Nondestructive testing system and method for structural uniformity of composite bulletproof helmet

Abstract

The invention discloses a nondestructive testing system and method for the structural uniformity of a composite bulletproof helmet, and relates to the technical field of composite mechanical manufacturing, and the method comprises the following steps: extracting a path curvature in a topological graph of a helmet material structure, carrying out polynomial fitting by using a least square method, generating an optimized scanning track, and determining the structural uniformity of the composite bulletproof helmet; the optimized scanning track is converted into a mechanical arm control instruction, an ultrasonic probe is driven to conduct multi-frequency focusing and incident angle deflection scanning on the helmet structure, and an original echo signal matrix is obtained; the original echo signal matrix is input into an acoustic physical neural model, the feature extraction layer executes time-frequency feature analysis and generates a sound wave propagation feature tensor, and the physical constraint layer executes sound velocity distortion correction and obtains a calibration echo signal matrix; according to the method, by constructing the structural topological graph and the acoustic physical neural model of the helmet material, quantitative analysis of distribution and communication characteristics of microdefects in the composite material is realized, and the overall precision and reliability of nondestructive testing are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of composite material mechanical manufacturing, and in particular to a non-destructive testing system and method for the structural uniformity of a composite material bulletproof helmet. Background Art

[0002] In recent years, with the rapid development of composite material technology, especially the wide application of high-performance composite materials in military equipment, higher requirements have been put forward for the safety, reliability, and performance of protective equipment such as bulletproof helmets. Among them, structural uniformity, as one of the key factors affecting the protective efficiency of bulletproof helmets, has received increasing attention. Traditional non-destructive testing methods such as X-ray imaging and ultrasonic testing can, to a certain extent, evaluate the state of the internal structure of composite materials.

[0003] However, there are still some deficiencies in the existing technologies. Firstly, the existing optical transmission detection method can only identify the macroscopic defect morphology, and fails to establish the mapping relationship between microscopic topological features such as pore distribution and connection paths and macroscopic mechanical parameters such as material stiffness and sound velocity gradient, resulting in the inability to predict the actual impact of defects on bulletproof performance. Secondly, the traditional ultrasonic testing uses a fixed scanning trajectory and a static signal analysis model, without considering the dynamic interference of the composite material microstructure on the acoustic wave propagation path, resulting in deviation in the correction of sound velocity distortion and difficulty in ensuring the detection accuracy. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a non-destructive testing method for the structural uniformity of a composite material bulletproof helmet, which solves the problems of insufficient detection accuracy and lack of micro-macro performance correlation in the existing detection methods.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: In a first aspect, the present invention provides a non-destructive testing method for the structural uniformity of a composite material bulletproof helmet, which includes collecting a dataset of the helmet microstructure, extracting the coordinate of the stress area at the pore position of the helmet to obtain the pore space coordinate, performing a mapping of the geometric attributes on the pore space coordinate to form nodes of the topological structure, applying a geometric adjacency algorithm to connect adjacent pores in the helmet connection path to generate physical channel line segments, performing a mapping of the transmission direction on the physical channel line segments to obtain the edges of the topological structure, and performing weighted fusion on the nodes of the topological structure and the edges of the topological structure to generate a topological map of the helmet material structure; Extracting the path curvature in the topological map of the helmet material structure, and performing polynomial fitting using the least squares method to generate an optimized scanning trajectory, converting the optimized scanning trajectory into a robotic arm control instruction, and driving an ultrasonic probe to perform multi-frequency focusing and incident angle deflection scanning on the helmet structure to obtain an original echo signal matrix; Input the original echo signal matrix into the acoustic physical neural model. When the feature extraction layer is executed, time-frequency feature analysis is performed to generate the acoustic wave propagation feature tensor. The physical constraint layer corrects the sound speed distortion to obtain the calibrated echo signal matrix. Perform multi-scale time-frequency analysis and non-linear conversion on the acoustic wave propagation feature tensor and the calibrated echo signal matrix, and output the non-destructive testing solution; Based on the non-destructive testing solution, perform node dynamic clustering and edge strength reallocation on the topological map of the helmet material structure through the improved incremental weight map algorithm, synchronously update the node weights, and optimize the connection strength of the edges.

[0007] As a preferred solution of the non-destructive testing method for the structural uniformity of the composite material bulletproof helmet described in the present invention, wherein: the helmet microstructure data set includes the helmet pore positions, helmet connection paths, helmet three-dimensional point clouds, and composite material performance data.

[0008] As a preferred solution of the non-destructive testing method for the structural uniformity of the composite material bulletproof helmet described in the present invention, wherein: extracting the path curvature in the topological map of the helmet material structure specifically includes the following steps, Use B-spline interpolation to parameterize and sample the topological map of the helmet material structure to obtain a discrete point set, and perform numerical differentiation on the discrete point set to generate the path curvature.

[0009] As a preferred solution of the non-destructive testing method for the structural uniformity of the composite material bulletproof helmet described in the present invention, wherein: generating the optimized scanning trajectory specifically includes the following operating steps, Perform noise suppression and feature enhancement on the path curvature to generate an optimized curvature distribution; use the least squares method to perform polynomial fitting on the optimized curvature distribution to form the optimized scanning trajectory.

[0010] As a preferred solution of the non-destructive testing method for the structural uniformity of the composite material bulletproof helmet described in the present invention, wherein: converting the optimized scanning trajectory into a robotic arm control instruction specifically includes the following operating steps, Apply dynamic time warping to discretize the time stamps of the optimized scanning trajectory to obtain an angle instruction sequence, and encapsulate the angle instruction sequence to form a robotic arm control instruction.

[0011] As a preferred solution of the non-destructive testing method for the structural uniformity of the composite material bulletproof helmet described in the present invention, wherein: obtaining the original echo signal matrix specifically includes the following operating steps, According to the robotic arm control instruction, drive the ultrasonic probe to scan the helmet structure with multi-frequency focusing and incident angle deflection to obtain multi-channel echo signals; Perform spatio-temporal alignment on the multi-channel echo signals to generate the original echo signal matrix.

[0012] As a preferred solution of the non-destructive testing method for the structural uniformity of the composite bulletproof helmet of the present invention, wherein: the acoustic physical neural model is specifically constructed as follows. Build and initialize the feature extraction layer and the physical constraint layer, and perform feature transfer and double stacking through residual connection to construct the acoustic physical neural model.

[0013] As a preferred solution of the non-destructive testing method for the structural uniformity of the composite bulletproof helmet of the present invention, wherein: the output non-destructive testing solution specifically includes the following operation steps. Input the original echo signal matrix into the acoustic physical neural model. The feature extraction layer performs multi-dimensional feature analysis through a three-dimensional convolution kernel to generate a sound wave propagation feature tensor. The physical constraint layer embeds the elastic constitutive equation to perform sound speed distortion correction on the sound wave propagation feature tensor to obtain a calibrated echo signal matrix. Apply the wavelet packet energy spectrum to perform multi-scale time-frequency analysis and non-linear conversion on the calibrated echo signal matrix to obtain the structural uniformity parameter and the defect probability distribution. Integrate the structural uniformity parameter and the defect probability distribution to output the non-destructive testing solution.

[0014] As a preferred solution of the non-destructive testing method for the structural uniformity of the composite bulletproof helmet of the present invention, wherein: updating the node weights of the topological graph of the helmet material structure and synchronously optimizing the connection strength of the edges specifically includes the following operation steps. Use wavelet packet decomposition to extract the defect-sensitive frequency band parameters in the non-destructive testing solution. According to the defect-sensitive frequency band parameters, perform node dynamic clustering and edge strength reallocation on the topological graph of the helmet material structure through an improved incremental weight graph algorithm to obtain the sound speed gradient coefficient and the stiffness distribution coefficient. Based on the sound speed gradient coefficient, perform gradient update on the node weights, and at the same time, perform coupling optimization on the connection strength of the edges according to the stiffness distribution coefficient.

[0015] In a second aspect, the present invention provides a non-destructive testing system for the structural uniformity of a composite bulletproof helmet, including a topological mapping module, a signal acquisition module, a solution generation module, and a dynamic optimization module. The topological mapping module is used to collect the microscopic structure data set of the helmet, extract the stress area coordinates of the pore positions of the helmet to obtain the pore space coordinates, map the geometric attributes of the pore space coordinates to form the nodes of the topological structure, apply the geometric adjacency algorithm to connect the adjacent pores of the helmet connection path to generate the physical channel line segment, map the transfer direction of the physical channel line segment to obtain the edges of the topological structure, and perform weighted fusion on the nodes of the topological structure and the edges of the topological structure to generate the topological graph of the helmet material structure. A signal acquisition module is used to extract the path curvature in the topological map of the helmet material structure, perform polynomial fitting using the least squares method to generate an optimized scanning trajectory, convert the optimized scanning trajectory into a robotic arm control instruction, drive the ultrasonic probe to perform multi-frequency focusing and incident angle deflection scanning on the helmet structure, and obtain an original echo signal matrix; A solution generation module is used to input the original echo signal matrix into an acoustic physical neural model. When the feature extraction layer executes time-frequency feature analysis, it generates an acoustic wave propagation feature tensor. The physical constraint layer performs sound speed distortion correction to obtain a calibrated echo signal matrix, performs multi-scale time-frequency analysis and non-linear conversion on the acoustic wave propagation feature tensor and the calibrated echo signal matrix, and outputs a non-destructive testing solution; A dynamic optimization module is used to, based on the non-destructive testing solution, perform node dynamic clustering and edge strength reallocation on the topological map of the helmet material structure through an improved incremental weight map algorithm, synchronously update the node weights, and optimize the connection strength of the edges.

[0016] In a third aspect, the present invention provides a computer device, including a memory and a processor. The memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the non-destructive testing method for the structural uniformity of the composite material bulletproof helmet as described in the first aspect of the present invention is implemented.

[0017] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, wherein: when the computer program is executed by the processor, any step of the non-destructive testing method for the structural uniformity of the composite material bulletproof helmet as described in the first aspect of the present invention is implemented.

[0018] The beneficial effects of the present invention are as follows: Through the topological map of the helmet material structure, a mapping relationship is established between the helmet pore positions and the helmet connection paths and mechanical parameters, realizing the quantitative analysis of the internal microscopic defect distribution and its connection characteristics of the composite material, and realizing the correlation between microscopic and macroscopic properties. By adopting a cooperative detection mechanism of an acoustic physical neural model and dynamic scanning trajectory optimization, the propagation modeling accuracy of ultrasonic waves in complex microstructures is improved, the correction deviation problem caused by sound speed distortion is significantly improved, and thus the overall accuracy and reliability of non-destructive testing are greatly improved. Description of the Drawings

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

[0020] Figure 1Flowchart of a non-destructive testing method for the structural uniformity of a composite bulletproof helmet.

[0021] Figure 2 Schematic diagram of a non-destructive testing system for the structural uniformity of a composite bulletproof helmet.

[0022] Figure 3 Flowchart for generating a topological map of the helmet material structure.

[0023] Figure 4 Flowchart for outputting a non-destructive testing plan. Detailed implementation manners

[0024] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe the detailed implementation manners of the present invention with reference to the accompanying drawings of the specification.

[0025] In the following description, many specific details are set forth to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0026] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation manner of the present invention. The appearances of "in one embodiment" in different places in this specification do not all refer to the same embodiment, nor are they separate or alternative embodiments that exclude each other with other embodiments.

[0027] Refer to Figures 1 to 4 , which is an embodiment of the present invention. This embodiment provides a non-destructive testing method for the structural uniformity of a composite bulletproof helmet, including the following steps: S1. Collect a dataset of the helmet's microscopic structure, map the pore positions of the helmet to the nodes of the topological structure, map the connected paths of the helmet to the edges of the topological structure, and generate a topological map of the helmet material structure.

[0028] The specific operation steps are as follows. S1.1. Collect a dataset of the helmet's microscopic structure. The dataset of the helmet's microscopic structure includes the pore positions of the helmet, the connected paths of the helmet, the 3D point cloud of the helmet, and the composite material performance data; Deploy an image sensor on the outer surface of the helmet to collect data on the pore positions of the helmet; deploy a distributed fiber optic sensor in the helmet sandwich to monitor data on the connected paths of the helmet; deploy a laser 3D scanner around the helmet to collect data on the 3D point cloud of the helmet; The performance data of the composite material include mechanical parameters, thermal parameters, electrical parameters, and acoustic parameters. The mechanical parameters are collected by a universal testing machine, the thermal parameters are collected by a differential scanning calorimeter, the electrical parameters are obtained by a high resistance meter, and the acoustic parameters are collected by an ultrasonic detection device. The helmet microstructure data set can not only accurately reflect the topological characteristics of the pore distribution and connection path inside the helmet, but also provide a high-precision data basis for subsequent defect evolution analysis, thus comprehensively improving the accuracy and reliability of the bulletproof helmet structure uniformity evaluation.

[0029] S1.2. Preprocess the helmet microstructure data set. Specifically, in the operation, the outlier detection method based on the 3σ criterion is used to identify and filter outliers in the helmet microstructure data set, and a moving average filter is used with a sliding window to eliminate the random fluctuations during the data collection process. At the same time, the sym8 wavelet basis is used to decompose the helmet microstructure data set into 5 layers to remove high-frequency noise and ensure the stability of the time-frequency characteristics in the helmet microstructure data set. B-spline interpolation is used to perform spatial resampling on the helmet microstructure data set to eliminate the resolution differences between different sensors. The ICP algorithm is used to perform data registration and spatial alignment on the helmet microstructure data set to achieve sub-pixel-level matching of cross-modal data, and dynamic time warping is used for timestamp alignment to eliminate the acquisition timing deviation. Then, min-max scaling is used to normalize the helmet microstructure data set to avoid dimensional differences, and the preprocessed helmet microstructure data set is output.

[0030] S1.3. Extract the force-bearing area coordinates of the pore positions on the helmet through the CAE topology analysis method to obtain the pore space coordinates, and map the geometric attributes of the pore space coordinates to form the nodes of the topological structure. Specifically, in the operation, the CAE topology analysis method is used to perform least squares surface fitting on the pore positions of the helmet to obtain the initial pore distribution data. Through dynamic time warping, the initial pore distribution data is aligned in time sequence and spatially located to obtain the timestamp and spatial position. The timestamp of the initial pore distribution data is used as the x-axis in the three-dimensional space, and the spatial position is used as the y-axis to construct a local coordinate system, and the grid density of the local coordinate system is adjusted to generate a non-uniform sampling grid. The Gaussian integration method is used to perform stress numerical sampling on the non-uniform sampling grid to obtain the stress values of each node, and cubic spline interpolation is performed on the stress values of each node to generate the force-bearing area distribution. DBSCAN density clustering is used to perform domain aggregation and core point extraction on the force-bearing area distribution to obtain the pore space coordinates, which are the mechanically sensitive positions that need to be focused on and can identify the potential failure origin area of the helmet under impact.

[0031] Perform feature marking on the pore space coordinates using Local Binary Pattern (LBP) to generate pore feature descriptors; perform multi-dimensional geometric quantization on the pore feature descriptors using a geometric parameter statistical algorithm to obtain the geometric properties of the pore space coordinates; perform dimensionality reduction on the geometric properties through principal component analysis to obtain the core feature vector with mechanical-geometric composite features; convert the core feature vector into JSON format through Z-score standardization and map it through the add_node function of the NetworkX library to generate the nodes of the topological structure.

[0032] S1.4. Apply the geometric adjacency algorithm to connect adjacent pores in the helmet connection path to generate physical channel line segments; perform mapping of the transfer direction on the physical channel line segments to obtain the edges of the topological structure. In specific operations, quickly retrieve the k-nearest neighbors (k = 5 - 8) of each helmet connection path through the KD-Tree spatial index, and perform triangulation on the k-nearest neighbors to obtain the initial adjacency relationship; according to the initial adjacency relationship, apply the geometric adjacency algorithm to perform spatial topological connection on the helmet connection path to generate initial connection lines, and perform adjacent pore connection on the initial connection lines to obtain the initial pore network; extract the principal stress direction of the initial pore network through stress tensor analysis, and perform path fitting through B-spline interpolation to form physical channel line segments.

[0033] Filter and smooth the physical channel line segments through a sliding window and assign stress gradient weights to generate weighted connection paths; use the principal stress tracking algorithm to extract the direction of the weighted connection paths to generate a standardized transfer direction, and map the standardized transfer direction through the GraphML (Graph Markup Language) protocol to obtain the edges of the topological structure.

[0034] S1.5. Perform weighted fusion of the nodes and edges of the topological structure to generate the topological graph of the helmet material structure. In specific operations, expand the neighborhood of the nodes of the topological structure through DBSCAN clustering, and cluster and merge the adjacent nodes of the topological structure to obtain a set of fused nodes. Synchronously apply the least squares method to perform non-uniform surface fitting on the set of fused nodes to generate a transition surface; perform stress gradient integration on the edges of the topological structure through the principal stress tracking algorithm to obtain the mechanical transfer path, and perform stress field coupling on the mechanical transfer path using Isogeometric Analysis (IGA) to obtain the stress-weighted path; perform weighted fusion of the transition surface and the stress-weighted path through the gradient descent optimization method to generate the topological graph of the helmet material structure.

[0035] S2. Extract the path curvature in the topological graph of the helmet material structure and perform polynomial fitting using the least squares method to generate an optimized scanning trajectory. Convert the optimized scanning trajectory into a robotic arm control instruction to drive the ultrasonic probe to perform multi-dimensional coupling scanning on the helmet structure to obtain the original echo signal matrix.

[0036] The specific operation steps are as follows: S2.1. Use B-spline interpolation to perform parametric sampling on the topological diagram of the helmet material structure, obtain a discrete point set, perform numerical differentiation on the discrete point set to generate path curvature. In specific operations, use wavelet transform to perform multi-scale decomposition on the topological diagram of the helmet material structure to extract geometric features; use principal component analysis to perform dimensionality reduction processing on the geometric features to obtain a key feature point set; apply B-spline interpolation to perform parametric sampling on the key feature point set to generate a parametric geometric representation, and perform equidistant segmentation on the parametric geometric representation to obtain a uniform parameter grid point set; perform coordinate mapping on the uniform parameter grid point set through isoparametric transformation, and use Gaussian integration to perform weight allocation to form a discrete point set; then apply the central difference method to the discrete point set for numerical differentiation, calculate the first derivative and the second derivative, and at the same time use the curvature formula in differential geometry to perform curvature field calculation on the first derivative and the second derivative to generate path curvature. The specific mathematical formula is as follows: ; Wherein, represents the path curvature, represents the parameter index of the discrete point set, represents the parameter 's first derivative, represents the second derivative of the parameter u.

[0037] S2.2. Perform noise suppression and feature enhancement on the path curvature to generate an optimized curvature distribution; use the least squares method to perform polynomial fitting on the optimized curvature distribution to form an optimized scanning trajectory. In specific operations, select the Symlets wavelet basis to perform multi-layer decomposition and frequency segmentation on the path curvature to obtain high-frequency coefficients, and use soft thresholding to perform noise suppression on the high-frequency coefficients, eliminate high-frequency noise while retaining the main curvature features, and obtain the denoised path curvature; extract the gradient change features of the denoised path curvature, and perform local interval division on the gradient change features through a sliding window to obtain gradient change intervals, use the difference operator to perform extreme value localization on the gradient change intervals to obtain curvature extreme points; use the Gaussian weighting function to perform weighted aggregation on the curvature extreme points to enhance the feature significance of the curvature mutation region, and perform amplitude adjustment through Min-Max normalization to generate a normalized curvature feature, and perform weighted integration on the normalized curvature feature to obtain an optimized curvature distribution; It should be noted that the soft threshold is defined based on the standard deviation of the high-frequency coefficients, and the value range is [0.5σ, 2σ] (σ is the standard deviation of the high-frequency coefficients); the difference operator is defined based on the central difference format of the gradient change interval, and the value range is [-1, 0, 1].

[0038] Perform matrix decomposition on the optimized curvature distribution through QR decomposition to obtain an upper triangular matrix. Perform least squares optimization on the upper triangular matrix to obtain the optimal coefficients. According to the optimal coefficients, use the least squares method to perform polynomial fitting on the optimized curvature distribution to obtain a parameterized trajectory. Apply the Savitzky-Golay filter to perform local smoothing on the parameterized trajectory to form a continuous scanning trajectory, and perform discretized interpolation on the continuous scanning trajectory to optimize the continuity of the curvature to obtain an optimized scanning trajectory.

[0039] S2.3. Apply dynamic time warping to perform timestamp discretization on the optimized scanning trajectory to obtain an angle command sequence, and encapsulate the angle command sequence to form a robotic arm control command. In specific operations, extract the curvature parameters of the optimized scanning trajectory, and perform parameter compression through Min-Max scaling to obtain key frames. Non-linearly align the time axis of the key frames with the standard time axis through the dynamic time warping algorithm to obtain the time offset of each key frame. Then perform uniform resampling on the time offset to form a discretized timestamp sequence. According to the discretized timestamp sequence, perform trajectory reconstruction on the optimized scanning trajectory through NURBS curve fitting to obtain the end pose, and perform coordinate mapping on the end pose to generate joint space coordinates. Use quintic polynomial interpolation to perform smooth interpolation on the joint space coordinates to obtain an angle command sequence. It should be noted that the standard time axis refers to the standard time template for the movement of the robotic arm, which is defined based on the specific task cycle time.

[0040] Pack and encapsulate the angle command sequence according to the EtherCAT industrial communication protocol to obtain an instruction data block. At the same time, add a safety check field and an emergency stop instruction slot to the instruction data block to form a robotic arm control command that can be directly issued.

[0041] S2.4. According to the robotic arm control instruction, drive the ultrasonic probe to perform multi-frequency focusing and incident angle deflection scanning on the helmet structure, obtain multi-channel echo signals, perform space-time alignment on the multi-channel echo signals, and generate the original echo signal matrix. Specifically, in the scanning end, send the robotic arm control instruction to the EtherCAT master station to drive the ultrasonic probe of the robotic arm to perform multi-plane dynamic scanning on the helmet structure. Meanwhile, during the scanning process, use the phased array transmitting unit of the ultrasonic probe to emit focused ultrasonic waves in a multi-frequency mode to ensure accurate focusing of the sound beam, and deflect the ultrasonic beam through the phase delay controller to achieve incident angle deflection within a fixed range (such as ±30°) to improve the detection coverage rate and form a spatial scanning grid. At the receiving end, synchronously sample the spatial scanning grid to obtain multi-channel echo signals. Meanwhile, use a digital filter to perform band-pass filtering on the multi-channel echo signals to generate frequency-domain separated signals, and then digitize the frequency-domain separated signals through a high-speed ADC to obtain a time-domain signal sequence. Perform time-delay compensation on the time-domain signal sequence through dynamic time warping to complete spatial alignment and generate the original echo signal matrix.

[0042] S3. Input the original echo signal matrix into the acoustic physics neural model. The feature extraction layer performs time-frequency feature analysis, and the physical constraint layer performs sound speed distortion correction, and outputs a non-destructive testing solution.

[0043] The specific operation steps are as follows. S3.1. Construct and train the acoustic physics neural model. Specifically, in the TensorFlow framework, call a three-dimensional convolutional network through Keras parameters. Set the size of the three-dimensional convolutional kernel to 5×5×5 and the stride to 1×1×1. After three-dimensional convolution, perform feature normalization operations through BatchNormalization (batch normalization) to improve the training stability. Synchronously use the ReLU activation function for non-linear transformation to complete the construction of the feature extraction layer. Call a three-layer fully connected network through nn.Module parameters, embed the elastic constitutive equation pair into the three-layer fully connected network, set the number of hidden layers to 3, the number of neurons to 128, and perform regularization through Dropout after the three-layer fully connected network to prevent overfitting. At the same time, perform weight decay through L2 regularization to complete the construction of the physical constraint layer. Use residual connections to perform multi-dimensional fusion on the feature extraction layer and the physical constraint layer to obtain a fused feature tensor, and perform feature transfer on the fused feature tensor to generate a high-level feature representation. Use the attention mechanism to perform weight allocation on the generated high-level feature representation to obtain weighted features. Apply the gating mechanism to dynamically modulate the weighted features to form optimized weights. According to the optimized weights, perform double stacking on the feature extraction layer and the physical constraint layer through cross-layer connections, and perform feature compression through average pooling to complete the construction of the acoustic physics neural model. Next, train the acoustic physical neural model. Further, divide the original echo signal matrix into a sample set, a training set, and a validation set; on the sample set, perform data augmentation using random flipping and data normalization using Z-score normalization to form enhanced sample data; on the training set, use backpropagation to update the gradients of the enhanced sample data, and synchronously apply the Adam optimizer to dynamically adjust the learning rate to obtain the training loss; on the validation set, use the mean squared error loss function to quantify the error of the training loss to obtain the validation error; when the validation error exceeds the convergence threshold for 5 consecutive epochs, the training terminates, and the trained acoustic physical neural model is output synchronously. It should be noted that the convergence threshold is defined based on the statistical distribution of the validation error, and its value range is [0.01 - 0.05].

[0044] S3.2. Input the original echo signal matrix into the acoustic physical neural model. The feature extraction layer performs multi-dimensional feature analysis through a three-dimensional convolution kernel to generate an acoustic wave propagation feature tensor. In specific operations, input the original echo signal matrix into the acoustic physical neural model through the Input interface, and perform multi-dimensional feature analysis on the original echo signal matrix through a three-dimensional convolution kernel. Among them, the convolution in the time dimension captures the acoustic wave propagation features, the convolution in the frequency dimension extracts the multi-frequency component features, and the convolution in the spatial dimension extracts the spatial correlation features. Concatenate the time-frequency feature analysis, multi-frequency component features, and spatial correlation features in the feature channels to obtain a multi-dimensional feature map; during the convolution process, use SAME padding to perform boundary padding on the multi-dimensional feature map to ensure the spatio-temporal dimension consistency of the multi-dimensional feature map; perform unit variance normalization on the padded multi-dimensional feature map through batch normalization to eliminate the dimensional differences between channels, and then perform a non-linear transformation through the ReLU activation function to suppress negative value features and enhance the expression ability, and output the normalized feature map; apply principal component analysis to perform dimensionality reduction processing on the normalized feature map to form an acoustic wave propagation feature tensor.

[0045] S3.3. The physical constraint layer embeds the elastic constitutive equation to perform sound speed distortion correction on the acoustic wave propagation feature tensor to obtain a calibrated echo signal matrix. In specific operations, input the acoustic wave propagation feature tensor into a three-layer fully connected network embedded with the elastic constitutive equation. The first layer network flattens the acoustic wave propagation feature tensor and maps it to the latent space through a fully connected function. At the same time, inject the composite material performance data as a physical constraint condition into the latent space through a 1×1 convolution to output the latent features; the second layer network calculates the sound speed distortion gradient through the elastic constitutive equation and performs a Hadamard product on the sound speed distortion gradient and the latent features to obtain the sound speed distortion correction value. The specific mathematical formula is as follows, ; Among them, represents the sound speed distortion correction value. represents the identity matrix, represents the flattened acoustic wave propagation feature tensor, represents the sound speed distortion gradient, represents the Hadamard product; It should be noted that the identity matrix refers to a square matrix with all main diagonal elements being 1 and the rest being 0, which is called through the numpy.eye function; According to the sound speed distortion correction value, a gating mechanism is applied to weight and correct the hidden features, and the corrected features are output; the third layer network performs a non - linear transformation on the corrected features using the ReLU activation function and performs gradient clipping through gradient truncation to form a calibrated echo signal matrix.

[0046] S3.4. Apply the wavelet packet energy spectrum to perform multi - scale time - frequency analysis and non - linear transformation on the calibrated echo signal matrix, obtain the structural uniformity parameter and the defect probability distribution, and integrate the structural uniformity parameter and the defect probability distribution to output a non - destructive testing scheme. Specifically, in the operation, the wavelet packet energy spectrum is used to perform multi - layer decomposition on the calibrated echo signal matrix to obtain sub - band signals in different frequency bands; the wavelet packet transform algorithm is used to perform energy integration on the sub - band signals in different frequency bands to obtain the sub - band energy spectrum, and multi - scale time - frequency analysis is performed on the sub - band energy spectrum to obtain the time - frequency energy matrix, which can clearly present the energy change characteristics of the internal structure of the composite material; then the Sigmoid function is used to perform non - linear transformation processing on the time - frequency energy matrix to obtain the normalized energy value, and the normalized energy value is mapped to the standard probability interval using the probability density function to generate the probability distribution matrix; through principal component analysis, dimension compression and parameter extraction are performed on the probability distribution matrix to obtain the structural uniformity parameter; kernel density estimation is used to perform probability density fitting on the structural uniformity parameter to form the defect probability distribution; Finally, the Bayesian fusion algorithm is used to perform weighted integration on the structural uniformity parameter and the defect probability distribution, and heat map rendering is performed through the Unity rendering engine to generate a comprehensive detection result; the comprehensive detection result is formatted and output according to the ISO standard template to form a non - destructive testing scheme.

[0047] S4. Based on the non - destructive testing scheme, the node weights of the helmet material structure topology map are updated through an improved incremental weight graph algorithm, and the connection strength of the edges is optimized synchronously.

[0048] The specific operation steps are as follows, S4.1. Extract the defect-sensitive frequency band parameters in the non-destructive testing scheme using wavelet packet decomposition. Specifically, in the operation, use a sliding window to segment and window the signals in the non-destructive testing scheme to obtain preprocessed signal segments. Then, perform wavelet packet decomposition on the preprocessed signal segments using the db4 wavelet basis function to form multi-scale sub-band components. Next, use the peak detection method to extract features from the multi-scale sub-band components to obtain the characteristic data of each frequency band. Apply linear discriminant analysis to the characteristic data of each frequency band for feature selection to generate the key frequency band feature vectors. Subsequently, perform weighted aggregation and parameter optimization on the key frequency band feature vectors through K-means clustering to form a set of sensitive frequency bands, and randomly sample the set of sensitive frequency bands to obtain the defect-sensitive frequency band parameters.

[0049] S4.2. According to the defect-sensitive frequency band parameters, perform node dynamic clustering and edge strength reallocation on the topological map of the helmet material structure through an improved incremental weight graph algorithm to obtain the sound velocity gradient coefficient and the stiffness distribution coefficient. Specifically, input the defect-sensitive frequency band parameters into the topological map of the helmet material structure through feature projection and perform non-linear transformation using the Gaussian kernel function to enhance the defect feature representation, obtaining a defect-enhanced topological map. Use a sliding window to scan the defect-enhanced topological map and perform regional division within the window to form local feature map sub-blocks. Use the improved incremental weight graph algorithm to perform local clustering and parameter extraction on the local feature map sub-blocks to obtain candidate feature nodes. Apply spectral clustering to the candidate feature nodes for dynamic aggregation to merge similar nodes and obtain the node clustering result. Synchronously use the Sobel operator to perform gradient analysis on the node clustering result to obtain the defect probability gradient. According to the defect probability gradient, use the gradient threshold to reassign the weights of the edges. For example, define the gradient threshold according to the statistical distribution of the defect probability gradient and set the gradient threshold to 0.7. When the defect probability gradient is higher than the gradient threshold, it is defined as a high-gradient region, and the edge strength weight is set to 1.5. When the defect probability gradient is lower than the gradient threshold, it is defined as a low-gradient region, and the edge strength weight is set to 0.8. It should be noted that the Sobel operator is directly called through the cv2.Sobel function in the OpenCV library.

[0050] Perform neighborhood expansion and path search on the defect-enhanced topological map that has completed node clustering and edge strength reallocation through breadth-first search to obtain the defect propagation path. Then, use the A* algorithm to quantify the cost of the defect propagation path to obtain the path cost matrix, and perform Min-Max normalization on the path cost matrix to obtain the sound velocity gradient coefficient, which reflects the degree of distortion of the sound wave propagation path. Synchronously use the Laplacian matrix to perform feature space decomposition on the defect-enhanced topological map with node clustering and edge strength reallocation, select the ratio of the first 3 non-zero eigenvalues, and perform logarithmic transformation to obtain the stiffness distribution coefficient, which characterizes the anisotropy of the material stiffness. It should be noted that the Laplacian matrix refers to the matrix representing the topological structure in the defect-enhanced topological graph, which is obtained by performing the difference set operation of the adjacency relationship and the node degree on the defect-enhanced topological graph; It should be noted that the improved incremental weight graph algorithm has made the following improvements on the existing incremental weight graph algorithm. On the one hand, it uses a dynamic weight adjustment mechanism based on the defect probability gradient to enhance the sensitivity of the defect region features. On the other hand, it combines the sliding window region division and spectral clustering dynamic aggregation to achieve the collaborative optimization of the node and edge strengths.

[0051] S4.3. Based on the sound velocity gradient coefficient, perform gradient update on the node weights, and at the same time, according to the stiffness distribution coefficient, perform coupling optimization on the connection strength of the edges. In the specific operation, use the gradient descent method with the sound velocity gradient coefficient as the learning rate adjustment factor to dynamically adjust the weights of each node in the defect-enhanced topological graph to obtain a preliminary optimized node set; according to the preliminary optimized node set, perform parameter fine-tuning through backpropagation, iterate and update the node weights to obtain a stable node weight distribution; at the same time, use linear interpolation to perform multi-scale sampling and feature fusion on the stiffness distribution coefficient, and perform intensity normalization through Z-score normalization to generate edge strength optimization parameters; according to the edge strength optimization parameters, perform coupling optimization on the edge connection strength, and simultaneously use Gaussian filtering for local smoothing processing to output the collaborative optimized helmet material structure topological graph.

[0052] This embodiment also provides a non-destructive detection system for the structural uniformity of a composite material bulletproof helmet, including: a topological mapping module, a signal acquisition module, a scheme generation module, and a dynamic optimization module; The topological mapping module is used to collect the helmet microstructure data set, extract the force area coordinates of the helmet pore positions, obtain the pore space coordinates, and perform geometric attribute mapping on the pore space coordinates to form the nodes of the topological structure. Apply the geometric adjacency algorithm to connect adjacent pores of the helmet connection path to generate physical channel line segments, perform mapping of the transmission direction on the physical channel line segments to obtain the edges of the topological structure, and perform weighted fusion of the nodes of the topological structure and the edges of the topological structure to generate the helmet material structure topological graph; The signal acquisition module is used to extract the path curvature in the helmet material structure topological graph, and use the least squares method for polynomial fitting to generate an optimized scanning trajectory, convert the optimized scanning trajectory into a robotic arm control instruction, drive the ultrasonic probe to perform multi-frequency focusing and incident angle deflection scanning on the helmet structure, and obtain the original echo signal matrix; A solution generation module is configured to input an original echo signal matrix into an acoustic physical neural model. When the feature extraction layer is executed, time-frequency feature analysis is performed to generate an acoustic wave propagation feature tensor. When the physical constraint layer is executed, sound speed distortion correction is performed to obtain a calibrated echo signal matrix. Multi-scale time-frequency analysis and non-linear conversion are performed on the acoustic wave propagation feature tensor and the calibrated echo signal matrix, and a non-destructive testing solution is output. A dynamic optimization module is configured to, based on the non-destructive testing solution, perform node dynamic clustering and edge strength reallocation on the helmet material structure topology map through an improved incremental weight graph algorithm, synchronously update the node weights, and optimize the connection strength of the edges.

[0053] This embodiment also provides a computer device applicable to the non-destructive testing method for the structural uniformity of a composite bulletproof helmet, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the non-destructive testing method for the structural uniformity of the composite bulletproof helmet as proposed in the above embodiment.

[0054] This computer device may be a terminal, and this computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of this computer device is used to provide computing and control capabilities. The memory of this computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of this computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be implemented through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of this computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of this computer device may be a touch layer covering the display screen, or a button, a trackball, or a touchpad provided on the outer shell of the computer device, or an external keyboard, a touchpad, or a mouse, etc.

[0055] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the non-destructive testing method for realizing the structural uniformity of the composite bulletproof helmet as proposed in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (Static Random Access Memory, abbreviated as SRAM), electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, abbreviated as EEPROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, abbreviated as EPROM), programmable read-only memory (Programmable Red-Only Memory, abbreviated as PROM), read-only memory (Read-Only Memory, abbreviated as ROM), magnetic memory, flash memory, magnetic disk or optical disc.

[0056] In summary, the present invention: through the topological diagram of the helmet material structure, establishes a mapping relationship between the pore positions and the connected paths of the helmet and the mechanical parameters, realizes the quantitative analysis of the internal microscopic defect distribution and its connected characteristics of the composite material, and realizes the correlation between the microscopic and macroscopic properties. Adopting the collaborative detection mechanism of the acoustic physical neural model and the dynamic scanning trajectory optimization improves the propagation modeling accuracy of ultrasonic waves in complex microstructures, significantly improves the correction deviation problem caused by sound speed distortion, and thus greatly improves the overall accuracy and reliability of non-destructive testing.

[0057] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A non-destructive testing method for the structural uniformity of a composite material bulletproof helmet, characterized in that: including, collecting a helmet microstructure dataset, extracting the coordinate of the force area at the pore position of the helmet to obtain the pore space coordinates, mapping the geometric attributes of the pore space coordinates to form nodes of the topological structure, applying a geometric adjacency algorithm to connect adjacent pores in the helmet connection path to generate physical channel line segments, performing a mapping of the transmission direction on the physical channel line segments to obtain the edges of the topological structure, and performing weighted fusion on the nodes and edges of the topological structure to generate a topological map of the helmet material structure; extracting the path curvature in the topological map of the helmet material structure, performing polynomial fitting using the least squares method to generate an optimized scanning trajectory, converting the optimized scanning trajectory into a robotic arm control instruction, and driving an ultrasonic probe to perform multi-frequency focusing and incident angle deflection scanning on the helmet structure to obtain an original echo signal matrix; inputting the original echo signal matrix into an acoustic physics neural model, where the feature extraction layer performs time-frequency feature analysis to generate a sound wave propagation feature tensor, the physical constraint layer performs sound speed distortion correction to obtain a calibrated echo signal matrix, and performing multi-scale time-frequency analysis and non-linear conversion on the sound wave propagation feature tensor and the calibrated echo signal matrix to output a non-destructive testing solution; based on the non-destructive testing solution, performing node dynamic clustering and edge strength reallocation on the topological map of the helmet material structure through an improved incremental weight graph algorithm, synchronously updating the node weights and optimizing the connection strength of the edges.

2. The non-destructive testing method for the structural uniformity of the composite material bulletproof helmet according to claim 1, characterized in that: The helmet microstructure dataset includes helmet pore positions, helmet connection paths, helmet three-dimensional point clouds, and composite material performance data.

3. The non-destructive testing method for the structural uniformity of the composite material bulletproof helmet according to claim 1, characterized in that: The extraction of the path curvature in the topological map of the helmet material structure specifically includes the following steps: using B-spline interpolation to parameterize and sample the topological map of the helmet material structure to obtain a discrete point set, and performing numerical differentiation on the discrete point set to generate the path curvature.

4. The non-destructive testing method for the structural uniformity of the composite material bulletproof helmet according to claim 3, characterized in that: The generation of the optimized scanning trajectory specifically includes the following operating steps: performing noise suppression and feature enhancement on the path curvature to generate an optimized curvature distribution; performing polynomial fitting on the optimized curvature distribution using the least squares method to form an optimized scanning trajectory.

5. The non-destructive testing method for the structural uniformity of the composite material bulletproof helmet according to claim 1, characterized in that: The conversion of the optimized scanning trajectory into a robotic arm control instruction specifically includes the following operating steps: applying dynamic time warping to discretize the time stamps of the optimized scanning trajectory to obtain an angle instruction sequence, and encapsulating the angle instruction sequence to form a robotic arm control instruction.

6. The non-destructive testing method for the structural uniformity of the composite material bulletproof helmet according to claim 5, characterized in that: The obtaining of the original echo signal matrix specifically includes the following operating steps: driving an ultrasonic probe to perform multi-frequency focusing and incident angle deflection scanning on the helmet structure according to the robotic arm control instruction to obtain multi-channel echo signals; performing spatio-temporal alignment on the multi-channel echo signals to generate an original echo signal matrix.

7. The non-destructive testing method for the structural uniformity of the composite material bulletproof helmet according to claim 1, characterized in that: The acoustic physics neural model is specifically constructed as follows: building and initializing a feature extraction layer and a physical constraint layer, and performing feature transfer and double stacking through residual connections to construct an acoustic physics neural model.

8. The non-destructive testing method for the structural uniformity of the composite material bulletproof helmet according to claim 1, characterized in that: The output of the non-destructive testing solution specifically includes the following operating steps: inputting the original echo signal matrix into the acoustic physics neural model, where the feature extraction layer performs multi-dimensional feature analysis through a three-dimensional convolution kernel to generate a sound wave propagation feature tensor; The physical constraint layer embeds the elastic constitutive equation to perform sound speed distortion correction on the acoustic wave propagation characteristic tensor, and obtains the calibrated echo signal matrix; Apply the wavelet packet energy spectrum to perform multi-scale time-frequency analysis and non-linear conversion on the calibrated echo signal matrix, and obtain the structural uniformity parameter and the defect probability distribution; Integrate the structural uniformity parameter and the defect probability distribution, and output the non-destructive testing scheme.

9. The non-destructive testing method for the structural uniformity of the composite material bulletproof helmet according to claim 8, characterized in that: Updating the node weights of the helmet material structure topology map and optimizing the connection strength of the edges synchronously, specifically including the following operation steps, Use wavelet packet decomposition to extract the defect-sensitive frequency band parameters in the non-destructive testing scheme; According to the defect-sensitive frequency band parameters, perform node dynamic clustering and edge strength reallocation on the helmet material structure topology map through the improved incremental weight graph algorithm, and obtain the sound speed gradient coefficient and the stiffness distribution coefficient; Based on the sound speed gradient coefficient, perform gradient update on the node weights, and at the same time, optimize the coupling of the connection strength of the edges according to the stiffness distribution coefficient.

10. A non-destructive testing system for the structural uniformity of a composite material bulletproof helmet, based on the non-destructive testing method for the structural uniformity of the composite material bulletproof helmet according to any one of claims 1 to 9, characterized in that: Including a topology mapping module, a signal acquisition module, a scheme generation module, and a dynamic optimization module; The topology mapping module is used to collect the helmet microstructure data set, extract the force area coordinates of the helmet pore positions, obtain the pore space coordinates, map the geometric attributes of the pore space coordinates to form the nodes of the topological structure, apply the geometric adjacency algorithm to connect the adjacent pores of the helmet connection path, generate the physical channel line segment, map the transfer direction of the physical channel line segment, obtain the edges of the topological structure, and perform weighted fusion on the nodes of the topological structure and the edges of the topological structure to generate the helmet material structure topology map; The signal acquisition module is used to extract the path curvature in the helmet material structure topology map, perform polynomial fitting using the least squares method to generate an optimized scanning trajectory, convert the optimized scanning trajectory into a robotic arm control instruction, drive the ultrasonic probe to scan the helmet structure with multi-frequency focusing and incident angle deflection, and obtain the original echo signal matrix; The scheme generation module is used to input the original echo signal matrix into the acoustic physical neural model. The feature extraction layer performs time-frequency feature analysis to generate the acoustic wave propagation characteristic tensor. The physical constraint layer performs sound speed distortion correction to obtain the calibrated echo signal matrix, and performs multi-scale time-frequency analysis and non-linear conversion on the acoustic wave propagation characteristic tensor and the calibrated echo signal matrix, and outputs the non-destructive testing scheme; The dynamic optimization module is used to perform node dynamic clustering and edge strength reallocation on the helmet material structure topology map through the improved incremental weight graph algorithm based on the non-destructive testing scheme, synchronously update the node weights, and optimize the connection strength of the edges.

Images