speed laser cladding process parameters of shaft parts according to claim 1, characterized in that, The hardness of the cladding layer is obtained by performing multiple micro Vickers hardness tests along the middle region of the cladding layer cross-section and taking the average value. The dilution rate is obtained by calculating the ratio of the projected area formed by the melting of the substrate in the molten pool to the total projected area of ​​the molten pool and the cladding layer. The dimensions of the heat-affected zone are obtained by measuring the width of the area where the microstructure of the substrate changes significantly along the normal direction of the fusion line, and averaging the measurement results at multiple locations.

4. The closed-loop optimization method for ultra-high-speed laser cladding process parameters of shaft parts according to claim 1, characterized in that, The backpropagation neural network multi-objective prediction model optimized by the genetic algorithm includes an input layer, a hidden layer, and an output layer. The number of nodes in the input layer corresponds to the number of variables to be optimized, and the number of nodes in the output layer corresponds to the number of quality evaluation indicators. The genetic algorithm is used to optimize the initial weights and thresholds of the neural network.

5. The closed-loop optimization method for ultra-high-speed laser cladding process parameters of shaft parts according to claim 1, characterized in that, The multi-objective particle swarm optimization algorithm uses laser power, scanning speed and powder feeding rate as decision variables. Among the constraints, the upper limit of surface roughness does not exceed a preset threshold, and the surface energy density range is predetermined based on a comprehensive analysis of the formable process window.

6. The closed-loop optimization method for ultra-high-speed laser cladding process parameters of shaft parts according to claim 1, characterized in that, The representative parameter scheme is divided into low heat input, medium heat input and high heat input intervals from the Pareto optimal solution set according to the surface energy density from low to high. The normalized comprehensive evaluation values ​​of cladding hardness, dilution rate, heat-affected zone size and surface roughness in each interval are used for screening. The rotational bending fatigue test is carried out under the condition of stress ratio of negative one, and the number of cycles or fatigue fracture is used as the test termination criterion.

7. The closed-loop optimization method for ultra-high-speed laser cladding process parameters of shaft parts according to claim 1, characterized in that, The optimal combination of process parameters is obtained by comprehensively screening based on the matching relationship between the hardness of the cladding layer and the hardness of the substrate, the gradient characteristics of the heat-affected zone, the uniformity of the microstructure, and the fatigue response performance.

8. The closed-loop optimization method for ultra-high-speed laser cladding process parameters of shaft parts according to claim 1, characterized in that, The medium-low temperature stress relief heat treatment is performed at a temperature of 500–600℃, and after holding at that temperature for 1–3 hours, the temperature is cooled to below 200℃ in the furnace before being removed and air-cooled.

9. The method according to claim 1, characterized in that, The criteria for determining the formable process window include: continuous cladding layer, no obvious unfusion defects or through cracks, dilution rate lower than a preset threshold, heat-affected zone size lower than a preset threshold, and surface roughness lower than a preset threshold.

10. A closed-loop optimization method for ultra-high-speed laser cladding process parameters of shaft parts according to any one of claims 1 to 9, characterized in that, The fixed process conditions include defocusing amount, overlap rate, laser spot diameter, powder feeding gas flow rate, and protective gas flow rate, which are kept constant throughout the entire test.

Images