Laser shock peening quality monitoring method based on acousto-optic holographic characteristics and related device

Abstract

The invention belongs to the field of laser shock peening quality monitoring, and discloses a laser shock peening quality monitoring method based on acousto-optic holographic characteristics and a related device.The method comprises the steps that acoustic emission signals on the surface of a target material, the surface of a clamp and the back face of an impact area and plasma optical signals generated by target elements in the target material are synchronously collected; time domain and frequency domain features of each signal are extracted, and adaptive prior fusion is carried out; a pre-trained Transform network-branch LSTM network enhanced quality monitoring model is used for processing fusion features of the acoustic emission signals and the plasma optical signals, laser pulse energy is output, and the restraint layer restrains the pressure intensity and the target material thickness. And finally, performing classification probability decision fusion on the quality parameters represented by the signals based on a probability enhanced decision tree to obtain a quality monitoring result. The objective of the invention is to monitor the laser shock peening quality through acousto-optic holographic multi-feature information so as to improve the monitoring precision.

Description

Technical Field

[0001] The present invention belongs to the field of laser shock peening quality monitoring, and particularly relates to a laser shock peening quality monitoring method and related device based on acousto-optic holographic features. Background Art

[0002] Laser shock peening (LSP) technology, as an efficient means of material surface modification, has advantages in improving the properties of metal materials in recent years. Its principle is based on high-power short-pulse laser irradiating the metal surface, inducing the rapid gasification of the protective layer to generate high-temperature and high-pressure plasma, and forming a transient shock wave load under the action of the constraint layer. When the impact pressure exceeds the dynamic yield strength of the material, the surface layer of the material undergoes dynamic plastic deformation, thereby improving its surface hardness, wear resistance, fatigue resistance and corrosion resistance. Thanks to the high efficiency of energy transfer during the millisecond-level transient action process, the LSP technology has been widely used in engineering fields such as aerospace, nuclear power, biomedicine and material forming, and has played an important role especially in improving the fatigue resistance of structural components and extending their service life.

[0003] Currently, there are some methods for monitoring the quality of the LSP process, such as a laser shock peening quality monitoring method and related device described in the patent with publication number CN118540840A. This method monitors the nanosecond-level plasma optical signal generated by the target element in the target material, divides the signal into a scintillation segment and an attenuation segment, extracts specific time-domain features from these two stages, and then inputs these features into a trained quality monitoring model to output the result characterizing the laser shock peening quality. Although it can achieve accurate evaluation of the quality of the laser shock peening process. However, this method mainly relies on a single information source (i.e., the time-domain features of the plasma optical signal), and there is still a problem that it cannot comprehensively capture the complex information of high pressure and rapid changes during the LSP process, thus limiting the further improvement of the monitoring accuracy. Summary of the Invention

[0004] Aiming at the problems existing in the prior art, the present invention provides a laser shock peening quality monitoring method and related device based on acousto-optic holographic features, aiming to monitor the laser shock peening quality by using acousto-optic holographic multi-feature information to better improve the monitoring accuracy.

[0005] To solve the above technical problems, the present invention is realized through the following technical solutions: According to the first aspect of the present invention, there is provided a laser shock peening quality monitoring method based on acousto-optic holographic features, including: Simultaneously collecting the first acoustic emission signal on the surface of the target material, the second acoustic emission signal on the surface of the fixture, the third acoustic emission signal on the back of the impact area, and the plasma optical signal generated by the target element in the target material during the laser shock peening process; Extract several time-domain features and several frequency-domain features of each acoustic emission signal respectively. Divide the plasma optical signal into a scintillation segment and an attenuation segment, and extract several time-domain features of the scintillation segment and the attenuation segment respectively. Perform adaptive prior fusion on several time-domain features and several frequency-domain features of each acoustic emission signal respectively to obtain the fused time-domain features and fused frequency-domain features of each acoustic emission signal. Perform adaptive prior fusion on several time-domain features of the scintillation segment and the attenuation segment respectively to obtain the fused time-domain features of the scintillation segment and the fused time-domain features of the attenuation segment. Use the pre-trained Transformer network-branched LSTM network enhanced quality monitoring model corresponding to each acoustic emission signal to process the fused time-domain features and fused frequency-domain features of each acoustic emission signal, and output the laser pulse energy, confinement layer confinement pressure, and target material thickness characterized by each acoustic emission signal. Use the pre-trained Transformer network-branched LSTM network enhanced quality monitoring model corresponding to the plasma optical signal to process the fused time-domain features of the scintillation segment and the fused time-domain features of the attenuation segment, and output the laser pulse energy, confinement layer confinement pressure, and target material thickness characterized by the plasma optical signal. Based on the probability-enhanced decision tree, perform decision fusion on the classification probabilities of the laser pulse energy, confinement layer confinement pressure, and target material thickness characterized by each acoustic emission signal, and the laser pulse energy, confinement layer confinement pressure, and target material thickness characterized by the plasma optical signal, and output the final quality monitoring result characterized by the laser pulse energy, confinement layer confinement pressure, and target material thickness.

[0006] In a possible implementation manner of the first aspect, the pre-trained Transformer network-branched LSTM network enhanced quality monitoring model corresponding to each acoustic emission signal and the pre-trained Transformer network-branched LSTM network enhanced quality monitoring model corresponding to the plasma optical signal both include two Transformer networks and four branched LSTM networks.

[0007] In a possible implementation manner of the first aspect, for the first acoustic emission signal, the two Transformer networks respectively perform deep processing on the fused time-domain features and fused frequency-domain features of the first acoustic emission signal, merge the fused time-domain features and fused frequency-domain features of the first acoustic emission signal after deep processing in the time dimension to obtain a time-frequency domain feature matrix, and input the time-frequency domain feature matrix into the linear layer to output the laser pulse energy characterized by the first acoustic emission signal. In the four-branch LSTM network, two of the branch LSTM networks respectively perform in-depth processing on the fused time-domain features and fused frequency-domain features of the first acoustic emission signal after in-depth processing, merge the outputs of the two branch LSTM networks in the time dimension, and finally input them into the linear layer to output the constraint pressure of the constraint layer characterized by the first acoustic emission signal; the other two branch LSTM networks respectively perform in-depth processing on the fused time-domain features and fused frequency-domain features of the first acoustic emission signal after in-depth processing, merge the outputs of the two branch LSTM networks in the time dimension, and finally input them into the linear layer to output the target thickness characterized by the first acoustic emission signal.

[0008] In a possible implementation manner of the first aspect, for the second acoustic emission signal, two Transformer networks respectively perform in-depth processing on the fused time-domain features and fused frequency-domain features of the second acoustic emission signal, merge the fused time-domain features and fused frequency-domain features of the second acoustic emission signal after in-depth processing in the time dimension to obtain a time-frequency domain feature matrix, and input the time-frequency domain feature matrix into the linear layer to output the laser pulse energy characterized by the second acoustic emission signal; In the four-branch LSTM network, two of the branch LSTM networks respectively perform in-depth processing on the fused time-domain features and fused frequency-domain features of the second acoustic emission signal after in-depth processing, merge the outputs of the two branch LSTM networks in the time dimension, and finally input them into the linear layer to output the constraint pressure of the constraint layer characterized by the second acoustic emission signal; the other two branch LSTM networks respectively perform in-depth processing on the fused time-domain features and fused frequency-domain features of the second acoustic emission signal after in-depth processing, merge the outputs of the two branch LSTM networks in the time dimension, and finally input them into the linear layer to output the target thickness characterized by the second acoustic emission signal.

[0009] In a possible implementation manner of the first aspect, for the third acoustic emission signal, two Transformer networks respectively perform in-depth processing on the fused time-domain features and fused frequency-domain features of the third acoustic emission signal, merge the fused time-domain features and fused frequency-domain features of the third acoustic emission signal after in-depth processing in the time dimension to obtain a time-frequency domain feature matrix, and input the time-frequency domain feature matrix into the linear layer to output the laser pulse energy characterized by the third acoustic emission signal; In the four-branch LSTM network, two of the branch LSTM networks respectively perform deep processing on the fused time-domain features and fused frequency-domain features of the third acoustic emission signal after deep processing, merge the outputs of the two branch LSTM networks in the time dimension, and finally input them into the linear layer to output the constrained layer constraint pressure characterized by the third acoustic emission signal; the other two branch LSTM networks respectively perform deep processing on the fused time-domain features and fused frequency-domain features of the third acoustic emission signal after deep processing, merge the outputs of the two branch LSTM networks in the time dimension, and finally input them into the linear layer to output the target thickness characterized by the third acoustic emission signal.

[0010] In a possible implementation manner of the first aspect, for the plasma optical signal, two Transformer networks respectively perform deep processing on the fused time-domain features of the scintillation segment and the fused time-domain features of the attenuation segment, merge the fused time-domain features of the scintillation segment and the fused time-domain features of the attenuation segment after deep processing in the time dimension to obtain a time-domain feature matrix, and input the time-domain feature matrix into the linear layer to output the laser pulse energy characterized by the plasma optical signal; In the four-branch LSTM network, two of the branch LSTM networks respectively perform deep processing on the fused time-domain features of the scintillation segment and the fused time-domain features of the attenuation segment after deep processing, merge the outputs of the two branch LSTM networks in the time dimension, and finally input them into the linear layer to output the constrained layer constraint pressure characterized by the plasma optical signal; the other two branch LSTM networks respectively perform deep processing on the fused time-domain features of the scintillation segment and the fused time-domain features of the attenuation segment after deep processing, merge the outputs of the two branch LSTM networks in the time dimension, and finally input them into the linear layer to output the target thickness characterized by the plasma optical signal.

[0011] In a possible implementation manner of the first aspect, the probability-enhanced decision tree performs decision fusion on the classification probabilities of the laser pulse energy, constrained layer constraint pressure, and target thickness characterized by each acoustic emission signal, and the laser pulse energy, constrained layer constraint pressure, and target thickness characterized by the plasma optical signal, specifically:

[0012]

[0013] where 、 、 and are the probability distribution matrices output by the sensors that collect the first acoustic emission signal, the second acoustic emission signal, the third acoustic emission signal, and the plasma optical signal respectively; is the total number of categories; represents the Confidence of a sensor for a category ; For the splitting decision process of the decision tree; For the sensor Weight; For single-sensor feature transformation; For the interaction weight between sensors; For cross-sensor interaction features; For the fusion function of the decision framework; i And j The values of are 1, 2, 3 or 4.

[0014] According to the second aspect of the present invention, there is provided a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for monitoring the quality of laser shock peening based on acousto-optic holographic features is implemented.

[0015] According to the third aspect of the present invention, there is provided a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the method for monitoring the quality of laser shock peening based on acousto-optic holographic features is implemented.

[0016] According to the fourth aspect of the present invention, there is provided a computer program product, and when the computer program product is executed by a processor, the method for monitoring the quality of laser shock peening based on acousto-optic holographic features is implemented.

[0017] Compared with the prior art, the present invention has at least the following beneficial effects: A method for monitoring the quality of laser shock peening based on acousto-optic holographic features provided by the present invention synchronously collects a variety of acoustic emission signals (including acoustic emission signals on the surface of the target material, the surface of the fixture, and the back of the impact area) and plasma optical signals during the laser shock peening process, and respectively extracts several time-domain features and frequency-domain features of these signals, realizing the capture of multi-dimensional information of the LSP process. Compared with the prior art that only relies on a single information source (the time-domain features of plasma optical signals), the present invention can more comprehensively and accurately reflect the high-pressure and rapidly changing complex information in the LSP process, thereby improving the monitoring accuracy. By using the adaptive prior fusion technology, the time-domain features and frequency-domain features of each acoustic emission signal, as well as the time-domain features of the scintillation section and the attenuation section of the plasma optical signal, are fused to obtain more comprehensive and accurate feature information. The fusion method not only improves the utilization rate of information but also enhances the model's processing ability for the complex information in the LSP process. Combining the quality monitoring model of shared Transformer-branched LSTM, the present invention is highly sensitive to the working condition parameters of laser pulse energy, constraint layer constraint pressure, and target material thickness. By deeply mining the time-varying information of the acousto-optic holographic signals in the laser shock peening process, the present invention can achieve accurate assessment of the quality of the LSP process. Using a probability-enhanced decision tree to perform decision fusion on the classification probabilities of the laser pulse energy, constraint layer constraint pressure, and target material thickness parameters characterized by each acoustic emission signal and plasma optical signal improves the accuracy of the monitoring results.

[0018] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following specifically enumerates preferred embodiments and, in conjunction with the accompanying drawings, makes a detailed description as follows. Description of the Drawings

[0019] To more clearly illustrate the technical solutions in the specific embodiments of the present invention, the following will briefly introduce the drawings required for use in the description of the specific embodiments. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0020] Figure 1 It is a flowchart of a method for monitoring the quality of laser shock peening based on acousto-optic holographic features of the present invention; Figure 2 It is the time-frequency domain features of the acoustic emission signal; Figure 3 It is the time-domain features of the scintillation section and the attenuation section of the plasma optical signal; Figure 4 It is a schematic diagram of the framework of the Transformer network-branched LSTM network for strengthening quality monitoring model; Figure 5It is a flowchart for decision fusion based on a probability-enhanced decision tree; Specific implementation mode To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0021] As Figure 1 shown, the embodiment of the present invention provides a method for monitoring the quality of laser shock peening based on acousto-optic holographic features, which specifically includes the following steps: S1. Synchronously collect the first acoustic emission signal on the surface of the target material, the second acoustic emission signal on the surface of the fixture, the third acoustic emission signal on the back of the impact area, and the plasma optical signal generated by the target element in the target material during the laser shock peening process.

[0022] Specifically, the first acoustic emission signal collected from the surface of the target material is denoted as AE1, the second acoustic emission signal collected from the surface of the fixture is denoted as AE2, the third acoustic emission signal collected from the back of the impact area is denoted as AE3, and the plasma optical signal generated by the target element in the target material is denoted as PMT.

[0023] Exemplarily, use a multi-channel acoustic emission acquisition system to collect the three-way acoustic emission signals AE1, AE2, and AE3. The acoustic emission signal acquisition system consists of a broadband acoustic emission sensor, an attenuator, an acquisition card, and acquisition software. The acoustic emission sensors used to collect AE1, AE2, and AE3 are all the same broadband acoustic emission sensors. The acoustic emission sensor used to collect AE1 is installed on the front of the target material, within a straight-line distance of 5 cm from the impact area. The acoustic emission sensor used to collect AE2 is installed on the surface of the fixture, at a flat surface of the fixture, as close to the impact area as possible. The acoustic emission sensor used to collect AE3 is installed on the back of the target material, directly behind the impact area.

[0024] Use an optical signal acquisition system with a microchannel photomultiplier tube as the core to collect the plasma optical signal. The optical signal acquisition system includes a microchannel photomultiplier tube and a narrowband filter. The narrowband filter can allow the nanosecond-level plasma optical signal with an amplitude evolving over time generated by the target element in the target material to pass through.

[0025] It should be noted that the target material refers to the TC4 titanium alloy plate in this embodiment, the size of the plate is 120*120, the surface of the target material refers to the surface of the target material that receives laser shock, the back of the impact area refers to the other side relative to the surface of the target material and directly opposite to the impact area, and the fixture refers to the device for clamping the target material.

[0026] S2. Extract several time-domain features and several frequency-domain features of each acoustic emission signal respectively. Divide the plasma optical signal into a flashing segment and an attenuation segment, and extract several time-domain features of the flashing segment and the attenuation segment respectively.

[0027] It should be noted that the flashing segment is a signal segment with periodically fluctuating amplitude and gradually tending to balance. Extract specific several time-domain features from the flashing segment and the attenuation segment respectively to obtain several time-domain features of the flashing segment and the attenuation segment.

[0028] Since the acoustic emission signal has rich information in both its time domain and frequency domain, its time-domain and frequency-domain features are both worthy of in-depth exploration. The flashing segment has only two obvious frequency-domain peaks, while the attenuation segment has almost no frequency-domain information. Therefore, the main information of the plasma optical signal exists in the time domain, and it is divided into the flashing segment and the attenuation segment to explore the time-domain feature information.

[0029] Exemplarily, such as Figure 2 and Figure 3 shown, extract specific several time-domain and frequency-domain features from each acoustic emission signal respectively to obtain several time-domain and frequency-domain features of AE1, AE2, and AE3.

[0030] S3. Perform adaptive prior fusion on several time-domain features and several frequency-domain features of each acoustic emission signal respectively to obtain the fused time-domain features and fused frequency-domain features of each acoustic emission signal. Perform adaptive prior fusion on several time-domain features of the flashing segment and the attenuation segment respectively to obtain the fused time-domain features of the flashing segment and the attenuation segment.

[0031] Specifically, the adaptive prior fusion is specifically as follows: (1) Introduce a key parameter, the Information Prior Vector (IPV), and perform preliminary processing on the feature matrix through matrix multiplication.

[0032]

[0033]

[0034] (2) During the processing, since the IPV is a one-dimensional vector, the feature matrix is vectorized during matrix multiplication.

[0035] (3) The selection of the IPV is composed of the L2-norm vector extracted by frame division of the feature matrix.

[0036] Exemplarily, taking AE1 as an example, denote its adaptive prior fusion module as composed of module 1 and module 2. Module 1 receives the input of the AE1 time-domain feature denoted as and module 2 receives the input of the AE1 frequency-domain feature denoted as , the fusion of time-domain features and the fusion of frequency-domain features are respectively completed by Module 1 and Module 2, and their outputs are the fused feature 1 and the fused feature 2, denoted as and . Similarly for AE2 and AE3, and their corresponding modules are denoted as Module 3, Module 4, Module 5, and Module 6. The difference between PMT and the above signals is that its adaptive prior fusion module consists of Module 7 and Module 8. Module 7 is used to process the time-domain features of the scintillation section, and Module 8 is used to process the time-domain features of the attenuation section.

[0037] S4. Use the pre-trained Transformer network-branched LSTM network enhanced quality monitoring model corresponding to each acoustic emission signal to process the fused time-domain features and fused frequency-domain features of each acoustic emission signal, and output the laser pulse energy, the confinement pressure of the confinement layer, and the thickness of the target material characterized by each acoustic emission signal. Use the pre-trained Transformer network-branched LSTM network enhanced quality monitoring model corresponding to the plasma optical signal to process the fused time-domain features of the scintillation section and the fused time-domain features of the attenuation section, and output the laser pulse energy, the confinement pressure of the confinement layer, and the thickness of the target material characterized by the plasma optical signal.

[0038] Specifically, as shown in Figure 4 , the pre-trained Transformer network-branched LSTM network enhanced quality monitoring model corresponding to each acoustic emission signal and the pre-trained Transformer network-branched LSTM network enhanced quality monitoring model corresponding to the plasma optical signal both include two Transformer networks and four branched LSTM networks.

[0039] For the first acoustic emission signal, the two Transformer networks respectively perform in-depth processing on the fused time-domain features and fused frequency-domain features of the first acoustic emission signal, merge the fused time-domain features and fused frequency-domain features of the first acoustic emission signal after in-depth processing in the time dimension to obtain a time-frequency domain feature matrix, and input the time-frequency domain feature matrix into the linear layer to output the laser pulse energy characterized by the first acoustic emission signal; among the four branched LSTM networks, two of the branched LSTM networks respectively perform in-depth processing on the fused time-domain features and fused frequency-domain features of the first acoustic emission signal after in-depth processing, merge the outputs of the two branched LSTM networks in the time dimension, and finally input them into the linear layer to output the confinement pressure of the confinement layer characterized by the first acoustic emission signal; the other two branched LSTM networks respectively perform in-depth processing on the fused time-domain features and fused frequency-domain features of the first acoustic emission signal after in-depth processing, merge the outputs of the two branched LSTM networks in the time dimension, and finally input them into the linear layer to output the thickness of the target material characterized by the first acoustic emission signal.

[0040] For the second acoustic emission signal, two Transformer networks respectively perform deep processing on the fused time-domain features and fused frequency-domain features of the second acoustic emission signal, merge the fused time-domain features and fused frequency-domain features of the second acoustic emission signal after deep processing in the time dimension to obtain a time-frequency domain feature matrix, input the time-frequency domain feature matrix into a linear layer, and output the laser pulse energy characterized by the second acoustic emission signal; among the four-branch LSTM networks, two of the branch LSTM networks respectively perform deep processing on the fused time-domain features and fused frequency-domain features of the second acoustic emission signal after deep processing, merge the outputs of the two branch LSTM networks in the time dimension, and finally input them into a linear layer to output the constraint layer constraint pressure characterized by the second acoustic emission signal; the other two branch LSTM networks respectively perform deep processing on the fused time-domain features and fused frequency-domain features of the second acoustic emission signal after deep processing, merge the outputs of the two branch LSTM networks in the time dimension, and finally input them into a linear layer to output the target material thickness characterized by the second acoustic emission signal.

[0041] For the third acoustic emission signal, two Transformer networks respectively perform deep processing on the fused time-domain features and fused frequency-domain features of the third acoustic emission signal, merge the fused time-domain features and fused frequency-domain features of the third acoustic emission signal after deep processing in the time dimension to obtain a time-frequency domain feature matrix, input the time-frequency domain feature matrix into a linear layer, and output the laser pulse energy characterized by the third acoustic emission signal; among the four-branch LSTM networks, two of the branch LSTM networks respectively perform deep processing on the fused time-domain features and fused frequency-domain features of the third acoustic emission signal after deep processing, merge the outputs of the two branch LSTM networks in the time dimension, and finally input them into a linear layer to output the constraint layer constraint pressure characterized by the third acoustic emission signal; the other two branch LSTM networks respectively perform deep processing on the fused time-domain features and fused frequency-domain features of the third acoustic emission signal after deep processing, merge the outputs of the two branch LSTM networks in the time dimension, and finally input them into a linear layer to output the target material thickness characterized by the third acoustic emission signal.

[0042] For the plasma optical signal, two Transformer networks respectively perform deep processing on the fused time-domain features of the scintillation segment and the fused time-domain features of the decay segment, merge the fused time-domain features of the scintillation segment and the fused time-domain features of the decay segment after deep processing in the time dimension to obtain a time-domain feature matrix, and input the time-domain feature matrix into a linear layer to output the laser pulse energy characterized by the plasma optical signal; among the four branch LSTM networks, two of the branch LSTM networks respectively perform deep processing on the fused time-domain features of the scintillation segment and the fused time-domain features of the decay segment after deep processing, merge the outputs of the two branch LSTM networks in the time dimension, and finally input them into a linear layer to output the confinement layer confinement pressure characterized by the plasma optical signal; the other two branch LSTM networks respectively perform deep processing on the fused time-domain features of the scintillation segment and the fused time-domain features of the decay segment after deep processing, merge the outputs of the two branch LSTM networks in the time dimension, and finally input them into a linear layer to output the target thickness characterized by the plasma optical signal.

[0043] Exemplarily, taking AE1 as an example, two Transformer networks with the same model dimension but different time steps are constructed, denoted as TRM1 and TRM2, to process the fused feature 1 and the fused feature 2 respectively, that is and . Denote the features after being processed by TRM1 and TRM2 as and , and and have three flow directions. First, and will be merged in the time dimension to form a new feature matrix, and this matrix will be input into a linear layer to be used for the classification task of laser pulse energy (5J, 3J, 1J). Second, and are respectively used as the inputs of LSTM1 and LSTM2, and the outputs of the LSTM will be merged in the time dimension and finally output to the linear layer, thereby completing the classification task of three confinement layer confinement pressures (118 kPa, 80 kPa, 40 kPa). Finally, and are respectively used as the inputs of LSTM3 and LSTM4, and their outputs are also merged in the time dimension and passed through the linear layer to achieve the classification task of three target thicknesses (6 mm, 4 mm, 2 mm).

[0044] AE2, AE3, and PMT are the same in principle and have the same processing flow as AE1.

[0045] Preferably, the Transformer network-branched LSTM network reinforcement quality monitoring model is trained using sample data. Specifically, during the training process, a Dynamic Balance Multi-Task Learning Loss (DynaBalance-MTL, DB-MTL) is designed as the loss function for the three tasks. This loss function combines the advantages of Focal Loss in dealing with class imbalance and the characteristics of Label Smoothing Loss in suppressing overfitting. The two complement each other to improve generalization, as shown in the formula:

[0046] where is the weight of Focal Loss, is the initial weight of Label Smoothing Loss, and satisfies . is the regularization coefficient, which controls the influence degree of the entropy term on the total loss. This weight will be automatically learned through gradient descent according to the training of each of the three subtasks. Among them, including the training of the weights of the loss through this adaptive loss function, the classification performance of each subtask can be improved, and the optimization strategy can be dynamically adjusted according to the characteristics of the tasks during the training process. For example, when the weight tends to 1 or 0, tends to 0 when is close to 0 or 1. However, due to the negative sign at the front of the formula, when the weight tends to extreme values, the entropy term will increase the total loss, thereby punishing this situation during the optimization process and prompting the weight to maintain a certain dynamic balance.

[0047] S5. Based on the probability-enhanced decision tree, perform decision fusion on the classification probabilities of the laser pulse energy, the constraint pressure of the constraint layer, and the target thickness characterized by each acoustic emission signal, as well as the laser pulse energy, the constraint pressure of the constraint layer, and the target thickness characterized by the plasma optical signal, and output the final quality monitoring result characterized by the laser pulse energy, the constraint pressure of the constraint layer, and the target thickness.

[0048] Specifically, the decision fusion of three-way acoustic emission and one-way optical signal sensors is realized through a Probability Enhanced Decision Tree (Prob-Enhanced DT). Let the probability distribution matrix output by the four sensors be denoted as:

[0049] In the formula, , , and Probability distribution matrices respectively output by sensors that collect the first acoustic emission signal, the second acoustic emission signal, the third acoustic emission signal, and the plasma optical signal; is the total number of categories; represents the -th sensor's confidence in category .

[0050] Construct a decision fusion framework that can display the interaction relationship between sensors and can dynamically adjust weights, as Figure 5 shown. This decision framework includes a fusion function:

[0051] where is the splitting decision process (nonlinear mapping) of the decision tree; is the weight of sensor ; is the single-sensor feature transformation (such as entropy, confidence difference); is the interaction weight between sensors; is the cross-sensor interaction feature (such as probability difference, joint confidence); is the fusion function of the decision framework; i and j take values of 1, 2, 3, or 4.

[0052] Table 1 Single-sensor and Cross-sensor Probability Feature Extraction

[0053] To adaptively adjust the contribution of each sensor in the decision, this embodiment uses the decision tree feature importance to automatically learn the weights of the sensors. Specifically, for each sensor , the contribution of each sensor is evaluated through the feature importance vector output by the decision tree model. These features include single-sensor features and cross-sensor interaction features , as shown in Table 1. The weights of single-sensor features are directly calculated through feature importance, while the weights of cross-sensor interaction features are proportionally allocated. The final weights are obtained through L2 normalization to ensure the comprehensive effect of all weights. The decision fusion process is divided into three stages: 1. The decision tree first selects the best splitting path based on the interaction features, preferentially selecting the feature with the largest information gain. 2. At the leaf nodes, the probability distributions of each sensor are fused through weighted voting. 3. Finally, the category corresponding to the maximum probability value is taken as the final decision result.

[0054] In another embodiment of the present invention, a computer device is provided. The computer device includes a processor and a memory. The memory is used to store a computer program, and the computer program includes program instructions. The processor is used to execute the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing core and control core of the terminal, and is suitable for implementing one or more instructions. Specifically, it is suitable for loading and executing one or more instructions in the computer storage medium to implement the corresponding method flow or corresponding function. The processor described in the embodiments of the present invention can be used for the operation of a laser shock peening quality monitoring method based on acousto-optic holographic features.

[0055] In another embodiment of the present invention, a storage medium is also provided, specifically a computer-readable storage medium (Memory). The computer-readable storage medium is a memory device in a computer device and is used to store programs and data. It can be understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and, of course, the extended storage medium supported by the computer device. The computer-readable storage medium provides a storage space that stores the operating system of the terminal. And in this storage space, one or more instructions suitable for being loaded and executed by the processor are also stored. These instructions can be one or more computer programs (including program codes). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk memory. One or more instructions stored in the computer-readable storage medium can be loaded and executed by the processor to implement the corresponding steps of the laser shock peening quality monitoring method based on acousto-optic holographic features in the above embodiments.

[0056] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0057] The present invention is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems), and computer program products according to embodiments of the present invention. It should be understood that each flow and / or block in the flowchart and / or block diagram, as well as the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, such that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in Figure 1 one or more of the flows Figure 1 or blocks or the combination of flows and / or blocks.

[0058] These computer program instructions can also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means that implement the functions specified in Figure 1 one or more of the flows Figure 1 or blocks or the combination of flows and / or blocks.

[0059] These computer program instructions can also be loaded onto a computer or other programmable data processing device, such that a series of operation steps are executed on the computer or other programmable device to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in Figure 1 one or more of the flows Figure 1 or blocks or the combination of flows and / or blocks.

[0060] The present invention also provides a computer program product, which is used to execute any one of the above-mentioned laser shock peening quality monitoring methods based on acousto-optic holographic features. Since the computer program product provided by the present invention and the above-mentioned laser shock peening quality monitoring method based on acousto-optic holographic features belong to the same inventive concept, the computer program product provided by the present invention has all the advantages of the above-mentioned laser shock peening quality monitoring method based on acousto-optic holographic features. Therefore, the beneficial effects of the computer program product provided by the present invention will not be elaborated one by one here.

[0061] In the present invention, the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples", etc. mean that the specific features, structures, materials, or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, without contradiction, those skilled in the art can combine and combine the different embodiments or examples described in this specification and the features of different embodiments or examples.

[0062] Finally, it should be noted that the above-mentioned embodiments are only specific embodiments of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: any person skilled in the art within the technical scope disclosed by the present invention can still modify the technical solutions recorded in the foregoing embodiments or can easily think of changes, or make equivalent replacements for some of the technical features; and these modifications, changes, or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for monitoring the quality of laser shock peening based on acousto-optic holographic characteristics, characterized in that, Including: Simultaneously collecting the first acoustic emission signal on the surface of the target material, the second acoustic emission signal on the surface of the fixture, the third acoustic emission signal on the back of the impact area, and the plasma optical signal generated by the target element in the target material during the laser shock peening process; Respectively extracting several time-domain features and several frequency-domain features of each acoustic emission signal, dividing the plasma optical signal into a flashing segment and a decaying segment, and respectively extracting several time-domain features of the flashing segment and the decaying segment; Respectively performing adaptive prior fusion on several time-domain features and several frequency-domain features of each acoustic emission signal to obtain the fused time-domain feature and the fused frequency-domain feature of each acoustic emission signal, and respectively performing adaptive prior fusion on several time-domain features of the flashing segment and the decaying segment to obtain the fused time-domain feature of the flashing segment and the fused time-domain feature of the decaying segment; Using the pre-trained Transformer network-branched LSTM network enhanced quality monitoring model corresponding to each acoustic emission signal to process the fused time-domain feature and the fused frequency-domain feature of each acoustic emission signal, and outputting the laser pulse energy, the constraint layer constraint pressure, and the target material thickness characterized by each acoustic emission signal. Using the pre-trained Transformer network-branched LSTM network enhanced quality monitoring model corresponding to the plasma optical signal to process the fused time-domain feature of the flashing segment and the fused time-domain feature of the decaying segment, and outputting the laser pulse energy, the constraint layer constraint pressure, and the target material thickness characterized by the plasma optical signal; Based on the probability-enhanced decision tree, making decision fusion on the classification probabilities of the laser pulse energy, the constraint layer constraint pressure, and the target material thickness characterized by each acoustic emission signal, and the laser pulse energy, the constraint layer constraint pressure, and the target material thickness characterized by the plasma optical signal, and outputting the final quality monitoring result characterized by the laser pulse energy, the constraint layer constraint pressure, and the target material thickness.

2. The laser shock peening quality monitoring method based on acousto-optic holographic features according to claim 1, wherein The pre-trained Transformer network-branched LSTM network enhanced quality monitoring model corresponding to each acoustic emission signal and the pre-trained Transformer network-branched LSTM network enhanced quality monitoring model corresponding to the plasma optical signal both include two Transformer networks and four branched LSTM networks.

3. The method for monitoring the quality of laser shock peening based on acousto-optic holographic features according to claim 2, characterized in that, For the first acoustic emission signal, the two Transformer networks respectively perform deep processing on the fused time-domain feature and the fused frequency-domain feature of the first acoustic emission signal, merge the fused time-domain feature and the fused frequency-domain feature of the first acoustic emission signal after deep processing in the time dimension to obtain a time-frequency domain feature matrix, and input the time-frequency domain feature matrix into the linear layer to output the laser pulse energy characterized by the first acoustic emission signal; In the four-branch LSTM network, two of the branch LSTM networks respectively perform deep processing on the fused time-domain features and fused frequency-domain features of the first acoustic emission signal after deep processing, merge the outputs of the two branch LSTM networks in the time dimension, and finally input them into the linear layer to output the constraint pressure of the constraint layer characterized by the first acoustic emission signal; the other two branch LSTM networks respectively perform deep processing on the fused time-domain features and fused frequency-domain features of the first acoustic emission signal after deep processing, merge the outputs of the two branch LSTM networks in the time dimension, and finally input them into the linear layer to output the target thickness characterized by the first acoustic emission signal.

4. The method for monitoring the quality of laser shock peening based on acousto-optic holographic features according to claim 2, wherein For the second acoustic emission signal, two Transformer networks respectively perform deep processing on the fused time-domain features and fused frequency-domain features of the second acoustic emission signal, merge the fused time-domain features and fused frequency-domain features of the second acoustic emission signal after deep processing in the time dimension to obtain a time-frequency domain feature matrix, and input the time-frequency domain feature matrix into the linear layer to output the laser pulse energy characterized by the second acoustic emission signal; In the four-branch LSTM network, two of the branch LSTM networks respectively perform deep processing on the fused time-domain features and fused frequency-domain features of the second acoustic emission signal after deep processing, merge the outputs of the two branch LSTM networks in the time dimension, and finally input them into the linear layer to output the constraint pressure of the constraint layer characterized by the second acoustic emission signal; the other two branch LSTM networks respectively perform deep processing on the fused time-domain features and fused frequency-domain features of the second acoustic emission signal after deep processing, merge the outputs of the two branch LSTM networks in the time dimension, and finally input them into the linear layer to output the target thickness characterized by the second acoustic emission signal.

5. A method for monitoring the quality of laser shock peening based on acousto-optic holographic features according to claim 2, characterized in that, For the third acoustic emission signal, two Transformer networks respectively perform deep processing on the fused time-domain features and fused frequency-domain features of the third acoustic emission signal, merge the fused time-domain features and fused frequency-domain features of the third acoustic emission signal after deep processing in the time dimension to obtain a time-frequency domain feature matrix, and input the time-frequency domain feature matrix into the linear layer to output the laser pulse energy characterized by the third acoustic emission signal; In the four-branch LSTM network, two of the branch LSTM networks respectively perform deep processing on the fused time-domain features and fused frequency-domain features of the third acoustic emission signal after deep processing, merge the outputs of the two branch LSTM networks in the time dimension, and finally input them into the linear layer to output the constraint pressure of the constraint layer characterized by the third acoustic emission signal; the other two branch LSTM networks respectively perform deep processing on the fused time-domain features and fused frequency-domain features of the third acoustic emission signal after deep processing, merge the outputs of the two branch LSTM networks in the time dimension, and finally input them into the linear layer to output the target thickness characterized by the third acoustic emission signal.

6. A method for monitoring the quality of laser shock peening based on acousto-optic holographic features according to claim 2, characterized in that, For the plasma optical signal, two Transformer networks respectively perform deep processing on the fused time-domain features of the scintillation segment and the fused time-domain features of the decay segment, merge the fused time-domain features of the scintillation segment and the fused time-domain features of the decay segment after deep processing in the time dimension to obtain a time-domain feature matrix, and input the time-domain feature matrix into a linear layer to output the laser pulse energy characterized by the plasma optical signal; Among the four-branch LSTM networks, two of the branch LSTM networks respectively perform deep processing on the fused time-domain features of the scintillation segment and the fused time-domain features of the decay segment after deep processing, merge the outputs of the two branch LSTM networks in the time dimension, and finally input them into a linear layer to output the confinement layer confinement pressure characterized by the plasma optical signal; the other two branch LSTM networks respectively perform deep processing on the fused time-domain features of the scintillation segment and the fused time-domain features of the decay segment after deep processing, merge the outputs of the two branch LSTM networks in the time dimension, and finally input them into a linear layer to output the target thickness characterized by the plasma optical signal.

7. A method for monitoring the quality of laser shock peening based on acousto-optic holographic features according to claim 1, characterized in that The probability-enhanced decision tree performs decision fusion on the classification probabilities of the laser pulse energy, confinement layer confinement pressure, and target thickness characterized by each acoustic emission signal, as well as the laser pulse energy, confinement layer confinement pressure, and target thickness characterized by the plasma optical signal, specifically: In the formula, , , and are the probability distribution matrices output by the sensors for collecting the first acoustic emission signal, the second acoustic emission signal, the third acoustic emission signal, and the plasma optical signal, respectively; is the total number of categories; represents the confidence of the th sensor in the category ; is the splitting decision process of the decision tree; is the sensor weight; is the single-sensor feature transformation; is the inter-sensor interaction weight; is the cross-sensor interaction feature; is the fusion function of the decision framework; i and j takes values of 1, 2, 3, or 4.

8. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements a method for monitoring the quality of laser shock peening based on acousto-optic holographic features as described in any one of claims 1 to 7.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements a method for monitoring the quality of laser shock peening based on acousto-optic holographic features as described in any one of claims 1 to 7.

10. A computer program product, characterized in that, When the computer program product is executed by the processor, it implements a method for monitoring the quality of laser shock peening based on acousto-optic holographic features as described in any one of claims 1 to 7.

Images