Digital twin driven data augmentation for landing gear laser shock peening monitoring method

Abstract

The application discloses a kind of digital twin driven data enhancement aircraft landing gear laser shock peening monitoring method, belong to the field of aviation structural parts remanufacturing, intelligent manufacturing and data-driven regulation and control field, including: for the alternating load in the service process of landing gear, establish fatigue evaluation system, identify high stress concentration area, build multi-modal monitoring system, use data enhancement method to expand impact data, combined with finite element analysis to build impact strengthening physical model, simulate shock wave propagation and residual stress evolution, and combined with experimental data to correct model, realize data-driven and mechanism modeling collaborative optimization.Based on deep learning, evaluate the impact quality, analyze the impact uniformity by fusing multi-modal data, and identify the abnormal area of strengthening. Use reinforcement learning to optimize the impact process parameters, realize intelligent feedback control. This method combines multi-modal monitoring, data enhancement, physical modeling and intelligent optimization, improves the stability and fatigue life of laser shock peening.

Description

Technical Field

[0001] This invention belongs to the fields of remanufacturing of aerospace structural components, intelligent manufacturing and data-driven control, and particularly relates to a data-enhanced laser shock strengthening monitoring method for aerospace landing gear driven by digital twins. Background Technology

[0002] As a crucial load-bearing structure of aircraft, landing gear endures complex alternating loads over extended periods. Its fatigue damage and surface stress distribution directly impact service safety and lifespan. Traditional landing gear strengthening primarily relies on methods such as shot peening, but these methods limit the depth of the strengthened layer and result in rapid residual stress decay, failing to meet long-term service requirements. Laser shock peening (LSP) technology utilizes high-energy laser pulses to generate high-amplitude shock waves on the metal surface, increasing the residual stress level and optimizing the microstructure, thereby significantly improving fatigue life. However, parameters such as impact energy, number of impacts, and laser spot diameter significantly influence the strengthening effect during LSP. Accurately monitoring impact quality and optimizing process parameters remains a significant technical challenge in the field of reinforced remanufacturing.

[0003] Currently, laser shock strengthening quality assessment mainly relies on post-experiment testing, lacking online monitoring methods and making it difficult to obtain real-time information on stress wave propagation, vibration response, and temperature changes during the shock process. Furthermore, optimization of strengthening parameters under different operating conditions still depends on empirical control, making adaptive optimization difficult. With the development of intelligent manufacturing and data-driven technologies, multimodal sensor monitoring, deep learning modeling, and reinforcement learning optimization have become important directions for improving shock strengthening quality. By combining generative data augmentation, physical mechanism modeling, and intelligent optimization strategies, the accuracy of shock quality assessment can be effectively improved, and intelligent adaptive control of the strengthening process can be achieved, providing technical support for the remanufacturing and service life extension of aircraft landing gear. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a digital twin-driven data-enhanced method for monitoring laser shock reinforcement of aircraft landing gear, comprising:

[0005] Acquire experimental data during the service life of the landing gear, and identify high stress concentration areas based on the experimental data;

[0006] A multimodal monitoring system is constructed, and data is collected on the laser shock process in the high stress concentration area based on the multimodal monitoring system to obtain laser shock data samples;

[0007] The laser shock data sample is expanded using a generative data augmentation method to obtain an expanded dataset;

[0008] A finite element mechanism model is constructed, and numerical simulation analysis is performed based on the finite element mechanism model to obtain numerical simulation data. The extended dataset and the numerical simulation data are then fused to obtain a fused dataset.

[0009] An impact quality assessment model based on a spatiotemporal attention map neural network is constructed. The fused dataset is input into the impact quality assessment model for calculation to obtain the impact quality assessment result.

[0010] Based on the impact quality assessment results and deep reinforcement learning methods, an intelligent feedback optimization model is constructed, and the laser shock process parameters are dynamically adjusted based on the intelligent feedback optimization model.

[0011] Preferably, the process of acquiring experimental data during the landing gear's service life includes:

[0012] Multi-channel force sensors are used to monitor load evolution during the service life of the landing gear;

[0013] By combining strain gauge measurements of strain distribution under different working conditions, the material response characteristics under load are analyzed.

[0014] The experimental data were generated by obtaining residual stress data of key parts using X-ray diffraction technology and monitoring thermal accumulation effect using infrared thermal imaging.

[0015] Preferably, the process of identifying high stress concentration areas based on the experimental data includes: processing complex alternating loads using the finite element method to obtain an elastoplastic mechanical model; simulating stress distribution based on the elastoplastic mechanical model; evaluating fatigue life using the Miner linear cumulative damage method; and identifying high stress concentration areas.

[0016] Preferably, the process of obtaining laser shock data samples includes:

[0017] A stress wave sensor was used to record the transient stress waveform induced by laser shock;

[0018] Vibration signals during the impact process were collected using a triaxial accelerometer, and the dynamic response of the impact region was analyzed.

[0019] The dynamic evolution of the heat-affected zone is obtained by monitoring temperature changes in the impact area using an infrared thermal imager.

[0020] The plasma radiation characteristics generated during laser shock are analyzed using a spectral sensor.

[0021] The laser shock data sample is obtained by performing multimodal data alignment on the transient stress waveform, the dynamic response of the impact region, the dynamic evolution of the heat-affected zone, and the plasma radiation characteristics based on the dynamic time warping method.

[0022] Preferably, the process of obtaining the extended dataset includes:

[0023] A generative adversarial network is used to augment the laser shock data samples to obtain a first dataset.

[0024] A second dataset is obtained by using a temporal generative adversarial network to enhance the impact process data in the laser impact data sample that conforms to time series characteristics;

[0025] A third dataset is obtained by probabilistically modeling the laser shock data samples based on a variational autoencoder.

[0026] A fourth dataset is obtained by generating cross-modal data from the laser shock data samples based on a cross-modal transformation network.

[0027] The first, second, third, and fourth datasets are filtered and enhanced to obtain the extended dataset.

[0028] Preferably, the process of constructing a finite element mechanism model and performing numerical simulation analysis based on the finite element mechanism model to obtain numerical simulation data includes:

[0029] A finite element analysis model of laser shock was established based on the theory of elastoplastic dynamics to simulate the transient stress changes and shock wave propagation characteristics inside the material during the impact process.

[0030] The Johnson-Cook constitutive model is used to describe the strain rate-dependent plastic behavior of the material and to simulate the evolution of residual stress under different laser shock energies;

[0031] The residual stress distribution under different working conditions is obtained through laser shock experiments. The residual stress distribution is compared with the transient stress change, shock wave propagation characteristics and residual stress evolution. The constitutive parameters are adjusted by the least squares optimization method to minimize the error between the numerically simulated stress distribution and the experimentally measured residual stress, thereby obtaining the numerical simulation data.

[0032] Preferably, the expression for calculating the impact characteristics of each monitoring point using the impact quality assessment model is as follows:

[0033] ;

[0034] in, Represents the impact characteristics of the l-th layer. For nodes The neighborhood set, and For trainable weights, It is a non-linear activation function.

[0035] Preferably, the attention weight expression of the impact quality assessment model is:

[0036] ;

[0037] in, For time-related scoring functions, This represents the attention weight.

[0038] Preferably, the intelligent feedback optimization model is generated based on a deep Q-network;

[0039] The objective function of the intelligent feedback optimization model is:

[0040] ;

[0041] in, As a discount factor, The current laser shock parameter adjustment strategy, For the reward function, This represents the current system status, including laser shock parameters, workpiece status, and monitoring data.

[0042] Compared with the prior art, the present invention has the following advantages and technical effects:

[0043] 1. This invention constructs a multimodal data acquisition and fatigue assessment system based on the complex alternating loads experienced by aircraft landing gear during service. Using force sensors, strain gauges, and X-ray diffraction (XRD) technology, the stress distribution of the landing gear under different service conditions is accurately monitored, and the evolution trend of high stress concentration areas and residual stress is calculated using the finite element method (FEM). The Miner linear cumulative damage method is employed to assess fatigue life, identify potential fatigue failure areas, and determine the target areas for laser shock stabilization.

[0044] 2. This invention combines a multimodal monitoring system with data augmentation technology to precisely control the laser shock process and ensure the stability of the strengthening effect. By integrating multiple sensors such as stress wave sensing, vibration analysis, infrared thermal imaging, and spectral detection, the dynamic mechanical response of the impact region is monitored in real time, and the dynamic time warping (DTW) method is used to align the multimodal data in time. Furthermore, generative adversarial networks (GANs) and time generative adversarial networks (TimeGANs) are used to expand the impact dataset, enhancing the diversity and physical consistency of the impact data, and improving the training quality of the deep learning model and the accuracy of finite element simulation.

[0045] 3. This invention employs an intelligent optimization method to improve the quality of laser shock strengthening. It utilizes a spatiotemporal attention graph neural network (STAGNN) to fuse multimodal data such as stress waves, vibrations, and temperature fields to accurately assess strengthening quality and identify potential shock anomaly areas. Combined with deep reinforcement learning (DRL), it optimizes shock parameters based on real-time monitoring data to ensure uniform residual stress distribution and intelligently adjusts shock energy, spot diameter, and the number of shocks to improve the consistency of strengthening. Attached Figure Description

[0046] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0047] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention. Detailed Implementation

[0048] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0049] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0050] Example 1

[0051] like Figure 1 As shown, this embodiment provides a digital twin-driven data-enhanced method for monitoring laser shock reinforcement of aircraft landing gear, including:

[0052] S1. To address the complex alternating loads experienced by aircraft landing gear during takeoff, landing, and taxiing, a fatigue assessment system was established to identify high-stress concentration areas, providing a target area basis for subsequent laser shock peening. Multi-channel force sensors were used to monitor the load evolution during landing gear service, including axial, tangential, and normal forces. Strain gauges were combined to measure strain distribution under different operating conditions, analyzing the material response characteristics under load. Residual stress data for key components were obtained using X-ray diffraction (XRD) technology, and infrared thermography was employed to monitor the thermal accumulation effect, assessing the impact of alternating stress on the material microstructure.

[0053] Based on experimental data, an elastoplastic mechanical model of the landing gear was constructed using the finite element method (FEM). Three-dimensional solid elements were used for mesh generation, and load boundary conditions were set for different service conditions. The stress distribution function of the landing gear under service conditions is assumed to be... It satisfies the momentum conservation equation:

[0054]

[0055] in, For stress tensor, For external forces, For material density, For the velocity field. The equivalent stress distribution is calculated based on the Von Mises criterion:

[0056]

[0057] in, , , The principal stress components are identified. For regions with high stress concentration, a local adaptive mesh refinement method is employed to improve the computational accuracy in areas with large stress gradient variations. The numerical model is then corrected using experimental measurement data.

[0058] Based on the fatigue damage accumulation theory, the fatigue life of the landing gear is calculated using the Miner linear cumulative damage method, and the damage factor is defined as follows:

[0059]

[0060] in, The number of cycles at a given stress level. This corresponds to the fatigue life. When... At this point, the material enters a state of fatigue failure. The fatigue limit is calculated using the Goodman modified criterion:

[0061]

[0062] in, The fatigue limit, For fatigue strength under symmetrical cyclic loading, For average stress, The ultimate strength of the material was determined. Based on the stress distribution results calculated by finite element method, the high fatigue damage region was analyzed in combination with experimental data, and further verified by the surface residual stress distribution data. Finally, the high stress concentration area on the landing gear surface was identified as the target area for laser shock peening, providing a basis for subsequent optimization of impact parameters.

[0063] S2. To comprehensively monitor stress wave propagation, vibration response, temperature changes, and plasma characteristics during laser shock peening, a multi-modal monitoring system is constructed to ensure that the strengthening effect in the impact area can be quantitatively evaluated. This system integrates multiple measurement methods such as stress wave sensing, vibration sensing, infrared thermal imaging, and spectral analysis, combined with data synchronization and feature fusion technologies, to achieve high-precision online monitoring of the impact process.

[0064] First, a stress wave sensor is used to record the transient stress waveform induced by laser shock, including the peak impact pressure, rise time, and attenuation characteristics, to assess the energy transfer efficiency of the shock. The propagation of the stress wave satisfies the wave equation:

[0065]

[0066] in, For shock wave pressure, The speed of sound is used. By analyzing the time-frequency characteristics of the shock wave, the uniformity of the impact intensity and the energy absorption can be determined.

[0067] Secondly, a triaxial accelerometer was used to collect vibration signals during the impact process, and the dynamic response of the impact region was analyzed. The vibration data was converted to the frequency domain using Fast Fourier Transform (FFT) to extract the main vibration modes.

[0068]

[0069] in, Representing the vibration amplitude at different frequencies, combined with the dominant frequency characteristics, it can be used to determine the transmission and attenuation of impact energy within the material.

[0070] Furthermore, infrared thermal imagers were used to monitor temperature changes in the impact area, capturing the dynamic evolution of the heat-affected zone (HAZ). During laser shock, the local temperature rises, and its heat conduction is affected by the thermal conductivity of the material. and specific heat capacity The influence and temperature distribution satisfy:

[0071]

[0072] in, This is the heat source term for laser input. Temperature changes in the impact area are recorded using high-frame-rate thermal imaging to ensure that excessive heat is not generated during the strengthening process, thus avoiding material performance degradation.

[0073] Finally, a spectral sensor was used to analyze the plasma radiation characteristics generated during the laser shock process. By monitoring the emission spectrum of the plasma, the coupling efficiency of the laser energy and the shock stability were evaluated.

[0074] Because the time bases of the various sensors differ, Dynamic Time Warping (DTW) is employed to align multimodal data, ensuring consistency in data timing. Finally, through multimodal feature fusion, a comprehensive monitoring framework for the impact area is constructed, providing data support for the accurate assessment of the enhancement effect.

[0075] S3. Due to the influence of multiple factors such as laser energy, number of impacts, spot diameter, and material properties during laser shock peening, the actual data samples collected are limited, and the data distribution has complex nonlinear characteristics. Directly training the monitoring model may lead to insufficient generalization ability. Therefore, a generative data augmentation method is introduced to expand the multimodal dataset of the shock process and improve the adaptability and robustness of the monitoring model under different working conditions.

[0076] Generative Adversarial Networks (GANs) are used to enhance the shock stress wave, vibration response, temperature field, and plasma spectral data to simulate the data distribution under different shock parameters and operating conditions. The real dataset is assumed to be:

[0077]

[0078] in, Represents stress wave data, Represents vibration data, Represents temperature data. Represents plasma spectral data. GAN consists of a generator. and discriminator The composition and optimization objective are:

[0079]

[0080] in, The distribution representing the actual shock data, The data is distributed as random noise. Through adversarial training, the generator is continuously optimized, and the generated data gradually approximates the true distribution, thereby expanding the dataset and improving the monitoring model's adaptability to different laser shock parameters.

[0081] During laser shock peening, the physical field data change dynamically over time. Simply generating independent data points based on GANs may overlook this temporal dependence. Therefore, a Time Generative Adversarial Network (TimeGAN) is used to generate shock process data that conforms to time series characteristics. Let the original time series data be:

[0082]

[0083] TimeGAN combines the structure of Generative Adversarial Networks (GANs) and Recurrent Neural Networks (RNNs) to map time series data to a latent variable space through an encoder-decoder architecture. Furthermore, time dependency modeling is performed in the latent variable space to optimize the objective function:

[0084]

[0085] in, The Kullback-Leibler divergence is used to ensure consistency between the generated data and the distribution of real data. TimeGAN is trained to enhance the dynamic characteristics of the impact process, enabling the monitoring system to adapt to different laser impact intensities and operating conditions, thereby improving prediction accuracy.

[0086] Directly using data-driven GANs to generate data may result in data that does not conform to the physical laws of impact. Therefore, physics-informed learning is introduced during data augmentation to ensure that the generated data conforms to the physical model of laser shock. For example, stress waves satisfy the one-dimensional wave equation:

[0087]

[0088] in, For shock wave pressure, Let be the stress wave propagation velocity of the material. This equation is used as a regularization constraint for the generator, ensuring that the data generated by the GAN conforms to the energy transfer laws of shock waves. Furthermore, the temperature field data satisfies the heat conduction equation:

[0089]

[0090] in, For specific heat capacity, Thermal conductivity, The heat source term is input for laser shock. Physical constraints are applied to the TimeGAN-generated data to ensure it conforms to the actual distribution of the heat-affected zone (HAZ), thus improving data reliability.

[0091] To address the nonlinear distribution of impact data, a variational autoencoder (VAE) is employed to probabilistically model the impact data, enhancing its diversity. Let the input data be... The VAE is mapped to a latent variable ZZ through the encoder, which follows a certain prior distribution. Then the decoder reconstructs it into Optimized lower bound of evidence (ELBO):

[0092]

[0093] in, For variational posterior distribution, The Kullback-Leibler divergence is used to generate shock data that conforms to the true distribution through VAE, and this data is then used for data augmentation to improve the generalization ability of the monitoring model.

[0094] In laser shock peening, complex nonlinear coupling relationships exist between various physical quantities (stress wave, vibration, temperature, and plasma spectrum). A Cross-Modal Translation Network (CMTN) is employed to learn the mapping relationships between different modal data. For example, under finite stress wave data, the CMTN predicts the corresponding temperature field data:

[0095]

[0096] Simultaneously, CycleGAN is used for cross-modal data generation to achieve the mapping from plasma spectral data to stress wave data:

[0097]

[0098] Here, GG and FF are two competing generators, and the loss is based on cycle consistency:

[0099]

[0100] Ensure cross-modal consistency of generated data, improve the joint modeling capability between different physical quantities, and enhance the integrity and consistency of impact process data.

[0101] Data generated by GAN, TimeGAN, PI-GAN, and VAE were combined, and data filtering and augmentation strategies were employed to select the dataset that best conforms to physical laws and covers different working conditions. Let the original data be... The generated data is Define the data augmentation loss:

[0102]

[0103] in, , To regularize the parameters, the data distribution was optimized to conform to actual working conditions. Finally, the enhanced data was used to train the laser shock reinforcement monitoring model, improving the prediction accuracy for different impact parameters, material types, and fatigue damage states, ensuring the stability and high adaptability of the monitoring system, and providing high-quality data support for subsequent reinforcement effect evaluation.

[0104] S4. During laser shock peening, the stress evolution within the material is influenced by multiple factors, including laser energy, number of shocks, spot diameter, and material properties. While experimental data can provide some information about the shock process, its distribution is limited and cannot directly analyze the internal stress state. Therefore, a finite element mechanism model is introduced to analyze shock wave propagation, residual stress redistribution, and material hardening effects through numerical simulation. This model is then modified based on experimental data to improve the physical consistency of the data enhancement and achieve synergistic optimization between data-driven and physical modeling.

[0105] First, a finite element analysis (FEA) model of laser shock is established based on elastoplastic dynamics theory to simulate the transient stress changes and shock wave propagation characteristics within the material during the impact process. It is assumed that the material obeys isotropic constitutive relations, and its dynamic stress distribution satisfies the momentum conservation equation:

[0106]

[0107] in, For stress tensor, For impact force, For material density, Let be the velocity vector. The transient propagation of the shock wave is described by the wave equation:

[0108]

[0109] in, To withstand pressure, The propagation velocity of the shock wave in the material is represented by the value denoted as . An adaptive mesh refinement technique is employed to improve the computational accuracy in high-gradient stress regions, ensuring accurate simulation of the residual stress distribution.

[0110] Under impact, a plastic deformation layer forms on the material surface, and the redistribution of residual stress is mainly influenced by the strengthening and strain hardening effects of the material. The strain rate-dependent plastic behavior of the material is described using the Johnson-Cook constitutive model, and its flow stress expression is as follows:

[0111]

[0112] Where A, B, C, n, m are material parameters. For equivalent plastic strain, To normalize the strain rate, The normalized temperature is used. By simulating the evolution of residual stress under different laser shock energies, the depth of the strengthening layer and the stress gradient distribution are analyzed to optimize the selection of shock parameters.

[0113] Experimental data were used to correct the finite element method (FEM) calculation results and improve the physical consistency of the model. First, residual stress distributions under different working conditions were obtained through laser shock experiments and compared with the stress results from numerical simulations. The least squares optimization method was then used to adjust the constitutive parameters to match the stress distribution obtained from the numerical simulations. Residual stress as measured experimentally The error between them is minimized:

[0114] Here, θ\theta represents the material parameter set of the finite element model, including strain hardening parameters, dynamic strengthening coefficients, and thermal softening coefficients. Through iterative optimization, it is ensured that the simulation data accurately reflects the stress evolution characteristics during the actual impact process, enabling the finite element calculation results to be used for data augmentation and improve the prediction accuracy of the data-driven model.

[0115] Ultimately, by synergistically enhancing the finite element mechanism model, the experimental data and numerical simulation data are integrated, supplementing the deficiencies of the experimental data. This makes the data augmentation method not only dependent on the generative model but also constrained by the physical mechanism, thereby improving the reliability and adaptability of the impact strengthening monitoring system.

[0116] S5. The quality of laser shock peening directly affects the fatigue life and service stability of the landing gear. Therefore, a comprehensive assessment of the stress distribution, strengthening layer depth, and impact uniformity after impact is required, along with anomaly detection technology to identify areas of insufficient strengthening or uneven impact. Since impact quality involves multimodal data such as stress waves, vibration signals, temperature fields, and plasma spectra, a data-driven deep learning method is used to construct a comprehensive evaluation model to improve the stability and robustness of the monitoring system.

[0117] To comprehensively analyze the dynamic changes during the impact process, an impact quality assessment model based on a spatio-temporal attention graph neural network (STAGNN) is constructed. This model combines the spatial information extraction capability of graph neural networks (GNNs) with the dynamic feature capture capability of the attention mechanism to improve the modeling ability of impact quality. Let the monitoring dataset within the impact area be:

[0118]

[0119] in, Represents stress wave data, Represents vibration data, Represents temperature data. This represents plasma spectral data. First, a graph neural network is used to establish the spatial topology of the impact region, and the graph formed by the monitoring points is defined. ,in For the set of monitoring points, This represents the dynamic relationship between physical quantities. The impact characteristics of each monitoring point are calculated using graph convolution.

[0120]

[0121] in, Represents the impact characteristics of the l-th layer. For nodes The neighborhood set, and For trainable weights, It is a non-linear activation function.

[0122] In the time dimension, an attention mechanism is introduced to calculate the importance of different time steps, ensuring that the model focuses on feature changes at critical moments. The attention weights are defined as follows:

[0123]

[0124] in, A time-dependent scoring function is used, and a learnable feedforward network is employed to calculate the feature importance at different time steps. Finally, by combining the spatial features of the graph neural network with the temporal dynamics information of the attention mechanism, the reinforcement quality of the impact region is predicted, and the impact uniformity is evaluated.

[0125] In the anomaly detection section, STAGNN is used to calculate the residual stress distribution in the impact region, and the result is compared with experimental measurement data to calculate the impact uniformity index.

[0126]

[0127] in, The average residual stress in the impact region, To determine the number of measurement points. When When the threshold is exceeded, it is considered that there is uneven impact in the area.

[0128] Finally, a complete impact quality assessment system was constructed based on STAGNN to achieve the fusion analysis of multimodal data, improve the accuracy of impact monitoring, provide data support for optimizing laser shock strengthening process parameters, and ensure the uniformity of impact and the stability of strengthening effect.

[0129] S6. The process parameters of laser shock peening directly affect the residual stress distribution and fatigue performance of the landing gear surface. Therefore, it is necessary to combine the impact quality assessment results to optimize process parameters such as laser energy, spot diameter, pulse width, and number of impacts, and construct an adaptive optimization model to improve the impact peening effect. Since the peening process involves complex nonlinear dynamic mechanisms, traditional experience-based parameter control methods are difficult to adapt to different working conditions. Therefore, Deep Reinforcement Learning (DRL) is used to construct an intelligent feedback optimization system, enabling process parameters to be dynamically adjusted based on historical peening data to achieve the optimal laser shock peening strategy.

[0130] The parameter space for laser shock peening is set as follows:

[0131]

[0132] in, Laser energy, The diameter of the light spot. The pulse width. The number of impacts. The uniformity of residual stress in the impact region. Depth of reinforcement layer and fatigue life improvement rate As the optimization objective of reinforcement learning, a mapping relationship between process parameters and reinforcement quality is constructed. A Deep Q-Network (DQN) is used for parameter optimization, defining the state... ,action and the reward function RR. Let the state of the impact process be:

[0133]

[0134] in, , , , Represents current impact monitoring data. These are the current process parameters. The reinforcement learning strategy is represented by the Q-function, and the optimization objective is:

[0135]

[0136] in, As a discount factor, The current laser shock parameter adjustment strategy, Let this be the reward function. The reward function is defined as:

[0137]

[0138] in, The rate of improvement of residual stress after impact. For stress uniformity, For temperature changes in the heat-affected zone, , , These are the weight parameters.

[0139] An ExperienceReplay mechanism was used to store monitoring data under different impact strategies, and the DQN model was trained to learn the optimal parameter adjustment strategy under different impact conditions. After multiple reinforcement training iterations, the model can automatically adjust parameters such as laser energy and number of impacts to optimize the uniformity of residual stress, avoid local stress concentration, and improve the quality of the reinforcement layer.

[0140] Ultimately, an intelligent feedback optimization system based on reinforcement learning enables adaptive control of the laser shock strengthening process, improving shock stability and optimization efficiency. This ensures optimal strengthening effects for different landing gear models, materials, and service conditions, thereby improving the fatigue life and service reliability of the landing gear.

[0141] Furthermore, the landing gear service load analysis and fatigue assessment in step S1 specifically includes the following methods:

[0142] During the service life of aircraft landing gear, the complex alternating loads it endures lead to localized high stress concentration areas. To optimize laser shock peening (S4), service load analysis and fatigue assessment are necessary. Multi-channel force sensors are used to collect axial, tangential, and normal force data. Strain gauges are used to measure strain distribution under different operating conditions, and X-ray diffraction (XRD) is employed to obtain residual stress in key areas. An elastoplastic mechanical model is constructed based on the finite element method (FEM) to simulate the stress distribution of the landing gear under typical load conditions, and the equivalent stress is calculated.

[0143]

[0144] in, , , These are the principal stress components. Based on Miner's linear cumulative damage theory, fatigue life is calculated, and the damage factor is defined as follows:

[0145]

[0146] in, The number of cycles at a given stress level. This corresponds to the fatigue life. When... When the material enters a fatigue failure state, laser shock peening is required in S4. The stress gradient data calculated in S1 provides benchmark data for the S2 monitoring system and serves as a real-world working condition sample for S3 data enhancement.

[0147] Furthermore, the multimodal monitoring system in step S2 specifically includes the following methods:

[0148] To monitor stress wave propagation, vibration response, temperature changes, and plasma characteristics during laser shock peening in real time, S2 employs multimodal sensors (stress wave, triaxial accelerometer, infrared thermography, and spectral analysis) for online monitoring of the shock process. Due to timing discrepancies in data collected from different sensors, Dynamic Time Warping (DTW) is used for timing alignment, with a matching error set as follows:

[0149]

[0150] in, , For different modal data sequences, To match the path, The distance function is used. Aligned data is used for S3 data augmentation to improve the stability of cross-condition monitoring. Impact energy distribution, temperature field evolution, and other data monitored in S2 are input into S5 for quality assessment and provide input for S6 feedback optimization.

[0151] Furthermore, the data augmentation method in step S3 specifically includes the following methods:

[0152] Due to the limited amount of actual impact data collected, S3 employs Generative Adversarial Networks (GANs) and Time Generative Adversarial Networks (TimeGANs) to expand the multimodal dataset and enhance the adaptability of the monitoring model. The GAN generates impact stress waves, temperature fields, and vibration data through adversarial training, and the objective function is optimized as follows:

[0153]

[0154] in, Generate synthetic data, To determine the authenticity of the data, TimeGAN further incorporates time-dependent modeling, giving the generated data realistic dynamic characteristics and enhancing the reliability of the S4 mechanism modeling. Furthermore, the generated data is used to train the S5 shock quality assessment model, improving its adaptability across different operating conditions.

[0155] Furthermore, the laser shock strengthening optimization in step S4 specifically includes the following methods:

[0156] Based on the high-stress region identified by S1 and the data generated by S3, S4 constructs a finite element analysis (FEM) model to simulate the propagation of the laser shock wave and the redistribution of residual stress. Residual stress With laser shock parameters (laser energy) Spot diameter Pulse width Number of impacts The relationship is:

[0157]

[0158] Experimental data are used to optimize this relationship and improve the consistency and stability of the strengthening process. After impact strengthening, the residual stress distribution data is input into S5 for quality assessment and provides an optimization reference for S6, enabling the strengthening process to be adaptively adjusted.

[0159] Furthermore, the impact quality assessment method in step S5 specifically includes the following methods:

[0160] S5 employs a spatiotemporal attention map neural network (STAGNN) to fuse stress wave, vibration, temperature, and plasma spectral data to predict the strengthening quality of the impact region and identify areas of aggravated strengthening. An impact quality uniformity index is defined as follows:

[0161]

[0162] in, The average residual stress in the impact region, To determine the number of measurement points. When If the set threshold is exceeded, the S6 reinforcement parameters need to be optimized. The reinforcement depth data evaluated in S5 is used to train the reinforcement learning (DRL) model in S6 to optimize the impulsive strategy and improve reinforcement consistency.

[0163] Furthermore, the laser shock process optimization method in step S6 specifically includes the following methods:

[0164] The S6 employs deep reinforcement learning (DRL) to optimize impact parameters, resulting in a more uniform distribution of residual stress and improving landing gear fatigue life. (State...) Given the current impact parameters and reinforced mass, the action... The optimization objective for parameter adjustment is:

[0165]

[0166] in, As a discount factor, The reward function is calculated as follows:

[0167]

[0168] in, The rate of improvement of residual stress after impact. For stress uniformity, This is to account for temperature changes in the heat-affected zone. S6 improves impact uniformity by optimizing reinforcement parameters and feeds this feedback to S5 for quality assessment, ensuring that the optimized reinforcement strategy maintains high reliability under different operating conditions.

[0169] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A digital twin driven data-augmented landing gear laser shock peening monitoring method, characterized in that, include: Acquire experimental data during the service life of the landing gear, and identify high stress concentration areas based on the experimental data; A multimodal monitoring system is constructed, and data is collected on the laser shock process in the high stress concentration area based on the multimodal monitoring system to obtain laser shock data samples; The laser shock data sample is expanded using a generative data augmentation method to obtain an expanded dataset; A finite element mechanism model is constructed, and numerical simulation analysis is performed based on the finite element mechanism model to obtain numerical simulation data. The extended dataset and the numerical simulation data are then fused to obtain a fused dataset. An impact quality assessment model based on a spatiotemporal attention map neural network is constructed. The fused dataset is input into the impact quality assessment model for calculation to obtain the impact quality assessment result. Based on the impact quality assessment results and deep reinforcement learning methods, an intelligent feedback optimization model is constructed, and the laser shock process parameters are dynamically adjusted based on the intelligent feedback optimization model. Based on the dynamic time warping method, multi-mode data alignment is performed on transient stress waveforms, dynamic response of the impact region, dynamic evolution of the heat-affected zone, and plasma radiation characteristics to obtain laser shock data samples; The expression for calculating the impact characteristics at each monitoring point using the impact quality assessment model is as follows: ； wherein, represents the impact feature of the 1st layer, is a neighborhood set of nodes , and are trainable weights, is a nonlinear activation function; The attention weight expression for the impact quality assessment model is as follows: ； in, For time-related scoring functions, Attention weights; The intelligent feedback optimization model is generated based on a deep Q-network. The objective function of the intelligent feedback optimization model is: ； in, As a discount factor, The current laser shock parameter adjustment strategy, For the reward function, This represents the current system status, including laser shock parameters, workpiece status, and monitoring data. The reward function is defined as: ； in, The rate of improvement of residual stress after impact. For stress uniformity, For temperature changes in the heat-affected zone, , , These are the weight parameters.

2. The method according to claim 1, characterized in that, The process of acquiring experimental data during the landing gear's service life includes: Multi-channel force sensors are used to monitor load evolution during the service life of the landing gear; By combining strain gauge measurements of strain distribution under different working conditions, the material response characteristics under load are analyzed. The experimental data were generated by obtaining residual stress data of key parts using X-ray diffraction technology and monitoring thermal accumulation effect using infrared thermal imaging.

3. The method according to claim 1, characterized in that, The process of identifying high stress concentration areas based on the experimental data includes: processing complex alternating loads using the finite element method to obtain an elastoplastic mechanical model; simulating stress distribution based on the elastoplastic mechanical model; evaluating fatigue life using the Miner linear cumulative damage method; and identifying high stress concentration areas.

4. The method according to claim 1, characterized in that, The process of obtaining laser shock data samples includes: A stress wave sensor was used to record the transient stress waveform induced by laser shock; Vibration signals during the impact process were collected using a triaxial accelerometer, and the dynamic response of the impact region was analyzed. The dynamic evolution of the heat-affected zone is obtained by monitoring temperature changes in the impact area using an infrared thermal imager. The plasma radiation characteristics generated during laser shock were analyzed using a spectral sensor.

5. The method according to claim 1, characterized in that, The process of obtaining the extended dataset includes: A generative adversarial network is used to augment the laser shock data samples to obtain a first dataset. A second dataset is obtained by using a temporal generative adversarial network to enhance the impact process data in the laser impact data sample that conforms to time series characteristics; A third dataset is obtained by probabilistically modeling the laser shock data samples based on a variational autoencoder. A fourth dataset is obtained by generating cross-modal data from the laser shock data samples based on a cross-modal transformation network. The first, second, third, and fourth datasets are filtered and enhanced to obtain the extended dataset.

6. The method according to claim 1, characterized in that, The process of constructing a finite element mechanism model, performing numerical simulation analysis based on the finite element mechanism model, and obtaining numerical simulation data includes: A finite element analysis model of laser shock was established based on the theory of elastoplastic dynamics to simulate the transient stress changes and shock wave propagation characteristics inside the material during the impact process. The Johnson-Cook constitutive model is used to describe the strain rate-dependent plastic behavior of the material and to simulate the evolution of residual stress under different laser shock energies; The residual stress distribution under different working conditions is obtained through laser shock experiments. The residual stress distribution is compared with the transient stress change, shock wave propagation characteristics and residual stress evolution. The constitutive parameters are adjusted by the least squares optimization method to minimize the error between the numerically simulated stress distribution and the experimentally measured residual stress, thereby obtaining the numerical simulation data.

Images