Device and method for monitoring state of laser shock peening absorption layer

Abstract

The invention discloses a device and method for monitoring the state of a laser shock peening absorption layer, and relates to the technical field of surface strengthening, and the device comprises a visual image sensor array, an image preprocessing module, an image analysis module and a state evaluation module. The visual image sensor array is used for acquiring visible spectrum images and near-infrared band images of a laser shock peening processing area from multiple angles; the image pre-processing module is used for pre-processing the acquired images to obtain pre-processed images at different angles; the image analysis module is deployed with an image recognition model based on a ResNet-UNet hybrid architecture and an attention mechanism, and is used for recognizing the preprocessed image to obtain an image analysis result; and the state evaluation module is used for evaluating by adopting a multi-threshold judgment mechanism according to the image analysis result to obtain an evaluation result. The state change of the absorption layer in the laser shock peening process can be monitored in real time, the damage condition can be found in time, and the stability and consistency of laser shock peening machining are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of surface strengthening technology, and more specifically to an apparatus and method for monitoring the state of a laser shock strengthening absorption layer. Background Technology

[0002] In laser shock peening (LSP), the absorber layer, as the key medium for converting laser energy into plasma shock waves, significantly impacts the strengthening effect due to its type and state. The integrity of the absorber layer plays a decisive role in the process. Excessive laser energy density, inappropriate pulse width, or improper repetition rate selection can lead to localized overheating or excessive impact on the absorber layer, causing it to break and detach. Once the absorber layer is damaged, the laser will directly act on the material surface. Lacking the energy conversion and protection of the absorber layer, the material surface will experience severe thermal effects, leading to localized melting, vaporization, or even ablation. This not only fails to achieve the desired strengthening effect but also degrades the surface quality, introduces microscopic defects, and reduces fatigue performance. Therefore, in actual processing, it is essential to optimize the combination of process parameters to ensure the absorber layer remains intact at all times to achieve the ideal strengthening effect.

[0003] Current monitoring of the absorption layer status during laser shock peening (LSP) is significantly inadequate. Traditional manual monitoring methods are not only inefficient but also susceptible to subjective factors, failing to promptly detect and respond to absorption layer damage. This lag can lead to surface quality defects, unstable processing quality, and increased production costs. Currently, absorption layer damage detection technology during LSP is still lacking. Existing research primarily focuses on optimizing absorption layer performance.

[0004] Therefore, how to monitor the changes in the state of the absorption layer in real time during laser shock peening and detect damage in a timely manner is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] In view of the above problems, the present invention is proposed to provide an apparatus and method for monitoring the state of a laser shock strengthening absorption layer to overcome or at least partially solve the above problems. Based on a multi-scale feature fusion neural network, it can monitor the changes in the state of the absorption layer in real time during laser shock strengthening, detect damage in a timely manner and trigger an early warning, effectively improving the stability and consistency of laser shock strengthening processing.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, embodiments of the present invention provide an apparatus for monitoring the state of a laser shock-enhanced absorption layer, comprising: a visual image sensor array, an image preprocessing module, an image analysis module, and a state assessment module; A visual image sensor array acquires visible spectral and near-infrared images of the laser-strengthened processing area from multiple angles. The image preprocessing module preprocesses the acquired images to obtain preprocessed images from different angles; The image analysis module deploys an image recognition model based on a ResNet-UNet hybrid architecture and an attention mechanism to recognize preprocessed images and obtain image analysis results. The image recognition model includes an encoder, an adaptive feature fusion module, an attention module, and a decoder. The encoder extracts multi-level features, the adaptive feature fusion module generates multi-view, multi-band fusion features based on the multi-level feature fusion, the attention module weights and fuses the multi-view, multi-band fusion features to obtain comprehensive fusion features, and the decoder performs pixel-level segmentation on the comprehensive fusion features to identify multiple typical damage features of the absorption layer and obtain image analysis results. The status assessment module uses a multi-threshold judgment mechanism to evaluate based on the image analysis results and obtain the assessment results.

[0007] Preferably, the visual image sensor array includes multiple high-speed industrial cameras; the multiple high-speed industrial cameras are respectively installed in front of the laser shock strengthening processing area, distributed in a cone shape, forming a multi-angle field of view coverage, and acquiring visible spectrum images and near-infrared band images at different angles; the high-speed industrial cameras are equipped with CMOS sensors.

[0008] Preferably, in the image recognition model, the encoder uses residual connections to solve the gradient vanishing problem in deep networks, and extracts multi-level features from low-level texture to high-level semantics from the preprocessed image. The adaptive feature fusion module employs a dual-path feature extraction and dynamic weighted fusion strategy to fuse the extracted multi-level features. It calculates weights for the multi-level features extracted from the preprocessed image corresponding to the visible spectrum image and the multi-level features extracted from the preprocessed image corresponding to the near-infrared band image, dynamically determines the importance of different modal data, and performs weighted fusion of the two sets of multi-level features according to the weights to obtain multi-view multi-band fused features. The attention module of the integrated spatial-channel attention mechanism includes a spatial attention layer, a channel attention layer, and a multi-scale fusion layer. The spatial attention layer highlights the damaged area by learning the importance weights of multi-view, multi-band fusion features in spatial location. The channel attention layer enhances the expressive power of key features by learning the importance of multi-view, multi-band fusion features in feature channels. The multi-scale fusion layer uses a multi-scale feature fusion strategy to weight and combine multi-view, multi-band fusion features according to the weights learned by the spatial and channel attention layers to obtain comprehensive fusion features and improve the recognition accuracy of damaged features of different sizes. The multi-scale fusion layer adopts a multi-scale feature pyramid structure to improve the detection accuracy of defects of different sizes. The decoder restores spatial resolution through upsampling and skip connections, achieving pixel-level accurate segmentation of integrated fusion features and identifying multiple typical damage features.

[0009] To achieve comprehensive detection, the image analysis module registers and fuses multi-view images acquired from different angles and multi-band images of different types. Combined with the extraction of features such as texture and edge, and leveraging the complementary advantages of multi-band information, the module achieves accurate detection and classification of surface and internal defects in laser shock-enhanced processing areas through joint analysis of multi-view and multi-band images and defect recognition algorithms, thereby improving the accuracy and robustness of detection.

[0010] Preferably, the image preprocessing module adopts a hardware acceleration architecture based on FPGA and DSP chips, and the preprocessing operations include improved bilateral filtering for noise reduction, adaptive histogram equalization enhancement, multi-point perspective transformation geometric correction, and standardization of images acquired from different angles into a unified format.

[0011] Preferred, improved bilateral filtering denoising employs the following steps: Texture and noise analysis: Calculate the local gray-level variance and texture gradient distribution of the image, and estimate the noise intensity and texture complexity; Initial weight calculation: Based on the spatial distance and gray-level difference between image pixels, calculate the spatial distance weight and gray-level similarity weight; Adaptive weight adjustment: Spatial distance weight is increased in regions of the image where the noise intensity is higher than a set intensity threshold, and gray-level similarity weight is increased in texture edge regions where the texture complexity change exceeds a set complexity threshold; the weight ratio is dynamically adjusted according to the real-time analysis results. Filtering application: Perform bilateral filtering on the image using adjusted weights to suppress high-frequency noise and spot interference while maintaining edge details.

[0012] Preferably, the multiple typical damage features include linear features, local bulges, and irregular missing areas; the image analysis results include no obvious damage features or one or more of the multiple typical damage features, as well as the relevant parameters of the damage area corresponding to the damage features, including boundary contours and morphological features.

[0013] Preferably, the evaluation index is calculated based on the relevant parameters in the image analysis results, and the evaluation is judged based on the evaluation index and the multi-threshold judgment mechanism. The obtained evaluation results include normal state, slightly damaged state and severely damaged state.

[0014] Preferably, the image recognition model is trained using the cross-entropy loss function and an optimizer.

[0015] Preferably, the system also includes an adaptive learning module that employs an online incremental learning strategy and knowledge distillation technology. This module continuously collects processing data during the laser shock peening process to optimize the image recognition model. The processing data includes multi-angle visible spectrum images and multi-band near-infrared images, corresponding defect annotation information, processing parameters (such as laser power, shock frequency, and material type), and environmental condition data. The online incremental learning strategy enables the image recognition model to dynamically update its parameters as new data arrives, avoiding forgetting issues caused by changes in data distribution. The knowledge distillation technology transfers knowledge from the new model to a lightweight student model, achieving efficient model updates and deployment. This process ensures the continuous adaptability and accuracy of the image recognition model in real-world production environments.

[0016] Preferably, the adaptive learning module includes a material recognition model for identifying various commonly used absorption layer materials (including aluminum foil, black paint, and black tape). The material recognition model includes a material feature encoder and a classifier based on a deep neural network. The material feature encoder preprocesses and extracts features from the acquired images, encoding the spectral and textural characteristics of different materials into feature vectors. The classifier is used to discriminate and identify the feature vectors, achieving accurate classification of the absorption layer material type. The material recognition model and the image recognition model adopt a joint training strategy, sharing some feature extraction layers. Through multi-task learning framework, they are collaboratively optimized, comprehensively utilizing material features and defect features to process the acquired data, ultimately generating more accurate image analysis results. This effectively supports state monitoring and defect detection in laser shock-enhanced processing areas, improving recognition accuracy and robustness.

[0017] Preferably, the image preprocessing module, image analysis module, and status assessment module are built into the host computer, and the visual image sensor array is connected to the host computer through a high-speed data transmission interface to achieve fast image transmission.

[0018] Preferably, it also includes a feedback control module, which triggers a multi-level early warning mechanism based on the evaluation results, issuing an early warning signal when the damage is minor and issuing an early warning signal and automatically suspending the processing flow when the damage is severe.

[0019] A method for monitoring the state of a laser-shock-enhanced absorption layer includes the following steps: Step 1: Acquire an unprocessed surface image of the absorption layer as a reference image to establish a reference image library; Step 2: Continuously acquire multi-angle images of the absorption layer surface during laser shock; Step 3: Preprocess the acquired multi-angle images; Step 4: Train an image recognition model using a reference image library, and use the image recognition model to identify multiple typical damage features in the preprocessed image to obtain image analysis results; Step 5: Determine the multi-threshold judgment mechanism based on the features of the baseline image in the reference image library, and use the multi-threshold judgment mechanism to judge the severity of the absorption layer damage based on the image analysis results to obtain the evaluation results.

[0020] Preferably, a corresponding early warning mechanism is triggered based on the evaluation results.

[0021] Preferably, online incremental learning continuously optimizes the damage feature recognition algorithm to improve system accuracy.

[0022] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a device and method for monitoring the state of the laser shock strengthening absorption layer, which can effectively avoid the melting, vaporization or ablation of the material surface caused by the damage of the absorption layer, ensure the stability and consistency of the laser shock strengthening process, be applicable to working conditions with different laser power and pulse width, have excellent damage detection accuracy, significantly improve processing quality and production efficiency, reduce the material surface defect rate, and provide a reliable guarantee for improving the fatigue life of high-performance key components. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0024] Figure 1 This is a schematic diagram of the working process of the device for monitoring the state structure of the laser shock-enhanced absorption layer provided in an embodiment of the present invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] This invention discloses a device for monitoring the state of a laser-shock-enhanced absorption layer, comprising: a visual image sensor array, an image preprocessing module, an image analysis module, and a state assessment module; A visual image sensor array acquires visible spectral and near-infrared images of the laser-strengthened processing area from multiple angles. The image preprocessing module preprocesses the acquired images to obtain preprocessed images from different angles; The image analysis module deploys an image recognition model based on a ResNet-UNet hybrid architecture and an attention mechanism to recognize preprocessed images and obtain image analysis results. The image recognition model includes an encoder, an adaptive feature fusion module, an attention module, and a decoder. The encoder extracts multi-level features, the adaptive feature fusion module generates multi-view, multi-band fusion features based on the multi-level feature fusion, the attention module weights and fuses the multi-view, multi-band fusion features to obtain comprehensive fusion features, and the decoder performs pixel-level segmentation on the comprehensive fusion features to identify multiple typical damage features of the absorption layer and obtain image analysis results. The status assessment module uses a multi-threshold judgment mechanism to evaluate based on the image analysis results and obtain the assessment results.

[0027] In one specific embodiment, the visual image sensor array 1 includes four high-speed industrial cameras, respectively installed at four angles of the laser shock-enhanced processing area, forming a 120° field of view coverage. Each camera is equipped with a 6-megapixel CMOS sensor, with a data acquisition frequency of 60 frames per second. Two cameras acquire visible spectrum images, and the other two acquire near-infrared (855nm) images. The cameras are connected to a host computer equipped with an image preprocessing module, an image analysis module, and a condition assessment module via a high-speed data transmission interface, achieving millisecond-level image transmission. This invention uses multiple industrial cameras to acquire images of the absorption layer surface in real time from different angles. Through dual-spectral imaging in the visible spectrum (400-700nm) and near-infrared band (700-1000nm), the physical change characteristics of the absorption layer under laser action are comprehensively captured. The visible spectrum mainly reflects surface morphology changes and color anomalies, while the near-infrared band can detect internal temperature distribution and microstructural changes in the material. Multi-angle acquisition eliminates blind spots from a single viewpoint, ensuring comprehensive detection of damage features.

[0028] In one specific embodiment, the image preprocessing module employs a hardware-accelerated architecture based on FPGA and DSP. It uses an improved bilateral filtering denoising method for Gaussian filtering, adaptive histogram equalization enhancement, and geometric correction on the original image. This standardizes images acquired from different angles to a unified format with a 1024×1024 pixel resolution, providing high-quality input for subsequent analysis. The hardware-accelerated architecture based on FPGA and dedicated DSP chips enables parallel image processing. The improved bilateral filtering algorithm combines spatial and intensity domain weighting functions to remove noise while preserving edge information. Adaptive histogram equalization dynamically adjusts contrast enhancement parameters based on the statistical characteristics of local regions, improving the visibility of minor damage features. Multi-point perspective transformation establishes geometric mapping relationships between different viewpoints, unifying multi-angle images to a standard coordinate system and providing a consistent data foundation for subsequent analysis.

[0029] Furthermore, the improved bilateral filtering denoising process employs the following steps: Texture and noise analysis: Calculate the local gray-level variance and texture gradient distribution of the image, and estimate the noise intensity and texture complexity; Initial weight calculation: Based on the spatial distance and gray-level difference between image pixels, calculate the spatial distance weight and gray-level similarity weight; Adaptive weight adjustment: Spatial distance weight is increased in regions of the image where the noise intensity is higher than a set intensity threshold, and gray-level similarity weight is increased in texture edge regions where the texture complexity change exceeds a set complexity threshold; the weight ratio is dynamically adjusted according to the real-time analysis results. Filtering application: Perform bilateral filtering on the image using adjusted weights to suppress high-frequency noise and spot interference while maintaining edge details.

[0030] In one specific embodiment, the image recognition model in the image analysis module adopts an improved ResNet-50 convolutional neural network structure. The encoder utilizes residual connections to address the gradient vanishing problem in deep networks, extracting multi-level features from the preprocessed image, ranging from low-level texture to high-level semantics. The adaptive feature fusion module employs a dual-path feature extraction and dynamic weighted fusion strategy to fuse the extracted multi-level features. Weights are calculated for the multi-level features extracted from the preprocessed image corresponding to the visible spectrum image and those extracted from the preprocessed image corresponding to the near-infrared band image, dynamically determining the importance of different modal data. Based on these weights, the two sets of multi-level features are weighted and fused to obtain multi-view, multi-band fused features. The spatial-channel attention mechanism comprises a spatial attention layer, a channel attention layer, and a multi-scale fusion layer. The spatial attention layer learns the importance weights of multi-view, multi-band fusion features in spatial location to highlight damaged areas. The channel attention layer learns the importance of multi-view, multi-band fusion features in feature channels to enhance the expressive power of key features. The multi-scale fusion layer uses a multi-scale feature fusion strategy to weight and combine the extracted multi-view, multi-band fusion features based on the weights learned by the spatial and channel attention layers, obtaining comprehensive fusion features to improve the recognition accuracy of damaged features of different sizes. The decoder restores spatial resolution through upsampling and skip connections, achieving pixel-level precise segmentation of the comprehensive fusion features. The multi-scale fusion layer employs a multi-scale feature pyramid structure to improve the detection accuracy of defects of different sizes. Trained with 5000 sets of labeled samples, it can identify three typical damaged features of the absorption layer: microcracks (linear features, width <0.5mm), bulge formation (local bulges, diameter 0.5-10mm), and local peeling (irregular area missing, area >2mm²).

[0031] Furthermore, by using the cross-entropy loss function and the Adam optimizer for training, a recognition accuracy of over 95% was achieved on the validation set.

[0032] In one specific embodiment, the image analysis results include one or more of the following: no obvious damage features or multiple typical damage features, as well as relevant parameters of the damaged area corresponding to the damage features. The relevant parameters include boundary contours and morphological features. The relevant parameters in the state assessment module calculate the assessment index, and the assessment is judged according to the assessment index and the multi-threshold judgment mechanism. The obtained assessment results include normal state, slightly damaged state, and severely damaged state, namely, normal (no obvious damage), slightly damaged (micro-cracks appear), and severely damaged (bulges or local peeling).

[0033] In one specific embodiment, the present invention captures the dynamic changes of the absorption layer surface during laser shock peening (LSP) processing using a multi-angle high-speed visual image sensor array. Combined with improved deep learning algorithms and hardware acceleration technology, it achieves intelligent monitoring and evaluation of the absorption layer state. In the evaluation stage, the state evaluation module not only directly utilizes the damage type output by the image recognition model, but also establishes a multi-dimensional evaluation index system based on the geometric morphology of the damage features to quantify the damage features and provide input data for the multi-threshold judgment mechanism.

[0034] The multi-dimensional evaluation index system includes multiple parameters such as the area of ​​the damaged region, aspect ratio, connectivity, and gradient change. The evaluation process of the state evaluation module is as follows: 1. Obtain the boundary contour and morphological features of the damaged area identified by the image recognition model; 2. Index Calculation: The damaged area is calculated based on the boundary contour to reflect the extent of damage. The aspect ratio is measured to determine the degree of shape abnormality. Connectivity is analyzed to identify the continuity and distribution characteristics of the damage. The gradient change is evaluated to reflect the sharpness and complexity of the damaged edge. The damaged area, aspect ratio, connectivity, and gradient change are used as evaluation indicators. 3. Material and process parameter reference: Call the current absorption layer material type (aluminum foil, black paint, black tape, etc.) and laser shock process parameters (laser power, shock frequency, scanning speed, etc.) as conditional factors affecting the threshold setting; 4. Threshold adjustment: Using a preset empirical model (statistical rules or machine learning models established based on historical processed data) and combined with conditional factors, the judgment thresholds of each evaluation indicator are dynamically adjusted. 5. Comprehensive assessment: The calculated multiple assessment indicators are compared with the adjusted judgment thresholds, and the damage level is output as the assessment result using a multi-threshold judgment mechanism, including normal state, minor damage state and severe damage state.

[0035] The empirical model, based on historical data from a large amount of data on different material types, process parameters, and damage cases, establishes a mapping relationship between damage indicators and damage levels through manual statistics or machine learning methods, enabling reasonable adjustment of thresholds under different conditions. The role of the absorber layer material type differs in this stage from that of the adaptive learning module: in the state assessment module, it is used to correct thresholds and avoid misjudgments caused by differences in material properties; while in the adaptive learning module, it is used to improve recognition accuracy during the training phase. This ensures that the assessment results have high adaptability and accuracy under different material and process conditions.

[0036] In one specific embodiment, the laser shock strengthening device is mounted on a robotic arm, which enables the device to move and adjust. The invention also includes a feedback control module that triggers a corresponding early warning mechanism to control the robotic arm based on the severity of the damage. When minor damage occurs, a yellow warning signal is issued; when severe damage occurs, a red warning is issued and the processing flow is automatically suspended to prevent damage to the substrate surface.

[0037] In one specific embodiment, the system further includes an adaptive learning module that employs an online incremental learning strategy and knowledge distillation technology. This module continuously collects processing data during the laser shock peening process to optimize the image recognition model. The processing data includes multi-angle visible spectrum images and multi-band near-infrared images, corresponding defect annotation information, processing parameters (such as laser power, shock frequency, and material type), and environmental condition data. The online incremental learning strategy enables the image recognition model to dynamically update its parameters as new data arrives, avoiding forgetting issues caused by changes in data distribution. The knowledge distillation technology transfers knowledge from the new model to a lightweight student model, achieving efficient model updates and deployment. This process ensures the continuous adaptability and accuracy of the image recognition model in actual production environments.

[0038] The adaptive learning module also includes a material recognition model for identifying various commonly used absorbent layer materials (including aluminum foil, black paint, and black tape). It can automatically distinguish between three common absorbent layer materials: aluminum foil (thickness 0.05-0.15mm), black paint (thickness 0.3-0.5mm), and black tape (thickness 0.2-0.3mm). The recognition parameters are adjusted according to the different material characteristics to improve the system's adaptability. The adaptive learning module employs an online incremental learning strategy, dynamically storing new training samples through a sample cache pool. Knowledge distillation technology uses the original model as the teacher network and the updated model as the student network, minimizing the difference in their output distributions to retain learned knowledge while simultaneously learning new feature patterns. This mechanism effectively avoids catastrophic forgetting, enabling the system to adapt to new material types and damage patterns without losing its original recognition capabilities. The material recognition model includes a material feature encoder and a deep neural network-based classifier. The material feature encoder extracts intrinsic properties of different absorbing layer materials (aluminum foil, black paint, black tape) through a deep learning network, including physical parameters such as surface texture, reflectivity, and thermal conductivity. These properties are encoded into high-dimensional feature vectors. The classifier is used to discriminate and identify these feature vectors, achieving accurate classification of absorbing layer material types. The material recognition model and the image recognition model employ a joint training strategy, sharing some feature extraction layers. Through a multi-task learning framework, they collaboratively optimize and comprehensively utilize material features and defect features to process the collected data, ultimately generating more accurate image analysis results. This effectively supports state monitoring and defect detection in laser-strengthened areas, improving recognition accuracy and robustness.

[0039] On the other hand, in one specific embodiment, a method for monitoring the state of a laser shock-enhanced absorption layer based on the above-described device, such as... Figure 1 As shown, the process is as follows: S1: After the system starts up, a reference image is first acquired on the unprocessed absorption layer to establish a reference model; S2: During the laser shock process, the visual image sensor array continuously acquires multi-angle images of the absorption layer surface; S3: The image preprocessing module performs noise reduction, enhancement, and standardization on the acquired raw images; S4: The image recognition model is trained using the reference model. The image analysis module uses the image recognition model to identify possible absorption layer damage features in the image and generates image analysis results. S5: Determine the multi-threshold judgment mechanism based on the features of the baseline image in the reference model. The state assessment module judges the severity of the absorption layer damage based on the image analysis results and generates the assessment results. S6: The feedback control module triggers the corresponding early warning mechanism based on the evaluation results; S7: The adaptive learning module continuously optimizes the recognition algorithm to improve system accuracy.

[0040] Experiments have verified that this invention can achieve a damage detection accuracy of over 95% and a response time of less than 100ms under working conditions of laser power of 2-20J and pulse width of 10-30ns. It effectively avoids material surface quality problems caused by absorption layer damage and improves the stability and consistency of laser shock strengthening processing.

[0041] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to the method section.

[0042] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A device for monitoring the state of a laser-shock-enhanced absorption layer, characterized in that, include: Visual image sensor array, image preprocessing module, image analysis module, and state assessment module; A visual image sensor array acquires visible spectral and near-infrared images of the laser-strengthened processing area from multiple angles. The image preprocessing module preprocesses the acquired images to obtain preprocessed images from different angles; The image analysis module is equipped with an image recognition model based on a ResNet-UNet hybrid architecture and an attention mechanism. It recognizes preprocessed images and obtains image analysis results. The image recognition model includes an encoder, an adaptive feature fusion module, an attention module, and a decoder. The encoder extracts multi-level features, the adaptive feature fusion module generates multi-view and multi-band fusion features based on the multi-level feature fusion, the attention module weights and fuses the multi-view and multi-band fusion features to obtain comprehensive fusion features, and the decoder performs pixel-level segmentation on the comprehensive fusion features to identify multiple typical damage features of the absorption layer and obtain image analysis results. The status assessment module uses a multi-threshold judgment mechanism to evaluate based on the image analysis results and obtain the assessment results.

2. The device for monitoring the state of a laser shock-enhanced absorption layer according to claim 1, characterized in that, The visual image sensor array includes multiple high-speed industrial cameras; these cameras are installed in front of the laser shock peening processing area in a conical distribution to acquire visible spectrum images and near-infrared images of different wavelengths from different angles.

3. The device for monitoring the state of a laser shock-enhanced absorption layer according to claim 1, characterized in that, In image recognition models, encoders utilize residual connections to extract multi-level features from preprocessed images, ranging from low-level texture to high-level semantics. The adaptive feature fusion module calculates weights for the multi-level features extracted from the preprocessed image corresponding to the visible spectrum image and the multi-level features extracted from the preprocessed image corresponding to the near-infrared band image, respectively, and then performs weighted fusion of the two sets of multi-level features according to the weights to obtain multi-view multi-band fused features. The attention module includes a spatial attention layer, a channel attention layer, and a multi-scale fusion layer; the spatial attention layer learns the importance weights of multi-view, multi-band fusion features in spatial location; The channel attention layer learns the importance weights of multi-view, multi-band fused features in the feature channels; The multi-scale fusion layer uses the weights learned by the spatial attention layer and the channel attention layer to weight and combine the multi-view and multi-band fusion features to obtain comprehensive fusion features; The decoder restores spatial resolution through upsampling and skip connections, performs pixel-level segmentation of the integrated features, and identifies multiple typical damage features.

4. The device for monitoring the state of a laser shock-enhanced absorption layer according to claim 1, characterized in that, The image preprocessing module adopts a hardware acceleration architecture based on FPGA and DSP chips. The preprocessing operations include improved bilateral filtering for noise reduction, adaptive histogram equalization enhancement, multi-point perspective transformation geometric correction, and standardization of images acquired from different angles into a unified format.

5. The device for monitoring the state of a laser shock-enhanced absorption layer according to claim 1, characterized in that, The typical damage features include linear features, local bulges, and irregular missing areas; the image analysis results include no obvious damage features or one or more of the typical damage features, as well as the relevant parameters of the damage area corresponding to the damage features.

6. The device for monitoring the state of a laser shock-enhanced absorption layer according to claim 1, characterized in that, Evaluation indicators are calculated based on image analysis results. The damage status of the absorption layer is assessed based on the evaluation indicators and a multi-threshold judgment mechanism. The obtained evaluation results include normal state, slightly damaged state, and severely damaged state.

7. The device for monitoring the state of a laser shock-enhanced absorption layer according to claim 1, characterized in that, The image recognition model is trained using the cross-entropy loss function and an optimizer.

8. The device for monitoring the state of a laser shock-enhanced absorption layer according to claim 1, characterized in that, It also includes an adaptive learning module, which uses an online incremental learning strategy and knowledge distillation technology to optimize the image recognition model by continuously collecting processing data during the laser shock enhancement process; It also deploys a material identification model, including a material feature encoder and a classifier; the material feature encoder preprocesses and extracts features from the collected processing data, encoding the spectral and texture characteristics of different materials into feature vectors; The classifier is used to identify the feature vectors and determine the type of the absorption layer material. The material identification model and the image identification model adopt a joint training strategy, share some feature extraction layers, and are collaboratively optimized through a multi-task learning framework. The collected data are processed using material features and damage features to optimize the image analysis results.

9. The device for monitoring the state of a laser shock-enhanced absorption layer according to claim 6, characterized in that, It also includes a feedback control module, which triggers a multi-level early warning mechanism based on the evaluation results. It issues an early warning signal when the damage is minor and an early warning signal and automatically suspends the processing flow when the damage is severe.

10. A method for monitoring the state of a laser-shock-enhanced absorption layer, characterized in that, An apparatus for monitoring the state of a laser-shock-enhanced absorption layer as described in any one of claims 1-9, comprising the following steps: Step 1: Acquire an unprocessed surface image of the absorption layer as a reference image to establish a reference image library; Step 2: Continuously acquire multi-angle images of the absorption layer surface during laser shock; Step 3: Preprocess the acquired multi-angle images; Step 4: Train an image recognition model using a reference image library, and use the image recognition model to identify multiple typical damage features in the preprocessed image to obtain image analysis results; Step 5: Determine the multi-threshold judgment mechanism based on the features of the baseline image in the reference image library, and use the multi-threshold judgment mechanism to judge the severity of the absorption layer damage based on the image analysis results to obtain the evaluation results.

Images