ETFE film performance testing method, device and system
Through multi-scale surface scanning and mechanical loading, combined with deep learning technology, the performance relationship matrix of ETFE film is constructed, which solves the problem of low accuracy in performance testing of ETFE films in the prior art, and achieves high-precision performance prediction.
Patent Information
- Application Number
- CN202411099464.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-12
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-08-12
AI Technical Summary
The prior art is difficult to fully reflect the complex performance characteristics of ETFE membranes in practical applications, and the accuracy of existing testing methods is low due to the complex correlation between microstructure and macro performance.
Multi-scale surface scanning, comprehensive membrane structure feature recognition, multi-modal mechanical loading and data acquisition are used to construct performance relationship matrix through feature extraction and relationship matrix transformation, and performance prediction is performed using attention-based cyclic graph convolutional network model.
It improves the accuracy of ETFE membrane performance testing, enhances the ability to capture complex nonlinear relationships and learns local structural features, and improves the accuracy of performance prediction.
Smart Images

Figure CN118629536B_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field of performance testing, and in particular to an ETFE film performance testing method, device and system. Background Art
[0002] As a high-performance polymer material, ETFE membrane has broad application prospects in the field of architecture. The performance test of ETFE membrane is a complex issue. Traditional testing methods are often limited to testing of a single scale or a single mode, which is difficult to fully reflect the complex performance characteristics of ETFE membrane in practical applications. In addition, due to the complex correlation between the microstructure and macroscopic performance of ETFE membrane, it is difficult to accurately predict its performance by relying solely on simple experimental data analysis. This leads to the low accuracy of existing technologies. Summary of the invention
[0003] The present application provides an ETFE film performance testing method, device and system, which are used to improve the accuracy of ETFE film performance testing.
[0004] In a first aspect, the present application provides a method for testing the performance of an ETFE film, the method comprising:
[0005] Perform multi-scale surface scanning on the standard ETFE film test sample to obtain the surface morphology data of the ETFE film;
[0006] Performing comprehensive membrane structure feature recognition on the surface morphology data of the ETFE membrane to construct a comprehensive structural model of the ETFE membrane;
[0007] Performing multi-mode mechanical loading and data acquisition on the comprehensive structural model of the ETFE membrane to obtain stress-strain relationship data and membrane structure evolution data;
[0008] Performing feature extraction and relationship matrix conversion on the stress-strain relationship data and the membrane structure evolution data to obtain a performance relationship matrix of the ETFE membrane;
[0009] The performance relationship matrix of the ETFE film is input into the attention-based recurrent graph convolutional network model to perform performance prediction, and the ETFE film performance prediction result is obtained.
[0010] In a second aspect, the present application provides an ETFE film performance testing device, the ETFE film performance testing device comprising:
[0011] A scanning module is used to perform multi-scale surface scanning on a standard ETFE film test sample to obtain surface morphology data of the ETFE film;
[0012] An identification module is used to perform comprehensive membrane structure feature identification on the surface morphology data of the ETFE membrane and construct a comprehensive structure model of the ETFE membrane;
[0013] An acquisition module, used for performing multi-mode mechanical loading and data acquisition on the comprehensive structural model of the ETFE membrane, to obtain stress-strain relationship data and membrane structure evolution data;
[0014] A conversion module, used for performing feature extraction and relationship matrix conversion on the stress-strain relationship data and the membrane structure evolution data to obtain a performance relationship matrix of the ETFE membrane;
[0015] The prediction module is used to input the performance relationship matrix of the ETFE film into the attention-based recurrent graph convolutional network model to perform performance prediction and obtain the ETFE film performance prediction result.
[0016] The third aspect of the present application provides an ETFE film performance testing system, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the ETFE film performance testing system executes the above-mentioned ETFE film performance testing method.
[0017] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned ETFE film performance testing method.
[0018] In the technical solution provided by the present application, by combining atomic force microscopy, white light interferometer and large-area optical scanning, comprehensive surface morphology data collection from nanometer to macroscopic scale is realized, and the accuracy and completeness of surface feature recognition are improved. A more accurate and detailed comprehensive structural model of ETFE membrane is constructed by adopting technologies such as multi-level structural unit analysis, structural topology map extraction and spatial autocorrelation analysis. Through finite element meshing and orthogonal experimental design, the optimal combination of uniaxial tension, biaxial tension and shear test is realized, and the mechanical properties of ETFE membrane under complex stress state are comprehensively evaluated. Combined with digital image analysis and X-ray diffraction real-time analysis, detailed stress-strain relationship data and membrane structure evolution data are obtained. Through multi-scale characteristic coefficient calculation, typical correlation analysis and tensor decomposition, the key performance parameters of ETFE membrane are effectively extracted, and a high-dimensional performance relationship matrix is constructed. Advanced deep learning technology is adopted to achieve high-precision prediction of ETFE membrane performance, improve the prediction model's ability to capture complex nonlinear relationships, enhance the model's learning ability for local structural features, and improve the accuracy of performance prediction. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.
[0020] Figure 1 This is a schematic diagram of an embodiment of the ETFE film performance testing method in the embodiment of the present application;
[0021] Figure 2 This is a schematic diagram of an embodiment of an ETFE film performance testing device in an embodiment of the present application. DETAILED DESCRIPTION
[0022] The embodiments of the present application provide a method, device and system for testing the performance of an ETFE membrane. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described here can be implemented in an order other than that illustrated or described here. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or system that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or systems.
[0023] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 , one embodiment of the ETFE film performance testing method in the embodiment of the present application includes:
[0024] Step S101, performing multi-scale surface scanning on a standard ETFE film test sample to obtain surface morphology data of the ETFE film;
[0025] It is understandable that the execution subject of the present application may be an ETFE film performance testing device, or a terminal or a server, which is not limited here. The present application embodiment is described by taking a server as the execution subject as an example.
[0026] Specifically, the surface height data at the nanometer level is obtained by atomic force microscope scanning technology, which provides the height distribution and microscopic characteristics of the ETFE film surface at the nanometer scale, which helps to understand the surface characteristics of the ETFE film at a very small scale; the surface profile data of the ETFE film at the micrometer scale is obtained by white light interferometer scanning technology, which provides the height change and surface structure information of the ETFE film at the medium scale; large-area optical scanning can obtain the surface morphology data of the ETFE film at the macroscale, providing detailed information on the overall surface morphology of the ETFE film. The surface height data at the nanometer level is Fourier transformed, and the height data in the spatial domain is converted into the surface roughness spectrum in the frequency domain. The roughness characteristics of the ETFE film surface at the nanometer scale are analyzed and understood through this spectrum. Based on the information of the surface roughness spectrum, the surface profile data at the micrometer level is adaptively filtered to optimize and refine the surface profile data, remove noise and unnecessary details, and obtain a more accurate and representative surface profile map. The optimized surface profile map is subjected to wavelet multi-resolution analysis to extract the multi-scale characteristic coefficients of the surface at different scales. These characteristic coefficients reflect the important characteristics and change trends of the ETFE film surface at different scales. According to the information of multi-scale characteristic coefficients, the surface morphology data at the macro level is segmented and divided into several surface structure units with different characteristics. The morphological features of the surface structure units are extracted to obtain the shape descriptors and texture features of each unit, so as to describe and analyze the morphological features and texture characteristics of the ETFE film surface. Based on the feature information, the corresponding ETFE film surface morphology data is constructed.
[0027] Step S102, performing comprehensive film structure feature recognition on the surface morphology data of the ETFE film, and constructing a comprehensive structural model of the ETFE film;
[0028] Specifically, the surface morphology data of the ETFE film is decomposed to obtain the characteristic coefficient matrix of different scales, and the data is refined and classified at different scales so that the features of each scale can be clearly identified and extracted. The surface morphology data is segmented by adaptive threshold according to the characteristic coefficient matrix to obtain a multi-level structural unit. The multi-level structural unit is subjected to morphological skeleton extraction to simplify the complex surface morphology into a skeleton structure, so that the basic morphology and connection relationship of the structure are retained, and the structural topology map is obtained. According to the structural topology map, the multi-level structural unit is subjected to regional growth processing, and the structural units with similar characteristics are aggregated to form a larger structural domain to obtain the structural domain. The structural domain can reflect the structural characteristics of the ETFE film surface at a larger scale. The grayscale co-occurrence matrix is calculated for the structural domain to obtain the texture feature vector. The grayscale co-occurrence matrix calculation can analyze the grayscale distribution and its mutual relationship inside the structural domain, and extract the feature vector reflecting the surface texture characteristics. The texture feature vector is used to classify the structural domain to obtain the structural type label. Through the classification processing of the texture feature vector, different structural domains are classified into different structural type labels. Spatial autocorrelation analysis of the structural type labels reveals the spatial distribution of the structural type labels and obtains the structural distribution pattern. Based on the structural distribution pattern, a preliminary three-dimensional structural representation of the ETFE membrane is constructed. Tensor decomposition is performed on the preliminary three-dimensional structural representation, and the three-dimensional structural data is reduced and extracted to extract the most important structural feature information and obtain the structural feature tensor. Based on the structural feature tensor, a comprehensive structural model of the ETFE membrane is generated.
[0029] Step S103, performing multi-mode mechanical loading and data acquisition on the comprehensive structural model of the ETFE membrane to obtain stress-strain relationship data and membrane structure evolution data;
[0030] Specifically, the comprehensive structural model of ETFE membrane is meshed by finite element, and the continuous ETFE membrane structure is discretized into a finite number of grid units to obtain a discretized calculation model. A multi-mode mechanical loading scheme is designed according to the discretized calculation model to obtain a loading sequence matrix. Under different mechanical conditions, the ETFE membrane is subjected to systematic loading tests to comprehensively evaluate its mechanical properties. The loading sequence matrix records the loading parameters and sequence under different loading modes to ensure the repeatability of the experiment and the reliability of the results. The loading sequence matrix is subjected to orthogonal experimental design to obtain the optimized loading parameter combination. Orthogonal experimental design is an effective experimental design method that optimizes the combination of multiple factors to obtain as much experimental information as possible with the least number of experiments. Based on the optimized loading parameter combination, the comprehensive structural model of ETFE membrane is subjected to uniaxial tensile test to obtain uniaxial stress-strain data. In the uniaxial tensile test, the stress-strain relationship data of the ETFE membrane under uniaxial tensile conditions is measured and recorded by applying a tensile force in a single direction. Real-time digital image analysis is performed on the uniaxial stress-strain data to obtain the local strain field distribution. Digital image analysis technology can capture and analyze the strain distribution of each area on the surface of ETFE film in real time during the tensile test, providing more detailed and accurate local strain information. According to the distribution of local strain field, the comprehensive structural model of ETFE film is subjected to biaxial tensile test to obtain biaxial stress-strain data. In the biaxial tensile test, by applying tensile forces in two directions at the same time, the stress-strain relationship data of ETFE film under biaxial tensile conditions are measured and recorded. The principal stress analysis is performed on the biaxial stress-strain data to obtain the stress principal axis direction field. The principal stress analysis analyzes the stress principal axis direction to understand the main force direction and stress concentration area of ETFE film during the force process. According to the stress principal axis direction field, the comprehensive structural model of ETFE film is subjected to shear test to obtain shear stress-strain data. In the shear test, by applying shear force, the stress-strain relationship data of ETFE film under shear conditions are measured and recorded, which helps to evaluate its anti-shear performance. The shear stress-strain data is subjected to Fourier transform to convert the data in the time domain or space domain into the frequency domain, thereby revealing the main frequency components and characteristic information in the data and obtaining stress-strain relationship data. At the same time, X-ray diffraction is used to analyze the membrane structure changes during the comprehensive structural model test in real time to obtain membrane structure evolution data. X-ray diffraction technology can monitor and analyze the crystal structure changes inside the ETFE membrane in real time during the test and obtain the evolution information of the membrane structure during mechanical loading.
[0031] Step S104, performing feature extraction and relationship matrix conversion on the stress-strain relationship data and the membrane structure evolution data to obtain a performance relationship matrix of the ETFE membrane;
[0032] Specifically, the corresponding multi-scale stress-strain characteristic coefficients are calculated according to the stress-strain relationship data. By performing multi-scale analysis on the stress-strain data, the stress-strain characteristic coefficients at different scales are extracted to describe the stress-strain characteristics of the ETFE membrane at different stress levels. At the same time, the membrane structure evolution data is characterized and the main structure evolution mode is obtained. The characteristic analysis can reveal the change law of the membrane structure during mechanical loading and identify the main structure evolution mode. Canonical correlation analysis is performed based on the multi-scale stress-strain characteristic coefficients and the main structure evolution mode to obtain the stress-structure correlation characteristics. Canonical correlation analysis is a statistical analysis method that reveals the intrinsic connection between stress characteristics and structural evolution by analyzing the correlation between multiple variables and extracts representative stress-structure correlation characteristics. The stress-structure correlation characteristics are subjected to nonlinear dimensionality reduction processing to obtain a low-dimensional representation vector. The nonlinear dimensionality reduction method can map high-dimensional data to a low-dimensional space while ensuring that the main information of the data is not lost, thereby simplifying the complexity of the data and improving the computational efficiency. A stress-structure mapping function is constructed based on the low-dimensional representation vector, and the representation vector in the low-dimensional space is mapped to the actual stress and structural characteristics to establish a corresponding relationship between stress and structure. A sensitivity analysis of the stress-structure mapping function was performed to obtain key performance parameters. Sensitivity analysis can evaluate the degree of influence of different parameters on performance, thereby identifying the parameters that play a key role in the performance of ETFE membrane. Resampling and enhancement were performed according to the key performance parameters to obtain an extended data set. Resampling and enhancement techniques enrich the diversity of the data set and improve the generalization ability of the model by expanding and optimizing the original data. Tensor decomposition was performed on the extended data set to obtain a multimodal feature tensor. Tensor decomposition can decompose complex multimodal data into core tensors and factor matrices, thereby extracting the main feature information and patterns in the data. The core tensor and factor matrix are calculated based on the multimodal feature tensor. The core tensor and factor matrix represent the global and local features of the data, respectively. The core tensor and factor matrix are transformed into a tensor network. By transforming and reorganizing the core tensor and factor matrix, a relationship matrix that can fully reflect the performance of ETFE membrane is constructed.
[0033] Step S105: input the performance relationship matrix of the ETFE film into the attention-based recurrent graph convolutional network model for performance prediction to obtain the ETFE film performance prediction result.
[0034] Specifically, the performance relationship matrix of ETFE film is input into the attention-based recurrent graph convolutional network model, and the performance relationship matrix of ETFE film is encoded by graph structure, and the performance relationship matrix is converted into an adjacency matrix and a node feature matrix to obtain the initial graph representation. The adjacency matrix describes the topological structure of the graph, while the node feature matrix contains the attribute information of each node. Multi-head self-attention is calculated according to the initial graph representation, and each attention head is linearly transformed, dot product and softmax operation is performed to obtain the attention weight matrix. The multi-head self-attention mechanism can capture the relationship between different parts of the graph and enhance the expression ability of the model by calculating multiple attention heads in parallel. The calculation of the attention weight matrix includes linear transformation of the initial graph representation to reduce the dimension and enhance the feature expression, then calculating the correlation between nodes by dot product, and normalizing the weights by softmax operation. The attention weight matrix and the node feature matrix are matrix multiplied and multiplied by the adjacency matrix, and the graph convolution operation is performed to obtain the local aggregated features. The graph convolution operation can effectively extract the local features in the graph by aggregating the features of the node with the features of its neighboring nodes. The calculation of local aggregate features is to multiply the attention weight matrix with the node feature matrix and then multiply it with the adjacency matrix to achieve weighted aggregation of features. The local aggregate features are input into the gated recurrent unit, and the hidden state features are obtained by updating the gate, resetting the gate and calculating the candidate hidden state. The gated recurrent unit controls the flow and memory of information through the update gate and the reset gate, and can effectively capture the temporal characteristics in the sequence data. The hidden state features are calculated by the gated recurrent unit to integrate the current local features and historical feature information. The hidden state features are residually connected to obtain multi-scale fusion features, and the mean and variance of the multi-scale fusion features are calculated. The layer normalization operation is performed to obtain standardized features with a mean of 0 and a variance of 1. The residual connection adds the original features in the feature calculation process to achieve the fusion of multi-scale features, which helps to improve the performance and stability of the model. The layer normalization operation standardizes the mean and variance of the features, so that the features have better numerical stability and training efficiency in subsequent calculations. The standardized features are subjected to global average pooling to obtain a graph-level vector representation, which is then input into the fully connected layer. The target feature vector is obtained by weight matrix multiplication and bias addition. The global average pooling operation averages the features of all nodes in the graph to obtain a global graph-level feature vector, which represents the feature information of the entire graph. The fully connected layer extracts the target feature vector by performing a linear transformation on the graph-level vector. The target feature vector is subjected to a nonlinear transformation to obtain an activated feature representation, which is then input into the fully connected layer for performance prediction calculation to obtain the performance prediction result of the ETFE film. The nonlinear transformation processes the target feature vector through an activation function (such as ReLU, etc.), introduces nonlinear features, and improves the expressiveness of the model.The fully connected layer calculates the performance prediction results of the ETFE film by performing linear transformation on the activated feature representation, thereby achieving accurate prediction of the performance of the ETFE film.
[0035] Perform matrix multiplication on the attention weight matrix and the node feature matrix so that the node features can be weighted according to their importance to obtain a weighted node feature matrix. Add self-loop connections to the adjacency matrix to obtain an enhanced adjacency matrix, and calculate the angle matrix based on the enhanced adjacency matrix. The purpose of adding self-loop connections is to ensure that each node can not only aggregate the information of neighboring nodes, but also retain its own information to obtain a more complete feature expression. The angle matrix is a diagonal matrix whose elements are the degrees of each row of the enhanced adjacency matrix, which is used to normalize the adjacency matrix. Operate on the angle matrix to obtain a normalized matrix, and perform left-right multiplication on the enhanced adjacency matrix based on the normalized matrix to obtain a symmetric normalized adjacency matrix. The normalization operation standardizes the adjacency matrix to prevent the influence of excessive node degree differences on the aggregation process, thereby ensuring the stability and effectiveness of feature aggregation. The symmetric normalized adjacency matrix is obtained by performing bilateral multiplication on the enhanced adjacency matrix, so that the matrix remains symmetric after normalization. Perform matrix multiplication on the symmetric normalized adjacency matrix and the weighted node feature matrix to obtain neighborhood aggregation features. By weighted summing the node features with the features of its neighboring nodes, effective information aggregation is achieved, so that the features of each node can integrate the information of its neighboring nodes to obtain richer feature expressions. Perform feature weight transformation according to the neighborhood aggregation features to obtain transformed features, and add bias items to the transformed features to obtain bias-adjusted features. Feature weight transformation is to further process the aggregated features through a linear transformation to adapt them to subsequent activation operations. Adding bias items is to introduce more flexibility so that the model can better fit the data. Perform nonlinear activation on the bias-adjusted features to obtain activated local features. Nonlinear activation usually uses activation functions such as ReLU. By introducing nonlinear transformations, the model has stronger expression capabilities. Features after nonlinear activation can better capture complex feature relationships. Perform dropout operation on the activated local features to obtain local aggregation features. The dropout operation is to prevent the model from overfitting. By randomly discarding some features, the model can be more robust during training and prevent over-reliance on certain features.
[0036] In the embodiment of the present application, by combining atomic force microscopy, white light interferometer and large-area optical scanning, comprehensive surface morphology data collection from nanometer to macroscopic scale is realized, and the accuracy and integrity of surface feature recognition are improved. A more accurate and detailed comprehensive structural model of ETFE membrane is constructed by adopting technologies such as multi-level structural unit analysis, structural topology map extraction and spatial autocorrelation analysis. The optimized combination of uniaxial tension, biaxial tension and shear test is realized through finite element meshing and orthogonal experimental design, and the mechanical properties of ETFE membrane under complex stress state are comprehensively evaluated. Combined with digital image analysis and X-ray diffraction real-time analysis, detailed stress-strain relationship data and membrane structure evolution data are obtained. Through multi-scale characteristic coefficient calculation, typical correlation analysis and tensor decomposition and other methods, the key performance parameters of ETFE membrane are effectively extracted, and a high-dimensional performance relationship matrix is constructed. Advanced deep learning technology is adopted to achieve high-precision prediction of ETFE membrane performance, improve the prediction model's ability to capture complex nonlinear relationships, enhance the model's learning ability for local structural features, and improve the accuracy of performance prediction.
[0037] In a specific embodiment, the process of executing step S101 may specifically include the following steps:
[0038] (1) Scanning the standard ETFE film test sample with an atomic force microscope to obtain nanoscale surface height data, scanning the standard ETFE film test sample with a white light interferometer to obtain micrometer-level surface profile data, and performing large-area optical scanning on the standard ETFE film test sample to obtain macroscopic-level surface morphology data;
[0039] (2) Performing Fourier transform on the nanoscale surface height data to obtain a surface roughness spectrum, and adaptively filtering the micron-scale surface profile data based on the surface roughness spectrum to obtain an optimized surface profile map;
[0040] (3) Perform wavelet multi-resolution analysis on the optimized surface contour map to obtain multi-scale characteristic coefficients, and then perform regional segmentation on the macroscopic surface morphology data based on the multi-scale characteristic coefficients to obtain surface structure units;
[0041] (4) The morphological features of the surface structural units are extracted to obtain shape descriptors and texture features, and the corresponding ETFE film surface morphology data are constructed based on the shape descriptors and texture features.
[0042] Specifically, an atomic force microscope is used to scan a standard ETFE film test sample to obtain nano-level height data of the sample surface. The atomic force microscope measures the height change of the surface through the interaction between the probe and the sample surface, and generates a three-dimensional image of the height data to understand the nano-level features of the ETFE film surface, such as tiny bumps and depressions. The same standard ETFE film test sample is scanned using a white light interferometer to obtain micron-level surface profile data. The white light interferometer measures the micron-level height change of the sample surface through interference phenomena. When white light is irradiated on the sample surface and reflected, surface points at different heights will cause changes in interference fringes, which can be converted into height data to obtain micron-level surface profile information. The standard ETFE film test sample is scanned using large-area optical scanning technology. Large-area optical scanning can quickly obtain the overall morphological information of the sample surface and provide macro-level surface features, such as large-scale ripples and overall morphological changes. Fourier transform is performed on the nano-level surface height data. Fourier transform is used to convert height data in the spatial domain into a surface roughness spectrum in the frequency domain. The Fourier transform formula is:
[0043] ;
[0044] in, represents the transformation result in the frequency domain, Represents the height data in the spatial domain, and are the horizontal and vertical dimensions of the data, and is a frequency variable. Through Fourier transform, the nanoscale surface height data is decomposed into different frequency components to obtain the surface roughness spectrum. According to the obtained surface roughness spectrum, the micron-level surface profile data is adaptively filtered to obtain the optimized surface profile. Adaptive filtering selectively filters the micron-level surface data by analyzing the importance of different frequency components in the surface roughness spectrum, removes noise and retains key structural information. Wavelet multi-resolution analysis is performed on the optimized surface profile to extract multi-scale feature coefficients. Wavelet transform is a mathematical tool for signal processing that performs analysis in both the time domain and the frequency domain. The wavelet transform formula is:
[0045] ;
[0046] in, represents the wavelet transform coefficients, is the signal to be analyzed, is the wavelet function, and Represent the scale and translation parameters respectively. Through wavelet transform, characteristic coefficients at different scales are obtained to reflect the multi-scale characteristics of the sample surface. According to the multi-scale characteristic coefficients, the macroscopic surface morphology data is segmented to obtain surface structural units. Regional segmentation identifies and divides regions with similar characteristics by analyzing the characteristic coefficients of different regions to form multiple surface structural units. Morphological features are extracted from the surface structural units to obtain shape descriptors and texture features. Morphological feature extraction extracts descriptors that can characterize these features by analyzing the geometric shape and surface texture of the structural units. Shape descriptors such as roundness and aspect ratio can describe the geometric characteristics of the structural units; texture features such as texture direction and texture density can describe the surface characteristics of the structural units. According to the shape descriptors and texture features, the corresponding ETFE film surface morphology data is constructed.
[0047] In a specific embodiment, the process of executing step S102 may specifically include the following steps:
[0048] (1) Decomposing the surface morphology data of ETFE film to obtain characteristic coefficient matrices of different scales, and performing adaptive threshold segmentation on the surface morphology data according to the characteristic coefficient matrix to obtain multi-level structural units;
[0049] (2) Perform morphological skeleton extraction on the multi-level structural units to obtain a structural topology map, and perform region growing processing on the multi-level structural units based on the structural topology map to obtain a structural domain;
[0050] (3) Calculate the gray-level co-occurrence matrix of the structural domain to obtain the texture feature vector, and classify the structural domain according to the texture feature vector to obtain the structural type label;
[0051] (4) Perform spatial autocorrelation analysis on the structural type labels to obtain the structural distribution pattern, and construct a preliminary three-dimensional structural representation of the ETFE membrane based on the structural distribution pattern;
[0052] (5) Performing tensor decomposition on the preliminary three-dimensional structural representation to obtain a structural feature tensor, and generating a comprehensive structural model of the ETFE membrane based on the structural feature tensor.
[0053] Specifically, the surface data is analyzed at multiple scales to obtain characteristic coefficient matrices at different scales. Wavelet transform is used to analyze the signal at different scales and capture time and frequency information at the same time. The formula of wavelet transform is:
[0054] ;
[0055] in, represents the wavelet transform coefficients, is the signal to be analyzed (the surface morphology data of the ETFE film), is the wavelet function, is the scale parameter, which indicates the degree of expansion or contraction of the wavelet function. is the translation parameter, which is used to translate the wavelet function to cover the entire signal. By performing wavelet transform on the surface data, characteristic coefficient matrices of different scales are obtained, which contain the characteristic information of the ETFE film at different scales. Adaptive threshold segmentation is performed on the surface morphology data according to the characteristic coefficient matrix to obtain multi-level structural units. Adaptive threshold segmentation is a method of dynamically adjusting the threshold. According to the characteristic changes of different scales in the characteristic coefficient matrix, the segmentation threshold is adaptively determined to divide the surface data into multiple levels of structural units. The structural unit reflects the surface characteristics of the ETFE film at different scales. Morphological skeleton extraction is performed on the multi-level structural unit to obtain a structural topology map. Morphological skeleton extraction gradually removes the edge pixels of the structural unit and retains its central skeleton to form a skeleton map reflecting the topological characteristics of the structural unit. According to the structural topology map, the multi-level structural unit is subjected to region growing processing to obtain a structural domain. The region growing processing starts from an initial point and gradually adds neighboring pixels to the region to form a connected structural domain. This process can be performed according to the connection relationship in the structural topology map to ensure the integrity and connectivity of the structural domain. The gray-level co-occurrence matrix of the structural domain is calculated to obtain the texture feature vector. Gray-level co-occurrence matrix is a statistical method to describe image texture. By calculating the co-occurrence relationship of gray value pairs in the image, a series of texture features are obtained. According to the texture feature vector, the structural domain is classified to obtain the structural type label. The classification process can use a variety of machine learning algorithms, such as support vector machines or convolutional neural networks. The structural domain is classified by training the model to generate the corresponding structural type label. Spatial autocorrelation analysis is performed on the structural type label to obtain the structural distribution pattern. Spatial autocorrelation analysis reveals the spatial distribution law by measuring the correlation between structural type labels at different positions. The Moran index is a commonly used spatial autocorrelation indicator, and its formula is:
[0056] ;
[0057] in, is the total number of observations, is the sum of the weight matrices, are the elements of the spatial weight matrix, and is the observed value, is the average value of the observed values. The calculation of the Moran index can reveal the spatial distribution pattern of the structural type label. According to the structural distribution pattern, a preliminary three-dimensional structural representation of the ETFE membrane is constructed. By extending the two-dimensional structural distribution information to three-dimensional space, a preliminary three-dimensional structural model of the ETFE membrane is generated. This model preliminarily reflects the structural characteristics and spatial distribution of the membrane at different levels. The preliminary three-dimensional structural representation is tensor decomposed to obtain a structural feature tensor. Tensor decomposition is a high-dimensional data decomposition method that extracts the main characteristic components in the structure by decomposing the three-dimensional structural representation into core tensors and factor matrices. The tensor decomposition formula is:
[0058] ;
[0059] in, is a tensor representing the preliminary three-dimensional structure, is the core tensor, It is a factor matrix that simplifies the complex structural information into several main characteristic components through tensor decomposition. Based on the structural characteristic tensor, a comprehensive structural model of the ETFE membrane is generated.
[0060] In a specific embodiment, the process of executing step S103 may specifically include the following steps:
[0061] (1) Perform finite element meshing on the comprehensive structural model of the ETFE membrane to obtain a discretized calculation model, and design a multi-mode mechanical loading scheme based on the discretized calculation model to obtain a loading sequence matrix;
[0062] (2) An orthogonal experimental design was performed on the loading sequence matrix to obtain the optimized loading parameter combination. Based on the optimized loading parameter combination, a uniaxial tensile test was performed on the comprehensive structural model of the ETFE membrane to obtain the uniaxial stress-strain data.
[0063] (3) Perform real-time digital image analysis on the uniaxial stress-strain data to obtain the local strain field distribution, and perform biaxial tensile testing on the comprehensive structural model of the ETFE membrane based on the local strain field distribution to obtain the biaxial stress-strain data;
[0064] (4) Perform principal stress analysis on the biaxial stress-strain data to obtain the stress principal axis direction field, and perform shear testing on the comprehensive structural model of the ETFE membrane based on the stress principal axis direction field to obtain shear stress-strain data;
[0065] (5) Perform Fourier transform on the shear stress-strain data to obtain stress-strain relationship data, and perform X-ray diffraction real-time analysis on the membrane structure changes during the comprehensive structural model test to obtain membrane structure evolution data.
[0066] Specifically, the continuous structural model is discretized to generate a finite element mesh. Finite element meshing decomposes the structural model into a number of finite small units, which are connected by nodes to form a whole. In this way, complex structural and mechanical problems are converted into discrete computational problems, which can be numerically solved by computer. In the process of finite element meshing, different types of units are usually used, such as quadrilateral units, triangular units or hexahedral units, to adapt to the geometric shape and analysis requirements of the structure. According to the discretized computational model, a multi-mode mechanical loading scheme is designed to obtain a loading sequence matrix. The multi-mode mechanical loading scheme includes different types of mechanical loading, such as tension, compression, shear, etc., which aims to comprehensively evaluate the performance of ETFE membrane under various mechanical conditions. The loading sequence matrix records the specific loading parameters and sequence under each loading mode to ensure the repeatability of the experiment and the reliability of the results. The loading sequence matrix is subjected to orthogonal experimental design to obtain the optimized loading parameter combination. Orthogonal experimental design is a systematic experimental design method that obtains as much information as possible with as few experiments as possible by reasonably arranging the experiments. Orthogonal experimental design arranges multiple experiments so that each experiment contains different level combinations of several factors, thereby systematically studying these factors and their interactions. The optimized loading parameter combination can comprehensively evaluate the mechanical properties of ETFE membrane while minimizing the number of experiments. According to the optimized loading parameter combination, the comprehensive structural model of ETFE membrane is subjected to uniaxial tensile test to obtain uniaxial stress-strain data. In the uniaxial tensile test, the tensile force in a single direction is applied to measure the stress-strain relationship of ETFE membrane in that direction. Stress The relationship between stress and strain (ε) is usually expressed by a stress-strain curve, and the formula is:
[0067] ;
[0068] in, represents stress, represents the elastic modulus of the material, Represents strain. Through uniaxial tensile testing, key mechanical performance parameters such as elastic modulus, yield strength and fracture strength of ETFE film are obtained. Real-time digital image analysis of uniaxial stress-strain data is performed to obtain the local strain field distribution. Digital image correlation method is a non-contact optical technology used to measure the surface deformation of an object. By tracking the changes in the surface image, the local strain field distribution can be accurately measured. The basic formula of the digital image correlation method is:
[0069] ;
[0070] in, Indicates at location The strain at and Along and The displacement increment in the direction. According to the local strain field distribution, the comprehensive structural model of the ETFE membrane is subjected to a biaxial tensile test to obtain the biaxial stress-strain data. The biaxial tensile test measures the stress-strain relationship of the material under biaxial loading conditions by applying tensile forces in two directions at the same time. The biaxial stress-strain data can reflect the mechanical properties of the ETFE membrane under complex stress states. The principal stress analysis is performed on the biaxial stress-strain data to obtain the stress principal axis direction field. The principal stress analysis determines the direction and magnitude of the principal stresses by calculating the principal stresses inside the material. is the maximum and minimum normal stress of the material in a specific direction, and the formula is:
[0071] ;
[0072] in, and are the normal stresses along the x-axis and y-axis, respectively, is the shear stress in the xy plane. Through principal stress analysis, the stress distribution and stress concentration area inside the material are determined. According to the stress principal axis direction field, the shear test is performed on the comprehensive structural model of the ETFE film to obtain the shear stress-strain data. In the shear test, the stress-strain relationship of the material under shear conditions is measured by applying shear force. Shear stress ( ) and shear strain ( The relationship is usually expressed in terms of shear modulus ( ) indicates that the formula is:
[0073] ;
[0074] in, represents the shear stress, represents the shear modulus, Represents shear strain. Shear test can evaluate the shear strength and deformation capacity of ETFE film. Perform Fourier transform on shear stress-strain data to obtain stress-strain relationship data. Fourier transform can convert stress-strain data from time domain or space domain to frequency domain, revealing the frequency components and characteristics in the data. The Fourier transform formula is:
[0075] ;
[0076] in, represents the transformation result in the frequency domain, Represents a signal in the time domain or space domain. The main frequency components and change patterns in the stress-strain data are analyzed through Fourier transform. During the test of the comprehensive structural model, the membrane structure changes are analyzed in real time by X-ray diffraction to obtain the membrane structure evolution data. X-ray diffraction technology can monitor the crystal structure changes inside the material in real time by detecting the diffraction pattern of X-rays inside the material. When X-rays are irradiated to the crystal, the atoms in the crystal will diffract the X-rays to form a diffraction pattern. By analyzing these diffraction patterns, the crystal structure information and changes of the material can be obtained.
[0077] In a specific embodiment, the process of executing step S104 may specifically include the following steps:
[0078] (1) Calculate the corresponding multi-scale stress-strain characteristic coefficients based on the stress-strain relationship data, and perform characteristic analysis on the membrane structure evolution data to obtain the main structure evolution mode;
[0079] (2) Based on the multi-scale stress-strain characteristic coefficients and the main structural evolution patterns, canonical correlation analysis is performed to obtain stress-structure correlation characteristics;
[0080] (3) Perform nonlinear dimensionality reduction processing on the stress-structure correlation characteristics to obtain a low-dimensional representation vector, and construct a stress-structure mapping function based on the low-dimensional representation vector;
[0081] (4) Perform sensitivity analysis on the stress-structure mapping function to obtain key performance parameters, and resample and enhance the key performance parameters to obtain an extended data set;
[0082] (5) Perform tensor decomposition on the extended data set to obtain a multimodal feature tensor, and calculate the core tensor and factor matrix based on the multimodal feature tensor;
[0083] (6) Perform tensor network transformation on the core tensor and factor matrix to obtain the performance relationship matrix of ETFE membrane.
[0084] Specifically, the collected stress-strain relationship data is subjected to multi-scale analysis. Multi-scale analysis decomposes the signal and represents its features at different scales to capture all information from micro to macro. For example, using wavelet transform, the formula is as follows:
[0085] ;
[0086] in, represents the wavelet transform coefficients, is the stress-strain signal, is the wavelet function, and Represent the scale and translation parameters respectively. Through wavelet transform, the characteristic coefficients at different scales are extracted to construct a multi-scale stress-strain characteristic coefficient matrix. These characteristic coefficients can fully reflect the mechanical properties of ETFE membranes at different stress levels. The characteristic analysis of the membrane structure evolution data is performed to obtain the main structural evolution mode. The characteristic analysis can use principal component analysis or other statistical methods to extract the main mode of structural evolution through dimensionality reduction technology. Principal component analysis projects the data into a low-dimensional space composed of principal components through eigenvalue decomposition, retaining the most important characteristic information. Canonical correlation analysis is performed based on the multi-scale stress-strain characteristic coefficients and the main structural evolution mode to obtain stress-structure correlation characteristics. Canonical correlation analysis is a multivariate statistical method used to analyze the correlation between two sets of variables. The goal of canonical correlation analysis is to find a linear transformation that maximizes the correlation between the two sets of variables after the transformation. Through canonical correlation analysis, the correlation characteristics between stress characteristics and structural characteristics are obtained. The stress-structure correlation characteristics are subjected to nonlinear dimensionality reduction processing to obtain a low-dimensional representation vector. Nonlinear dimensionality reduction methods such as t-SNE or UMAP can effectively map high-dimensional data to low-dimensional space while retaining the global and local structures of the data. The low-dimensional characterization vector can simplify the data structure. Based on the low-dimensional characterization vector, a stress-structure mapping function is constructed. The mapping function can be trained using a machine learning model such as a neural network, and a nonlinear mapping model is established by fitting the relationship between stress characteristics and structural characteristics. The mapping function can convert stress characteristics into corresponding structural characteristics. A sensitivity analysis is performed on the constructed stress-structure mapping function to determine the key performance parameters. Sensitivity analysis determines the key parameters that have the greatest impact on the results by analyzing the impact of changes in input parameters on the output results. These key performance parameters can help optimize the design and improve the robustness of the model. For example, local sensitivity analysis and global sensitivity analysis are used to identify the parameters that have the greatest impact on the performance of ETFE membranes. Resampling and enhancement are performed according to the key performance parameters to obtain an extended data set. Data enhancement techniques such as random noise addition, rotation, and translation can expand the size of the data set and improve the generalization ability of the model. Resampling techniques such as SMOTE can balance the sample distribution in the data set to avoid the model being biased towards certain specific samples. The extended data set is tensor decomposed to obtain a multimodal feature tensor. Tensor decomposition is a high-dimensional data processing method that can extract the main features of the data by decomposing the high-dimensional tensor into low-dimensional core tensors and factor matrices. The tensor decomposition formula is as follows:
[0087] ;
[0088] in, represents the original data tensor, is the core tensor, It is a factor matrix. Through tensor decomposition, complex multimodal data is converted into a low-dimensional representation that is easy to handle. According to the multimodal feature tensor, the core tensor and factor matrix are calculated, and the core tensor and factor matrix are transformed into a tensor network to obtain the performance relationship matrix of the ETFE membrane. Tensor network transformation is an advanced data processing technology. By further processing the core tensor and factor matrix, a performance relationship matrix is constructed to fully reflect the mechanical properties and structural characteristics of the ETFE membrane.
[0089] In a specific embodiment, the process of executing step S105 may specifically include the following steps:
[0090] (1) The performance relationship matrix of ETFE film is input into the attention-based recurrent graph convolutional network model, and the performance relationship matrix of ETFE film is encoded by graph structure, and the performance relationship matrix is converted into an adjacency matrix and a node feature matrix to obtain the initial graph representation;
[0091] (2) Calculate the multi-head self-attention according to the initial graph representation, and perform linear transformation, dot product and softmax operations on each attention head to obtain the attention weight matrix;
[0092] (3) Perform matrix multiplication on the attention weight matrix and the node feature matrix and multiply them with the adjacency matrix to perform graph convolution operations to obtain local aggregate features;
[0093] (4) Input the local aggregated features into the gated recurrent unit, and obtain the hidden state features by updating the gate, resetting the gate, and calculating the candidate hidden state;
[0094] (5) Perform residual connection on the hidden state features to obtain multi-scale fusion features, calculate the mean and variance of the multi-scale fusion features, perform layer normalization operation, and obtain standardized features with a mean of 0 and a variance of 1;
[0095] (6) Perform a global average pooling operation on the standardized features to obtain a graph-level vector representation, and input the graph-level vector representation into the fully connected layer. By multiplying the weight matrix and adding the bias term, the target feature vector is obtained.
[0096] (7) Perform a nonlinear transformation on the target feature vector to obtain an activated feature representation, and input the activated feature representation into the fully connected layer for performance prediction calculation to obtain the performance prediction result of the ETFE film.
[0097] Specifically, the performance relationship matrix of the ETFE film is encoded into a graph structure, and the performance relationship matrix is converted into a graph representation, including an adjacency matrix and a node feature matrix. The adjacency matrix represents the connection relationship between nodes in the graph, while the node feature matrix contains the attribute information of each node. Multi-head self-attention is calculated based on the initial graph representation. The multi-head self-attention mechanism captures the relationship between different parts of the graph by calculating multiple attention heads in parallel, thereby enhancing the expressive power of the model. For each attention head, a linear transformation is performed to map the node feature matrix to a low-dimensional space. A dot product operation is performed to calculate the similarity between nodes, and the formula is:
[0098] ;
[0099] in, is the dimension of the key. Through the softmax operation, we get the attention weight matrix and apply it to the value matrix The output of each attention head is obtained. The output results of multiple attention heads are concatenated to obtain the overall attention output. Matrix multiplication is performed on the attention weight matrix and the node feature matrix, and multiplied with the adjacency matrix to perform a graph convolution operation. The graph convolution operation achieves effective information transmission and fusion by aggregating the features of a node with the features of its neighboring nodes. The formula for graph convolution is:
[0100] ;
[0101] in, is the updated node feature matrix, is the normalized adjacency matrix, is the weight matrix, is an activation function (such as ReLU). Through this operation, the local aggregate features of each node are obtained. The local aggregate features are input into the gated recurrent unit, and the hidden state features are obtained by updating the gate, resetting the gate, and calculating the candidate hidden state. The gated recurrent unit is a commonly used recurrent neural network unit that models sequence data by controlling the flow and memory of information. The formula of the gated recurrent unit is as follows:
[0102] ;
[0103] in, It is the update gate. It is the reset gate. is a candidate hidden state, is the current hidden state, is the sigmoid function, is an element-by-element multiplication operation. Through the calculation of the gated recurrent unit, the hidden state features of each node in the sequence are obtained. The hidden state features are residually connected to obtain multi-scale fusion features. The residual connection realizes the fusion of multi-scale features by adding the original features in the feature calculation process. The formula is:
[0104] ;
[0105] The mean and variance of the multi-scale fusion features are calculated, and the layer normalization operation is performed so that the mean of the features is 0 and the variance is 1, thereby improving numerical stability and training efficiency. The standardized features are subjected to global average pooling to obtain a graph-level vector representation. Global average pooling generates a global graph representation by averaging all node features. The formula is:
[0106] ;
[0107] in, is the graph-level vector representation, is the node feature, is the number of nodes. The graph-level vector representation is input into the fully connected layer, and the target feature vector is obtained by weight matrix multiplication and bias term addition. The formula is:
[0108] ;
[0109] in, is the weight matrix of the fully connected layer, is the bias term. Perform a nonlinear transformation on the target feature vector to obtain the activated feature representation. Nonlinear transformation usually uses an activation function (such as ReLU), and the formula is:
[0110] ;
[0111] in, is the activation function. The activated feature representation is input into the fully connected layer to perform performance prediction calculation, and finally the performance prediction result of the ETFE film is obtained.
[0112] In a specific embodiment, the execution step performs matrix multiplication on the attention weight matrix and the node feature matrix and multiplies them with the adjacency matrix, performs a graph convolution operation, and obtains the local aggregation feature. The process may specifically include the following steps:
[0113] (1) Perform matrix multiplication on the attention weight matrix and the node feature matrix to obtain the weighted node feature matrix;
[0114] (2) Add self-loop connections to the adjacency matrix to obtain an enhanced adjacency matrix, and calculate the angle matrix based on the enhanced adjacency matrix;
[0115] (3) Operate the angle matrix to obtain a normalized matrix, and perform left-right multiplication on the enhanced adjacency matrix according to the normalized matrix to obtain a symmetric normalized adjacency matrix;
[0116] (4) Perform matrix multiplication on the symmetric normalized adjacency matrix and the weighted node feature matrix to obtain the neighborhood aggregation features;
[0117] (5) Perform feature weight transformation according to the neighborhood aggregation features to obtain the transformed features, and add bias items to the transformed features to obtain bias-adjusted features;
[0118] (6) Perform nonlinear activation on the bias-adjusted features to obtain activated local features, and perform dropout operation on the activated local features to obtain local aggregated features.
[0119] Specifically, the attention weight matrix and the node feature matrix are dimensionally matched. Assume that the node feature matrix The dimension is ,in is the number of nodes, is the feature dimension of each node. Attention weight matrix The dimension is , represents the attention weight between nodes. Through matrix multiplication , get the weighted node feature matrix , its dimension is still . For the adjacency matrix Add self-loop connections to obtain an enhanced adjacency matrix , ensuring that each node retains its own characteristics while aggregating the characteristics of its neighbors. The operation formula for adding self-loop connections is:
[0120] ;
[0121] in, is the identity matrix, which adds self-loop connections on the diagonal of the adjacency matrix. The diagonal matrix is calculated based on the enhanced adjacency matrix , whose elements is the enhanced adjacency matrix No. The sum of all elements in the row. The formula for the angle matrix is:
[0122] ;
[0123] Operate the angle matrix to get the normalized matrix The purpose of the normalization matrix is to standardize the adjacency matrix to prevent the influence of different node degrees. According to the normalization matrix, the enhanced adjacency matrix is multiplied left and right to obtain the symmetric normalized adjacency matrix The formula for the symmetric normalized adjacency matrix is:
[0124] ;
[0125] Ensure that the adjacency matrix remains symmetric after normalization, so that information propagation is more stable in the graph structure. Perform matrix multiplication on the symmetric normalized adjacency matrix and the weighted node feature matrix to obtain the neighborhood aggregation feature By taking a weighted average of the features of a node and its neighboring nodes, effective information aggregation is achieved. The formula is:
[0126] ;
[0127] in, It is the neighborhood aggregation feature matrix, which reflects the aggregation information of each node after considering the features of neighboring nodes. The feature weights are transformed according to the neighborhood aggregation features to obtain the transformed features. The feature weights are transformed through a linear transformation matrix , reproject the aggregated features, the formula is:
[0128] ;
[0129] in, is a weight matrix, which is learned through training and maps the aggregated features to the new feature space. Add a bias term to the transformed features , and obtain the bias adjustment characteristic The formula for adding the bias term is:
[0130] ;
[0131] in, is the bias vector, which is obtained through training and learning, and can add more flexibility to feature transformation. Nonlinear activation is performed on the bias adjustment feature to obtain the activated local feature . Non-linear activation usually uses the ReLU activation function, the formula is:
[0132] ;
[0133] Among them, the ReLU function is defined as , nonlinearity is introduced, so that the model can capture more complex feature relationships. Dropout operation is performed on the activated local features to obtain the final local aggregation features Dropout is a regularization technique that prevents overfitting by randomly discarding some features. The formula is:
[0134] ;
[0135] in, is the probability of dropout, indicating the proportion of discarded features.
[0136] The above describes the ETFE film performance test method in the embodiment of the present application. The following describes the ETFE film performance test device in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the ETFE film performance testing device includes:
[0137] A scanning module 201 is used to perform multi-scale surface scanning on a standard ETFE film test sample to obtain surface morphology data of the ETFE film;
[0138] The recognition module 202 is used to perform comprehensive film structure feature recognition on the surface morphology data of the ETFE film and construct a comprehensive structure model of the ETFE film;
[0139] The acquisition module 203 is used to perform multi-mode mechanical loading and data acquisition on the comprehensive structural model of the ETFE membrane to obtain stress-strain relationship data and membrane structure evolution data;
[0140] The conversion module 204 is used to perform feature extraction and relationship matrix conversion on the stress-strain relationship data and the membrane structure evolution data to obtain the performance relationship matrix of the ETFE membrane;
[0141] The prediction module 205 is used to input the performance relationship matrix of the ETFE film into the attention-based recurrent graph convolutional network model to perform performance prediction and obtain the ETFE film performance prediction result.
[0142] Through the synergy of the above components, combined with atomic force microscopy, white light interferometer and large-area optical scanning, comprehensive surface morphology data collection from nanometer to macro scale is achieved, and the accuracy and completeness of surface feature recognition are improved. A more accurate and detailed comprehensive structural model of ETFE membrane is constructed by adopting technologies such as multi-level structural unit analysis, structural topology map extraction and spatial autocorrelation analysis. Through finite element meshing and orthogonal experimental design, the optimal combination of uniaxial tension, biaxial tension and shear test is achieved, and the mechanical properties of ETFE membrane under complex stress state are comprehensively evaluated. Combined with digital image analysis and X-ray diffraction real-time analysis, detailed stress-strain relationship data and membrane structure evolution data are obtained. Through multi-scale characteristic coefficient calculation, canonical correlation analysis and tensor decomposition, the key performance parameters of ETFE membrane are effectively extracted, and a high-dimensional performance relationship matrix is constructed. Advanced deep learning technology is adopted to achieve high-precision prediction of ETFE membrane performance, improve the prediction model's ability to capture complex nonlinear relationships, enhance the model's learning ability for local structural features, and improve the accuracy of performance prediction.
[0143] The present application also provides an ETFE film performance testing system, which includes a memory and a processor. The memory stores computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor executes the steps of the ETFE film performance testing method in the above-mentioned embodiments.
[0144] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the ETFE film performance testing method.
[0145] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0146] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer system (which can be a personal computer, server, or network system, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.
[0147] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. A method for testing the performance of ETFE film, characterized in that: The ETFE film performance testing method comprises: Perform multi-scale surface scanning on the standard ETFE film test sample to obtain the surface morphology data of the ETFE film; Performing comprehensive membrane structure feature recognition on the surface morphology data of the ETFE membrane to construct a comprehensive structural model of the ETFE membrane; The comprehensive structural model of the ETFE membrane is subjected to multi-mode mechanical loading and data collection to obtain stress-strain relationship data and membrane structure evolution data; specifically comprising: performing finite element meshing on the comprehensive structural model of the ETFE membrane to obtain a discretized calculation model, and designing a multi-mode mechanical loading scheme according to the discretized calculation model to obtain a loading sequence matrix; performing orthogonal experimental design on the loading sequence matrix to obtain an optimized loading parameter combination, and performing a uniaxial tensile test on the comprehensive structural model of the ETFE membrane according to the optimized loading parameter combination to obtain uniaxial stress-strain data; and performing an orthogonal experimental design on the loading sequence matrix to obtain an optimized loading parameter combination to obtain uniaxial stress-strain data. Performing real-time digital image analysis to obtain local strain field distribution, and performing biaxial tensile testing on the overall structural model of the ETFE membrane according to the local strain field distribution to obtain biaxial stress-strain data; performing principal stress analysis on the biaxial stress-strain data to obtain stress principal axis direction field, and performing shear testing on the overall structural model of the ETFE membrane according to the stress principal axis direction field to obtain shear stress-strain data; performing Fourier transformation on the shear stress-strain data to obtain stress-strain relationship data, and performing X-ray diffraction real-time analysis on membrane structure changes during the overall structural model testing process to obtain membrane structure evolution data; The stress-strain relationship data and the membrane structure evolution data are subjected to feature extraction and relationship matrix conversion to obtain the performance relationship matrix of the ETFE membrane; specifically, the method comprises: calculating the corresponding multi-scale stress-strain characteristic coefficients according to the stress-strain relationship data, and performing feature analysis on the membrane structure evolution data to obtain the main structure evolution mode; performing canonical correlation analysis on the multi-scale stress-strain characteristic coefficients and the main structure evolution mode to obtain stress-structure correlation characteristics; performing nonlinear dimensionality reduction processing on the stress-structure correlation characteristics to obtain a low-dimensional characterization vector, and constructing a stress-structure mapping function according to the low-dimensional characterization vector; performing sensitivity analysis on the stress-structure mapping function to obtain key performance parameters, and performing resampling and enhancement according to the key performance parameters to obtain an extended data set; performing tensor decomposition on the extended data set to obtain a multi-modal characteristic tensor, and calculating a core tensor and a factor matrix according to the multi-modal characteristic tensor; performing tensor network transformation on the core tensor and the factor matrix to obtain the performance relationship matrix of the ETFE membrane; The performance relationship matrix of the ETFE film is input into the attention-based recurrent graph convolutional network model to perform performance prediction, and the ETFE film performance prediction result is obtained.
2. The ETFE film performance testing method according to claim 1, wherein: The multi-scale surface scanning of the standard ETFE film test sample is performed to obtain the surface morphology data of the ETFE film, including: Scanning the standard ETFE film test sample with an atomic force microscope to obtain nanometer-level surface height data, scanning the standard ETFE film test sample with a white light interferometer to obtain micrometer-level surface profile data, and performing large-area optical scanning on the standard ETFE film test sample to obtain macroscopic-level surface morphology data; Performing Fourier transformation on the nanometer-level surface height data to obtain a surface roughness spectrum, and adaptively filtering the micrometer-level surface profile data according to the surface roughness spectrum to obtain an optimized surface profile map; Performing wavelet multi-resolution analysis on the optimized surface contour map to obtain multi-scale characteristic coefficients, and performing regional segmentation on the macroscopic surface morphology data according to the multi-scale characteristic coefficients to obtain surface structure units; Morphological features are extracted from the surface structural unit to obtain shape descriptors and texture features, and corresponding ETFE film surface morphology data are constructed according to the shape descriptors and the texture features.
3. The ETFE film performance testing method according to claim 2, wherein: The method of performing comprehensive film structure feature recognition on the surface morphology data of the ETFE film and constructing a comprehensive structure model of the ETFE film comprises: Decomposing the surface morphology data of the ETFE film to obtain characteristic coefficient matrices of different scales, and performing adaptive threshold segmentation on the surface morphology data according to the characteristic coefficient matrix to obtain a multi-level structural unit; Performing morphological skeleton extraction on the multi-level structural unit to obtain a structural topology map, and performing region growing processing on the multi-level structural unit according to the structural topology map to obtain a structural domain; Calculating the gray level co-occurrence matrix of the structural domain to obtain a texture feature vector, and classifying the structural domain according to the texture feature vector to obtain a structural type label; Performing spatial autocorrelation analysis on the structure type labels to obtain a structure distribution pattern, and constructing a preliminary three-dimensional structure representation of the ETFE film according to the structure distribution pattern; The preliminary three-dimensional structure representation is subjected to tensor decomposition to obtain a structural feature tensor, and a comprehensive structural model of the ETFE membrane is generated based on the structural feature tensor.
4. The ETFE film performance testing method according to claim 1, wherein: The performance relationship matrix of the ETFE film is input into the attention-based recurrent graph convolutional network model to perform performance prediction, and the ETFE film performance prediction result is obtained, including: Inputting the performance relationship matrix of the ETFE film into an attention-based recurrent graph convolutional network model, performing graph structure encoding on the performance relationship matrix of the ETFE film, converting the performance relationship matrix into an adjacency matrix and a node feature matrix, and obtaining an initial graph representation; Calculate multi-head self-attention according to the initial graph representation, and perform linear transformation, dot product and softmax operations on each attention head to obtain an attention weight matrix; Performing a matrix multiplication operation on the attention weight matrix and the node feature matrix and multiplying them with the adjacency matrix, performing a graph convolution operation, and obtaining local aggregation features; Input the local aggregated features into a gated recurrent unit, and obtain hidden state features by calculating an update gate, a reset gate, and candidate hidden states; Performing residual connection on the hidden state features to obtain multi-scale fusion features, calculating the mean and variance of the multi-scale fusion features, performing layer normalization operation, and obtaining standardized features with a mean of 0 and a variance of 1; Performing a global average pooling operation on the standardized features to obtain a graph-level vector representation, and inputting the graph-level vector representation into a fully connected layer, and obtaining a target feature vector by weight matrix multiplication and bias term addition; The target feature vector is subjected to nonlinear transformation to obtain an activated feature representation, and the activated feature representation is input into a fully connected layer to perform performance prediction calculation to obtain a performance prediction result of the ETFE film.
5. The ETFE film performance testing method according to claim 4, characterized in that: The matrix multiplication operation is performed on the attention weight matrix and the node feature matrix and multiplied with the adjacency matrix, and a graph convolution operation is performed to obtain local aggregation features, including: Performing a matrix multiplication operation on the attention weight matrix and the node feature matrix to obtain a weighted node feature matrix; Adding self-loop connections to the adjacency matrix to obtain an enhanced adjacency matrix, and calculating an angle matrix based on the enhanced adjacency matrix; The angle matrix is operated to obtain a normalized matrix, and the enhanced adjacency matrix is multiplied left and right according to the normalized matrix to obtain a symmetric normalized adjacency matrix; Performing a matrix multiplication operation on the symmetric normalized adjacency matrix and the weighted node feature matrix to obtain neighborhood aggregation features; Performing feature weight transformation according to the neighborhood aggregation feature to obtain transformed features, and adding a bias item to the transformed features to obtain a bias-adjusted feature; The bias adjustment feature is nonlinearly activated to obtain an activated local feature, and a dropout operation is performed on the activated local feature to obtain a local aggregation feature.
6. An ETFE film performance testing device, characterized in that: Used to perform the ETFE film performance testing method as described in any one of claims 1 to 5, the device comprises: A scanning module is used to perform multi-scale surface scanning on a standard ETFE film test sample to obtain surface morphology data of the ETFE film; An identification module is used to perform comprehensive membrane structure feature identification on the surface morphology data of the ETFE membrane and construct a comprehensive structure model of the ETFE membrane; The acquisition module is used for performing multi-mode mechanical loading and data acquisition on the comprehensive structural model of the ETFE membrane to obtain stress-strain relationship data and membrane structure evolution data; specifically comprising: performing finite element meshing on the comprehensive structural model of the ETFE membrane to obtain a discretized calculation model, and designing a multi-mode mechanical loading scheme according to the discretized calculation model to obtain a loading sequence matrix; performing orthogonal experimental design on the loading sequence matrix to obtain an optimized loading parameter combination, and performing a uniaxial tensile test on the comprehensive structural model of the ETFE membrane according to the optimized loading parameter combination to obtain uniaxial stress-strain data; and performing an orthogonal experimental design on the loading sequence matrix to obtain an optimized loading parameter combination. The ETFE film is subjected to real-time digital image analysis on the strain data to obtain the local strain field distribution, and the overall structural model of the ETFE film is subjected to biaxial tensile test according to the local strain field distribution to obtain the biaxial stress-strain data; the biaxial stress-strain data is subjected to principal stress analysis to obtain the stress principal axis direction field, and the overall structural model of the ETFE film is subjected to shear test according to the stress principal axis direction field to obtain the shear stress-strain data; the shear stress-strain data is subjected to Fourier transformation to obtain the stress-strain relationship data, and the membrane structure change during the overall structural model test is subjected to X-ray diffraction real-time analysis to obtain the membrane structure evolution data; A conversion module is used to perform feature extraction and relationship matrix conversion on the stress-strain relationship data and the membrane structure evolution data to obtain a performance relationship matrix of the ETFE membrane; specifically comprising: calculating the corresponding multi-scale stress-strain characteristic coefficients according to the stress-strain relationship data, and performing feature analysis on the membrane structure evolution data to obtain the main structure evolution mode; performing canonical correlation analysis on the multi-scale stress-strain characteristic coefficients and the main structure evolution mode to obtain stress-structure correlation characteristics; performing nonlinear dimensionality reduction processing on the stress-structure correlation characteristics to obtain a low-dimensional characterization vector, and constructing a stress-structure mapping function based on the low-dimensional characterization vector; performing sensitivity analysis on the stress-structure mapping function to obtain key performance parameters, and resampling and enhancing according to the key performance parameters to obtain an extended data set; performing tensor decomposition on the extended data set to obtain a multi-modal characteristic tensor, and calculating a core tensor and a factor matrix based on the multi-modal characteristic tensor; performing tensor network transformation on the core tensor and the factor matrix to obtain a performance relationship matrix of the ETFE membrane; The prediction module is used to input the performance relationship matrix of the ETFE film into the attention-based recurrent graph convolutional network model to perform performance prediction and obtain the ETFE film performance prediction result.
7. An ETFE film performance testing system, characterized in that: The ETFE film performance testing system comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instructions in the memory so that the ETFE film performance testing system executes the ETFE film performance testing method as described in any one of claims 1-5.
8. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the ETFE film performance testing method according to any one of claims 1 to 5 is implemented.
