Aero-engine blade laser shock remanufacturing method based on digital twinning

Abstract

The invention discloses an aero-engine blade laser shock remanufacturing method based on digital twinning, and belongs to the field of deep transfer learning, and the method comprises the steps: collecting cutting force, vibration, temperature and other multi-modal data for the non-uniform stress distribution caused by tool wear, and constructing a monitoring system; and feature alignment is performed on different working condition data by using a maximum mean difference method, a wear diagnosis and stress prediction model is constructed in combination with deep transfer learning, and a stress concentration area is identified. And laser shock strengthening remanufacturing is implemented in a stress concentration area, shock parameters are optimized, a small sample data set is expanded in combination with an adversarial generative network, and the remanufacturing adaptability is improved. Incremental transfer learning is adopted, the model is dynamically adjusted based on a freezing-fine adjustment mechanism, and rapid optimization under the new working condition is achieved. The blade residual stress control precision and the machining stability are improved by firstly monitoring tool wear and stress distribution, then implementing laser shock peening and optimizing a remanufacturing strategy in combination with incremental transfer learning.

Description

Technical Field

[0001] The present invention belongs to the field of deep transfer learning, and particularly relates to a laser shock remanufacturing method for aero-engine blades based on digital twin. Background Technique

[0002] The manufacturing and remanufacturing processes of aero-engine blades involve high-precision numerical control machining. The surface quality and residual stress state of the blades directly affect their service performance and service life. However, during the numerical control machining process, the cumulative effect of tool wear will lead to fluctuations in cutting force, changes in heat input, and non-uniformity of material deformation, thus causing the imbalance of the residual stress distribution on the blade surface and even triggering local stress concentration problems. This stress non-uniformity phenomenon may reduce the fatigue resistance of the blade and affect the effect of subsequent laser shock peening remanufacturing. Therefore, how to accurately monitor the tool wear state during the machining process, predict the stress concentration area, and optimize the laser shock remanufacturing parameters is the key to improving the machining quality and service stability of the blade.

[0003] Traditional blade machining quality control mainly relies on the adjustment of empirical parameters, and it is difficult to accurately predict the influence of tool wear on stress distribution under complex working conditions, resulting in limitations in stress optimization strategies. In recent years, with the development of intelligent manufacturing and data-driven technologies, using multi-sensors to monitor the machining state in real time and combining deep learning and transfer learning methods to model tool wear and stress evolution has become an important direction to improve the numerical control machining accuracy. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides a laser shock remanufacturing method for aero-engine blades based on digital twin, including:

[0005] Collecting multi-modal data during the numerical control machining process of aero-engine blades, preprocessing the multi-modal data to obtain a preprocessed data set;

[0006] Constructing a tool wear diagnosis and stress distribution prediction model, inputting the preprocessed data set into the tool wear diagnosis and stress distribution prediction model, and obtaining tool wear diagnosis and stress distribution prediction results;

[0007] Based on the tool wear diagnosis and stress distribution prediction results, identifying the stress concentration area caused by tool wear and optimizing the laser shock peening strategy;

[0008] Implementing laser shock peening on the stress concentration area based on the optimized laser shock peening strategy to obtain feedback data;

[0009] Based on the feedback data, optimizing the tool wear diagnosis and stress distribution prediction model through an incremental transfer learning method to obtain a multi-condition adaptation model;

[0010] Predict the remanufacturing of different blades based on the multi - working - condition adaptation model.

[0011] Preferably, the process of obtaining the pre - processed data set includes: collecting multi - modal data in the numerical control machining process of aero - engine blades, and performing dimensionality reduction on the multi - modal data based on sparse principal component analysis and variational auto - encoder to obtain the pre - processed data set.

[0012] Preferably, the expression of the optimization objective function of the sparse principal component analysis is:

[0013] ;

[0014] where, is the projection matrix, is the sparse regularization parameter, is the original data matrix;

[0015] The calculation expression of the variational auto - encoder is:

[0016] ;

[0017] where, is the variational posterior distribution, is the latent variable, is the prior distribution, is the Kullback - Leibler divergence.

[0018] Preferably, the process of obtaining the tool wear diagnosis and stress distribution prediction results includes:

[0019] Align the data features of different wear states in the pre - processed data set based on the maximum mean discrepancy method to obtain an aligned data set;

[0020] Construct a tool wear diagnosis and stress distribution prediction model based on the self - attention time - series network, input the time - series data set into the tool wear diagnosis and stress distribution prediction model for calculation, and obtain the tool wear diagnosis and stress distribution prediction results.

[0021] Preferably, the expression of the maximum mean discrepancy method is:

[0022] ;

[0023] where, is the feature mapping mapped to the reproducing kernel Hilbert space, is the reproducing kernel Hilbert space, is the source - domain data distribution, is the target - domain data distribution, is a source domain sample, is a target domain sample.

[0024] Preferably, based on the tool wear diagnosis and stress distribution prediction results, the process of identifying the stress concentration area caused by tool wear and optimizing the laser shock peening strategy includes:

[0025] Constructing a stress concentration area dataset based on the tool wear diagnosis and stress distribution prediction results;

[0026] Calculating the correlation of the data in the stress concentration area dataset and constructing a stress concentration area identification model based on the correlation;

[0027] Calculating the change trend of each physical variable during the machining process based on the stress concentration area identification model, establishing a stress evolution model, and screening the areas with significant stress anomalies based on the stress evolution model;

[0028] Obtaining the laser shock peening area based on the area with significant stress anomalies and optimizing the laser shock parameters based on the distribution of the laser shock peening area;

[0029] Optimizing the laser shock peening strategy based on the optimized laser shock parameters.

[0030] Preferably, the process of optimizing the laser shock parameters based on the distribution of the laser shock peening area includes:

[0031] Setting different parameter combinations of the laser shock parameters, conducting experimental verification, and establishing a mapping relationship between the parameters and the influence on the surface residual stress;

[0032] Optimizing the parameters based on the mapping relationship, performing laser shock, recording the change of the plasma pressure in the shock area and the shock wave propagation characteristics, and analyzing the shock wave attenuation law under different laser parameters;

[0033] Verifying the distribution of the residual stress by using the finite element simulation method and correcting it in combination with the experimental measurement data to obtain the optimized laser shock peening process parameters.

[0034] Preferably, based on the feedback data, the process of optimizing the tool wear diagnosis and stress distribution prediction model by the incremental transfer learning method to obtain a multi-condition adaptation model further includes: during the model parameter update process, adopting a freeze-fine-tuning strategy, freezing some layers of the deep neural network, updating the parameters of specific layers, keeping the parameters of the low-level feature extraction layer unchanged to enable the low-level feature extraction layer to continue extracting general features; for the high-level decision-making layer, allowing parameter fine-tuning to enable the model to adapt to the new condition data.

[0035] Preferably, in the incremental transfer learning method, the expression of the optimization objective is:

[0036] ;

[0037] Wherein, is the loss of the source operating condition model, is the distribution deviation calculated by the maximum mean difference, and are the weight coefficients, are the parameters of the source operating condition training model, are the parameters of the target operating condition training model.

[0038] Compared with the prior art, the present invention has the following advantages and technical effects:

[0039] 1. Based on the influence of tool wear on stress distribution during the numerical control machining process of aeroengine blades, the present invention collects multi-modal data such as cutting force, vibration, temperature field, and stress distribution, and uses sparse principal component analysis (SPCA) and variational autoencoder (VAE) for dimensionality reduction and feature extraction to construct a tool wear diagnosis model. By calculating the change of tool wear rate and stress concentration area, stress anomalies during the machining process are accurately identified, and the feature distribution under different operating conditions is optimized by combining the maximum mean difference (MMD) to improve the adaptability of the model to complex operating conditions and achieve efficient and accurate stress anomaly detection.

[0040] 2. The present invention uses incremental transfer learning to construct a deep learning model to realize cross-operating condition knowledge transfer of different blade models, tool wear states, and impact parameters. The MMD method is used to align the features of the source domain and target domain data, and combined with the freeze-fine-tuning strategy, while maintaining the learning results of the original machining conditions, the learning ability of the model under new operating conditions is improved, the prediction accuracy and calculation efficiency are improved, and the generalization ability of intelligent monitoring under multiple operating conditions is ensured.

[0041] 3. Aiming at the stress concentration problem caused by tool wear, the present invention accurately identifies the strengthening area based on a data-driven method and optimizes the laser shock parameters to ensure uniform residual stress distribution. Combining plasma pressure monitoring and shock wave propagation analysis, the laser shock power, pulse width, and shock times are dynamically adjusted to improve the strengthening uniformity and stability. Through fatigue life testing, microstructure analysis, and surface roughness measurement, the strengthening effect is quality evaluated, and based on the feedback optimization model, the reliability and service stability of blade remanufacturing are further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] The drawings constituting a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application. In the drawings:

[0043] Figure 1Schematic diagram of the method process according to the embodiments of the present invention. Detailed implementation manners

[0044] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments may be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0045] It should be noted that the steps shown in the flowchart of the accompanying drawings may be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in an order different from that here.

[0046] Embodiment 1

[0047] As Figure 1 shown, in this embodiment, a laser shock remanufacturing method for aeroengine blades based on digital twin is provided, including:

[0048] S1. Collect multi-modal data during the numerical control machining process of aeroengine blades, and set the data set , where is the cutting force data, including the normal force, tangential force and feed force, and is monitored in real time by a force sensor; is the vibration signal data, and the dynamic contact characteristics of the tool-workpiece are recorded by a triaxial accelerometer to characterize the machining stability; is the temperature field data, and the surface temperature distribution of the blade and the characteristics of the heat affected zone (HAZ) are recorded by an infrared thermal imager; is the stress distribution data, and in combination with strain gauges and digital speckle pattern interferometry (DSPI), the evolution trend of the residual stress on the blade surface during the machining process is monitored.

[0049] Due to the high data dimension, the existence of redundant information and the uneven distribution of different modal data, sparse principal component analysis (SPCA) and variational autoencoder (VAE) are used for dimensionality reduction to extract the key features during the machining process, improving the data utilization rate and calculation efficiency. Let the original data matrix be , where is the number of samples, is the data dimension. SPCA introduces regularization constraints on the basis of traditional principal component analysis (PCA), and optimizes the objective function as follows:

[0050] ;

[0051] is the projection matrix, is the sparse regularization parameter, ensuring that the data after dimensionality reduction still retains the most discriminative feature information.

[0052] For the non - linear feature relationship, the VAE is further used for feature extraction. Let the input data be mapped to the latent variable through the encoder, which follows a certain prior distribution , and then reconstructed by the decoder into

[0053] . Minimize the evidence lower bound (ELBO):

[0054] where is the variational posterior distribution, is the Kullback - Leibler divergence. By jointly optimizing SPCA and VAE, the dimensionality - reduced features of different - modality data have local discriminative ability while maintaining the global distribution information, providing data support for subsequent tool wear diagnosis and stress distribution modeling.

[0055] S2. Based on the multi - modality data collected during the NC machining process, construct a tool wear diagnosis and stress distribution prediction model to identify the stress concentration areas caused by tool wear. Set the tool wear state data set , where is the cutting force data at different wear stages, is the corresponding vibration signal data, is the temperature field distribution on the tool surface during the wear process, is the residual stress measurement data. By analyzing the influence relationship of tool wear on different physical fields, construct a mapping model between tool wear states and stress distributions.

[0056] Since there are domain - to - domain biases in the data distributions at different tool wear stages, the maximum mean discrepancy (MMD) is used to align the data features of different wear states. Let the source - domain data distribution be , and the target - domain data distribution be . The formula for calculating the maximum mean discrepancy is as follows:

[0057] ;

[0058] where is the feature mapping mapped to the reproducing kernel Hilbert space, is the reproducing kernel Hilbert space, is the source - domain data distribution, is the target - domain data distribution, is the source - domain sample, is the target - domain sample. By minimizing the MMD to optimize data alignment, improve the cross - operating - condition prediction ability.

[0059] In terms of stress distribution prediction, a Self-Attention Temporal Network (SATN) is adopted to model the dynamic evolution of the stress concentration region during the wear process. The time series data of tool wear is set as , where is the temperature change, is the vibration signal, is the cutting force, is the stress distribution. The SATN model uses the self-attention mechanism to calculate the weighted feature representation at time step :

[0060] ;

[0061] where are the query, key and value matrices respectively, and is the scaling factor of the feature dimension. By introducing the self-attention mechanism, the model can capture the long-term influence of tool wear on the stress concentration region and improve the prediction accuracy of the stress distribution for the wear state.

[0062] Combining MMD transfer learning with SATN time series modeling, cross-condition tool wear diagnosis and stress distribution prediction are realized, providing a basis for identifying high-precision stress concentration regions for laser shock peening.

[0063] S3. Based on the tool wear diagnosis and stress distribution prediction results, identify the stress concentration regions caused by tool wear and optimize the laser shock peening strategy. Set the stress concentration region dataset , where is the residual stress distribution data, is the temperature field data, is the cutting force anomaly point, is the vibration anomaly signal. Adopt data-driven methods to analyze the high-stress regions on the blade surface under different tool wear states and screen the parts suitable for laser shock peening.

[0064] By calculating the correlations between tool wear characteristics, cutting force characteristics, temperature field characteristics and stress distribution, a stress concentration region identification model is constructed. Assume that there is a mapping relationship between the formation of the stress concentration region and these characteristic variables , where represents the distribution of the stress concentration region, and represents each machining characteristic variable. Normalization processing is adopted to keep the characteristics of different data modalities within the same numerical range and improve the stability of model prediction.

[0065] For the identified stress concentration regions, calculate the changing trends of various physical variables during the machining process, establish a stress evolution model, screen out the regions with significant stress anomalies, and label them as laser shock peening regions. Let the change amplitude of the stress anomaly region be When exceeds the set threshold, it is considered that there is a stress concentration phenomenon in this region, and it needs to be optimized by laser shock peening.

[0066] Combined with the experimental data, verify the identified laser shock peening regions to ensure that the selected regions conform to the machining stress evolution law. Finally, generate a distribution map of the blade stress concentration regions, providing a basis for the subsequent optimization of laser shock peening parameters.

[0067] S4. Implement laser shock peening on the stress concentration regions to improve the uniformity of the residual stress distribution on the blade surface and enhance the fatigue resistance. Set the laser shock remanufacturing data set where is the residual stress data on the blade surface after impact, is the plasma pressure distribution during the impact process, is the impact force signal, is the dynamic vibration characteristic of the impact region.

[0068] According to the distribution of the stress concentration regions, optimize the laser shock parameters, including laser power , spot diameter , pulse width and the number of impacts . Set different parameter combinations, conduct experimental verification, and establish a mapping relationship between the parameters and the influence on the surface residual stress. Let the residual stress and the laser shock parameters have a non-linear relationship:

[0069] ;

[0070] Fit this relationship through experimental data, optimize the laser shock parameters, make the distribution of the residual stress on the blade surface more uniform, and avoid new stress concentration phenomena.

[0071] During the laser shock process, use an on-line monitoring system to record the plasma pressure change and shock wave propagation characteristics in the impact region, and analyze the shock wave attenuation law under different laser parameters. Let the shock wave pressure gradient affect the microstructural changes of the material. When is lower than the critical value , it may lead to insufficient strengthening effect, and it is necessary to increase the laser power or adjust the pulse width to improve the impact energy transfer efficiency.

[0072] For different impact conditions, finite element simulation is used to verify the distribution of residual stress, and it is corrected in combination with experimental measurement data. Finally, optimized laser shock peening process parameters are obtained to ensure uniform distribution of residual stress on the blade surface and improve the remanufacturing quality.

[0073] S5. Based on the data feedback in the laser shock remanufacturing process, an incremental transfer learning strategy is adopted to optimize the model and improve its adaptability under different blade models, damage states and impact parameters. Set the model training data set , where is the data of the residual stress on the blade surface after impact, is the plasma pressure distribution, is the impact force signal, is the dynamic vibration characteristic of the impact area. Since there are data distribution deviations in the remanufacturing conditions of different blades, the directly trained model may perform poorly under new conditions. Therefore, incremental transfer learning is used for optimization to enable the model to maintain good generalization ability after new data is added.

[0074] The maximum mean discrepancy (MMD) is used to align the features of the new condition data to ensure that the data distribution of the source condition and the data distribution of the target condition are minimized. During the incremental learning process, define the model parameter trained under the source condition and the model parameter trained under the target condition. The optimization objective is as follows:

[0075] ;

[0076] Among them, is the loss of the source condition model, is the distribution deviation calculated by the maximum mean discrepancy, and are weight coefficients.

[0077] During the model parameter update process, a freeze - fine - tuning strategy is adopted, that is, some layers of the deep neural network are frozen, and only the parameters of specific layers are updated to avoid damage to the original model by new data and ensure stability during the incremental learning process. For the low - level feature extraction layer, keep its parameters unchanged to continue extracting general features; for the high - level decision - making layer, allow parameter fine - tuning to enable the model to adapt to new condition data and improve the prediction accuracy.

[0078] After the incremental transfer learning optimization, the model can maintain high - precision prediction ability under different laser shock remanufacturing conditions, adapt to the remanufacturing requirements of different blades, and ensure the processing stability and the control effect of residual stress.

[0079] After laser shock remanufacturing, quality assessment is carried out on the surface residual stress distribution, fatigue strength and microstructural changes of the blade, and the shock process parameters are optimized based on data feedback. Set the blade quality assessment data set , where is the residual stress data of the blade after remanufacturing, is the fatigue life measurement result, is the information on the change of the material microstructure, is the surface roughness feature.

[0080] For different blade regions, analyze the change of residual stress before and after laser shock peening, and calculate the stress improvement rate to evaluate the shock peening effect.

[0081] ;

[0082] When exceeds the set optimization threshold , it is considered that the strengthening effect meets the expectation; if not, the laser shock parameters need to be adjusted and the shock strategy is re-optimized.

[0083] Use high-cycle fatigue test (HCF) to measure the fatigue life of the blade, record the fatigue crack growth rate and analyze the crack initiation situation. Assume that the crack growth follows the Paris formula, and the fatigue life assessment is as follows:

[0084] ;

[0085] Among them, and are material constants, is the stress intensity factor range. If the fatigue life is lower than the set threshold, it indicates that the laser shock peening is insufficient or the parameters are unreasonable, and the shock power or shock coverage needs to be adjusted.

[0086] Measure the surface quality, and use white light interferometry to calculate the surface roughness , ensuring that the surface roughness after laser shock meets the aviation blade remanufacturing standard. For the data feedback under different processing conditions, construct a process parameter adaptive optimization model to dynamically adjust the laser shock strategy to ensure the quality stability of remanufacturing under different blade models and damage states. Finally, form a blade remanufacturing quality assessment system to improve the reliability and service life of the blade.

[0087] Furthermore, the multi-modal data acquisition in step S1 specifically includes the following:

[0088] In the numerical control machining process of aero-engine blades, multi-modal data such as cutting force, vibration, temperature field, and stress distribution are collected to provide an optimization basis for tool wear diagnosis (S2), stress concentration area identification (S3), and laser shock peening (S4). Since the distribution characteristics of different data modalities are different, directly using the original data may affect the stability of the model. Therefore, after data collection, sparse principal component analysis (SPCA) is used for data dimensionality reduction to extract key features from high-dimensional data. At the same time, a variational autoencoder (VAE) is used to further enhance the expression ability of non-linear features, reduce data redundancy, and improve data quality. The dimensionality-reduced data is expressed as:

[0089] ;

[0090] where, is the original data matrix, is the data feature after dimensionality reduction. The dimensionality reduction result is input into the tool wear diagnosis model (S2) to ensure the stability and generalization ability of subsequent predictions.

[0091] Furthermore, the tool wear diagnosis and stress distribution prediction in step S2 specifically include the following:

[0092] Based on the data collected in S1, a tool wear diagnosis model is constructed to predict the stress concentration area caused by wear. Tool wear is a dynamic evolution process. As the machining time progresses, the wear degree gradually increases, thereby affecting the stress distribution. The tool wear rate is calculated as follows:

[0093] ;

[0094] where, is the current tool wear amount, is the time interval. This rate is used to judge the current state of the tool and, combined with stress data, predict the change trend of the high-stress area. Since the data distribution varies under different machining conditions, in S2, the maximum mean discrepancy (MMD) is used for feature alignment to enable the model to adapt to the machining conditions of different blades. The MMD is calculated as follows:

[0095] ;

[0096] where, is the feature mapping mapped to the reproducing kernel Hilbert space, is the reproducing kernel Hilbert space, is the source domain data distribution, is the target domain data distribution, is the source domain sample, is the target domain sample. By minimizing the MMD to optimize data alignment, the cross-condition prediction ability is improved.

[0097] Further, the identification of the stress concentration region in step S3 specifically includes the following:

[0098] Based on the predicted tool wear state and stress distribution in S2, accurately identify the high stress concentration region to provide an optimization strategy for subsequent laser shock peening (S4). The stress change rate ( ) is calculated as follows:

[0099] ;

[0100] where and respectively represent the residual stress values at the current moment and the previous moment. When exceeds the set threshold, the region is marked as a high stress concentration region and is included in the laser shock optimization range. In addition, based on data-driven, classify and analyze the change trends of cutting force, temperature field, and vibration signals during the machining process to ensure that the laser shock peening strategy in S4 can accurately act on the region that most needs optimization, rather than blindly applying laser shocks.

[0101] Further, the laser shock peening in step S4 specifically includes the following:

[0102] Based on the high stress region identified in S3, determine the specific location of laser shock peening and optimize the shock parameters. The key parameters of laser shock peening include laser power ( ), spot diameter ( ), pulse width ( ), and number of shocks ( ). The relationship between the residual stress and these parameters is as follows:

[0103] ;

[0104] Optimize this relationship through experimental data and finite element simulation to make the stress distribution on the surface of the blade after strengthening more uniform and improve the fatigue resistance. During the laser shock process, real-time monitor the plasma pressure and shock wave propagation characteristics to ensure the stability of laser shock peening. If it is found that the shock wave energy decays rapidly, it may be necessary to adjust the shock power or coverage rate to improve the strengthening effect. In addition, establish a closed-loop optimization control system to monitor the stress distribution on the surface of the strengthened blade and compare it with the prediction result in S3. If the strengthening effect does not meet the optimization requirements, automatically adjust the laser shock parameters for secondary optimization to make the improvement rate of residual stress reach the optimal.

[0105] Further, the incremental transfer learning optimization in step S5 specifically includes the following:

[0106] Under the conditions of different blade models, damage states, and changes in impact parameters, an incremental transfer learning strategy is adopted to enable the model to continuously optimize and improve the prediction accuracy. The optimization objectives are as follows:

[0107] ;

[0108] Among them, Calculate the difference in feature distributions between the source working condition and the target working condition, Control the adjustment range of the model parameters to ensure the stability of the model while adapting to new data. To prevent the model from forgetting the knowledge it has learned, during the training process, a freeze-fine-tuning strategy is adopted, that is, the low-level feature extraction layer is frozen, and only the parameters of the high-level decision-making layer are updated to improve the adaptability to new working conditions.

[0109] Furthermore, the blade quality evaluation and feedback optimization in step S6 specifically include the following:

[0110] After the laser shock remanufacturing is completed, to ensure the strengthening effect, it is necessary to conduct residual stress evaluation on the blade surface, fatigue life test, and microstructure analysis, and optimize the shock process parameters based on the measurement data. The stress improvement rate is calculated as follows:

[0111] ;

[0112] When is lower than the optimization threshold adjust the laser shock parameters for secondary optimization. In addition, a high-cycle fatigue test (HCF) is used to evaluate the fatigue life of the blade, and its reliability is analyzed in combination with the crack propagation rate. Based on the feedback data, a data-driven optimization model is constructed, and the feedback data of S6 is used to optimize the incremental transfer learning process of S5, enabling the model to continuously adapt to new processing conditions and improve the quality stability of blade remanufacturing.

[0113] The above is only a preferred specific implementation manner of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A laser shock remanufacturing method for aircraft engine blades based on digital twin, characterized in that: include: Collecting multimodal data during the numerical control machining of aircraft engine blades, preprocessing the multimodal data to obtain a preprocessed data set; Constructing a tool wear diagnosis and stress distribution prediction model, inputting the preprocessed data set into the tool wear diagnosis and stress distribution prediction model, and obtaining tool wear diagnosis and stress distribution prediction results; Based on the tool wear diagnosis and stress distribution prediction results, the stress concentration area caused by tool wear is identified and the laser shock strengthening strategy is optimized; Based on the optimized laser shock peening strategy, laser shock peening is performed on the stress concentration area to obtain feedback data; Based on the feedback data, the tool wear diagnosis and stress distribution prediction model is optimized by an incremental transfer learning method to obtain a multi-condition adaptation model; Predicting the remanufacturing of different blades based on the multi-condition adaptation model; The process of obtaining tool wear diagnosis and stress distribution prediction results includes: Based on the maximum mean difference method, the data features of different wear states in the preprocessed data set are distributed and aligned to obtain an aligned data set; Constructing a tool wear diagnosis and stress distribution prediction model based on a self-attention temporal network, inputting the aligned data set into the tool wear diagnosis and stress distribution prediction model for calculation, and obtaining the tool wear diagnosis and stress distribution prediction results; Based on the tool wear diagnosis and stress distribution prediction results, the process of identifying the stress concentration area caused by tool wear and optimizing the laser shock peening strategy includes: Constructing a stress concentration area data set based on the tool wear diagnosis and stress distribution prediction results; Calculating the correlation of data in the stress concentration area data set, and constructing a stress concentration area identification model based on the correlation; Calculating the variation trend of various physical variables during the processing based on the stress concentration area identification model, establishing a stress evolution model, and screening areas with significant stress anomalies based on the stress evolution model; Acquire a laser shock peening region based on the region with significant stress anomaly, and optimize laser shock parameters based on the distribution of the laser shock peening region; The laser shock processing strategy is optimized based on the optimized laser shock parameters.

2. The method according to claim 1, characterized in that The process of obtaining the preprocessed data set includes: collecting multimodal data during the numerical control machining of aircraft engine blades, reducing the dimension of the multimodal data based on sparse principal component analysis and variational autoencoder, and obtaining the preprocessed data set.

3. The method according to claim 2, characterized in that The expression of the optimization objective function of the sparse principal component analysis is: ； in, is the projection matrix, is the sparse regularization parameter, is the original data matrix; The calculation expression of the variational autoencoder is: ； in, is the variational posterior distribution, is a hidden variable, is the prior distribution, is the Kullback-Leibler divergence.

4. The method according to claim 1, characterized in that: The expression of the maximum mean difference method is: ； in, is the feature map mapped to the reproducing kernel Hilbert space, is the reproducing kernel Hilbert space, is the source domain data distribution, is the target domain data distribution, is the source domain sample, is the target domain sample.

5. The method according to claim 1, characterized in that The process of optimizing the laser shock parameters based on the distribution of the laser shock peening area includes: Set different combinations of laser shock parameters, conduct experimental verification, and establish a mapping relationship between the effects of parameters on surface residual stress; Optimize parameters based on the mapping relationship, perform laser shock, record the plasma pressure change and shock wave propagation characteristics in the shock area, and analyze the shock wave attenuation law under different laser parameters; The finite element simulation method is used to verify the distribution of residual stress, and corrections are made based on experimental measurement data to obtain optimized laser shock peening process parameters.

6. The method according to claim 1, characterized in that Based on the feedback data, the tool wear diagnosis and stress distribution prediction model is optimized by an incremental transfer learning method, and the process of obtaining a multi-working condition adaptation model also includes: in the process of updating the model parameters, a freeze-fine-tuning strategy is adopted to freeze some layers of the deep neural network, update the parameters of specific layers, and keep the parameters of the low-level feature extraction layer unchanged so that the low-level feature extraction layer continues to extract common features; for the high-level decision layer, parameter fine-tuning is allowed to adapt the model to the new working condition data.

7. The method according to claim 4, characterized in that In the incremental transfer learning method, the expression of the optimization objective is: ； in, is the source condition model loss, The deviation of the distribution calculated as the maximum mean difference, and is the weight coefficient, Train the model parameters for the source condition, Train the model parameters for the target operating conditions.

Images