Online monitoring system and method for composite cloth production line

Abstract

The invention discloses an on-line monitoring system and method for a composite cloth production line, and relates to the technical field of automatic control, and the method comprises the steps: carrying out the space-time registration of a data set of the composite cloth production line through employing an iterative nearest point algorithm, obtaining the spatial registration data, carrying out the multi-source feature aggregation of the spatial registration data, outputting a composite cloth quality feature matrix, and carrying out the real-time monitoring of the composite cloth production line. Inputting the composite cloth quality feature matrix into a hot melt penetration prediction model, performing multi-physical field coupling solution by a physical coupling layer, performing feature difference transfer by a residual correction layer, obtaining a composite cloth penetration depth distribution map, and performing abnormal region detection and risk level quantification through an improved gradient direction clustering method according to the composite cloth penetration depth distribution map to obtain a composite cloth penetration depth prediction result. And obtaining a leakage risk heat map, and performing temperature difference comparison analysis on the leakage risk heat map. Through the iterative nearest point algorithm, the multi-scale feature fusion method and the hot melt penetration prediction model, the accuracy of the online monitoring scheme and the consistency of the product quality of composite cloth production are effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automation control, in particular to an online monitoring system and method for a composite cloth production line. BACKGROUND

[0002] With the development of science and technology, as a high-performance material, composite cloth has been widely used in aerospace, medical care, sports clothing and other fields. Its production process is complex, involving hot melt composite and coating penetration and other key links, and the product quality directly affects the safety and reliability of the terminal application. In order to ensure the stability of the production process and the consistency of the product performance, modern composite cloth production lines gradually use online monitoring mechanism. The traditional production monitoring method relies on manual inspection and simple sensor data monitoring. In recent years, the rapid development of automation and intelligent technology provides a new solution for the online monitoring of composite cloth production lines.

[0003] However, the existing technology still has some deficiencies. First, the traditional monitoring method often has difficulty in achieving efficient and accurate spatial registration and feature aggregation when dealing with complex and variable production line data, which limits the in-depth understanding of the quality characteristics of the composite cloth. Second, the existing hot melt penetration prediction model usually ignores the interaction between multiple physical fields, resulting in inaccurate penetration depth distribution prediction and inability to achieve accurate monitoring of the quality of the composite cloth product. SUMMARY

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

[0005] Therefore, the present application provides an online monitoring method for a composite cloth production line to solve the problems of low spatial registration accuracy of multi-source data and insufficient penetration depth prediction accuracy in existing monitoring methods.

[0006] To solve the above technical problems, the present application provides the following technical solutions:

[0007] In a first aspect, the present application provides an online monitoring method for a composite cloth production line, which comprises: using an iterative closest point algorithm to perform spatial registration on a composite cloth production line data set, obtaining spatial registration data, and performing multi-source feature aggregation on the spatial registration data to output a composite cloth quality feature matrix;

[0008] Inputting the composite cloth quality feature matrix into a hot melt penetration prediction model, performing multi-physical field coupling solution by a physical coupling layer, and performing feature difference transmission by a residual error correction layer to obtain a composite cloth penetration depth distribution map;

[0009] According to the composite cloth penetration depth distribution map, performing abnormal area detection and risk level quantification by an improved gradient direction clustering method to obtain a leakage risk heat map, and performing temperature difference comparison analysis on the leakage risk heat map to form an online monitoring scheme.

[0010] The online monitoring scheme is converted into comprehensive control instructions to drive the PLC control unit to perform real-time monitoring on the composite cloth production line in terms of permeation risk and abnormal temperature distribution.

[0011] As a preferred scheme of the online monitoring method of the composite cloth production line, the composite cloth production line data set comprises production line equipment operation parameters, process control parameters and composite cloth product quality data.

[0012] As a preferred scheme of the online monitoring method of the composite cloth production line, the output composite cloth quality feature matrix specifically comprises the following operation steps,

[0013] The multi-scale downsampling and nearest neighbor search are performed on the composite cloth production line data set by using the iterative nearest point algorithm, the matching point set is obtained, the matching point set is time-space registered, and the space registration data is output;

[0014] The multi-source feature aggregation is performed on the space registration data by using the multi-scale feature fusion method, the multi-dimensional joint tensor is generated, the multi-dimensional joint tensor is subjected to slice reorganization, and the composite cloth quality feature matrix is output.

[0015] As a preferred scheme of the online monitoring method of the composite cloth production line, the composite cloth permeation depth distribution map is obtained, and the following operation steps are specifically included,

[0016] The physical coupling layer and the residual correction layer are built, and the gradient optimization and parameterized stacking are performed by using the residual connection, so as to construct the hot melt permeation prediction model;

[0017] The composite cloth quality feature matrix is input into the hot melt permeation prediction model, the physical coupling layer performs multi-physical field coupling solution by using the heat conduction-convection differential equation, and the permeation depth distribution is formed;

[0018] The residual correction layer performs feature difference transmission by using the dilated convolution, and obtains the permeation depth correction amount;

[0019] The permeation depth distribution and the permeation depth correction amount are weighted and spliced in the feature channel, and the permeation depth prediction value is generated;

[0020] The permeation depth prediction value is subjected to space domain projection by using the bilinear interpolation, and the composite cloth permeation depth distribution map is obtained.

[0021] As a preferred scheme of the online monitoring method of the composite cloth production line, the leakage risk heat map is obtained, and the following operation steps are specifically included,

[0022] The composite cloth permeation depth distribution map is divided into local area blocks to form a grid analysis unit;

[0023] Anomaly detection and clustering purification are performed on the grid analysis unit by using the improved gradient direction clustering method, a candidate anomaly point set is obtained, and spatial neighborhood aggregation is performed on the candidate anomaly point set to generate an anomaly region set;

[0024] The hierarchical analysis method is applied to quantize the risk level of the anomaly region set, and a risk classification vector is output; the risk classification vector is subjected to kernel density estimation fitting to obtain a leakage risk heat map.

[0025] As a preferred scheme of the online monitoring method of the composite cloth production line, the online monitoring scheme comprises the following operation steps,

[0026] The Gaussian interpolation is applied to perform discrete sampling on the leakage risk heat map to extract temperature distribution parameters; and the temperature difference gradient analysis method is used to perform temperature difference comparison analysis on the temperature distribution parameters to obtain a temperature anomaly vector;

[0027] The time sequence analysis and weighted fusion are performed on the temperature anomaly vector and the leakage risk heat map to form the online monitoring scheme.

[0028] As a preferred scheme of the online monitoring method of the composite cloth production line, the online monitoring scheme comprises the following operation steps,

[0029] The online monitoring scheme is packaged into an instruction by using the OPC protocol to form a comprehensive control instruction;

[0030] According to the comprehensive control instruction, the scanning cycle analysis and register writing are performed by the PLC control unit to output a high-low level signal;

[0031] Based on the high-low level signal, the permeation risk of the composite cloth production line is monitored in real time, and the PID controller is used to dynamically adjust the abnormal temperature distribution.

[0032] In a second aspect, the present application provides an online monitoring system for a composite cloth production line, comprising a registration aggregation module, a coupling correction module, a scheme generation module, and a monitoring control module:

[0033] The registration aggregation module uses the iterative closest point algorithm to perform space-time registration on the composite cloth production line data set to obtain space registration data, and performs multi-source feature aggregation on the space registration data to output a composite cloth quality feature matrix;

[0034] The coupling correction module inputs the composite cloth quality feature matrix into a hot melt permeation prediction model, performs multi-physical field coupling solution by a physical coupling layer, and performs feature difference transmission by a residual correction layer to obtain a composite cloth permeation depth distribution map;

[0035] A scheme generation module detects an abnormal area and quantifies a risk level through an improved gradient direction clustering method according to the composite cloth permeation depth distribution map, obtains a leakage risk heat map, performs temperature difference comparison analysis on the leakage risk heat map, and forms an online monitoring scheme.

[0036] A monitoring control module converts the online monitoring scheme into comprehensive control instructions to drive a PLC control unit to perform real-time monitoring on the permeation risk and abnormal temperature distribution of the composite cloth production line.

[0037] In a third aspect, the present application provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and wherein the computer program, when executed by the processor, implements any step of the online monitoring method of the composite cloth production line according to the first aspect of the present application.

[0038] In a fourth aspect, the present application provides a computer readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any step of the online monitoring method of the composite cloth production line according to the first aspect of the present application.

[0039] The present application has the following beneficial effects: through the iterative nearest point algorithm and the multi-scale feature fusion method, the spatio-temporal registration accuracy of multi-source heterogeneous data in the composite cloth production line is improved, and efficient alignment and integration of multi-source data are realized. At the same time, a hot melt permeation prediction model is constructed, fully considering the interaction between multiple physical fields such as heat conduction and convection, enhancing the prediction ability of the permeation depth distribution, effectively improving the accuracy and real-time performance of the online monitoring scheme, and further improving the stability of the composite cloth production process and the consistency of the product quality. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0041] Fig. 1 The flowchart of the online monitoring method of the composite cloth production line.

[0042] Fig. 2 The schematic diagram of the online monitoring system of the composite cloth production line.

[0043] Fig. 3 The flowchart of the permeation depth distribution map generation process of the composite cloth.

[0044] Fig. 4 The flowchart of the leakage risk heat map generation and monitoring scheme formation. DETAILED DESCRIPTION

[0045] In order to make the above objectives, features and advantages of the present application more clear and easily understood, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0046] In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present application. The present application, however, can be practiced in a variety of ways other than those specifically described herein, and the present application should not be construed as being limited to the embodiments described herein as these embodiments are presented solely for the purposes of enabling a complete and enabling disclosure of the present application.

[0047] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. The "in one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an embodiment that is independent of or selected from other embodiments.

[0048] Reference Figs. 1-4 For one embodiment of the present application, the embodiment provides an online monitoring method of a composite cloth production line, comprising the following steps:

[0049] S1, using the iterative closest point algorithm to perform space-time registration on the composite cloth production line data set, obtaining space registration data, and performing multi-source feature aggregation on the space registration data, and outputting a composite cloth quality feature matrix.

[0050] The specific operation steps are as follows,

[0051] S1.1, collect the composite cloth production line data set, which includes production line equipment operating parameters, process control parameters and composite cloth product quality data.

[0052] The production line equipment operating parameters include motor speed, hot melt roller temperature and air pressure data; the encoder is installed at the output shaft end of the driving motor of the production line to collect the motor speed data; the thermocouple is arranged on the surface of the hot melt roller and the inner wall of the heating cavity to obtain the hot melt roller temperature data; the pressure sensor is deployed in the air pressure pipeline to collect the air pressure data;

[0053] The process control parameters include coating thickness, hot melting time and production line speed data; the laser thickness gauge is installed at the outlet of the coating machine to collect the coating thickness data; the timer is arranged inside the hot melting device to collect the hot melting time data; the speed sensor is deployed at the transmission wheel position of the production line to collect the production line running speed data;

[0054] The composite cloth product quality data includes permeation parameters, tensile strength, and surface uniformity image data; an X-ray perspective instrument is installed at a production line quality detection point to collect permeation parameter data; a tensile testing machine is arranged in a finished product inspection area to obtain tensile strength data; and an optical microscope is used to synchronously collect surface uniformity image data;

[0055] The composite cloth production line data set provides multi-dimensional and high-time-efficiency online monitoring for intelligent industrial control, and provides key data support for dynamic optimization and quality closed-loop regulation of the production process.

[0056] S1.2, preprocessing the collected composite cloth production line data set, in the specific operation, for the production line equipment operation parameters, the production line equipment operation parameters are divided into multiple sample time sequence segments by using the sliding window statistical method, the abnormal values are detected and removed, and the Daubechies wavelet basis is used for multi-layer decomposition to suppress high-frequency noise and ensure data smoothness; for process control parameters, use kriging spatial interpolation to perform outlier detection and dynamic removal to ensure numerical validity, and simultaneously use median filtering to perform sliding smoothing of process control parameters in a fixed window (such as 7 sampling points) to eliminate transient interference; for composite cloth product quality data, fast Fourier transform is used for frequency domain feature extraction to separate vibration noise, and noise suppression is performed through Wiener filtering to ensure data purity;

[0057] The production line equipment operation parameters, process control parameters and composite cloth product quality data after preprocessing are time-aligned by dynamic time warping, the time base is unified, and linear interpolation is applied for resampling to unify the sampling frequency of each sensor; next, Z-score standardization is used for normalization to eliminate dimension differences, and finally principal component analysis is performed for dimension compression to output the preprocessed composite cloth production line data set.

[0058] S1.3, using the iterative closest point algorithm to perform spatiotemporal registration on the composite cloth production line data set to obtain spatial registration data, in the specific operation, the iterative closest point algorithm is used to perform point cloud feature extraction on the composite cloth production line data set to obtain original point cloud data; the original point cloud data is deconstructed by curvature variation, and local surface fitting is performed by least squares method to obtain a curvature distribution map; the curvature distribution map is segmented into sub-blocks by using a sliding window to obtain local area blocks, and the local area blocks are multi-scale down-sampled by voxel grid filtering to reduce the data amount while preserving the geometric features, and the down-sampled point cloud is output; then, the nearest neighbor search is performed on the down-sampled point cloud, and the outlier points are segmented by using the spectral clustering algorithm to obtain matching point pairs, and the matching point pairs are synchronously aggregated by weighting to obtain a matching point set;

[0059] The matching point set is subjected to nonlinear feature extraction by kernel principal component analysis (KPCA) to form a high-dimensional feature manifold, spatial-temporal registration is performed on the high-dimensional feature manifold, and spatial transformation projection is performed by the cut space mapping method to output a preliminary registration result; the gradient descent method is used to iteratively optimize and exponentially smooth the preliminary registration result to output spatial registration data.

[0060] S1.4, multi-source feature aggregation is performed on the spatial registration data to output a composite fabric quality feature matrix, in specific operations, the spatial registration data is subjected to smoothing processing by a Gaussian filter to obtain denoised registration data; multi-scale feature fusion is applied to the denoised registration data to perform pyramid decomposition and feature extraction to obtain multi-scale features; texture analysis and feature description are performed on the multi-scale features by a local binary pattern (LBP) to obtain microscopic texture features, and a histogram of oriented gradients (HOG) is simultaneously applied to gradient statistics of the multi-scale features to capture macroscopic structure features; multi-source feature aggregation is performed on the microscopic texture features and the macroscopic structure features to form a joint feature space, and tensor splicing is performed on the joint feature space to generate a multi-dimensional joint tensor;

[0061] The multi-dimensional joint tensor is subjected to region division and coordinate positioning by a spatial grid mapping method to obtain sub-region spatial positions; then the multi-dimensional joint tensor is sliced according to the 16x16 regular grid based on the sub-region spatial positions to obtain local feature blocks; the local feature blocks are projected to a two-dimensional space by t-SNE dimension reduction, and K-means is used for similarity clustering in the two-dimensional space to obtain feature clustering results; feature compression is performed on the feature clustering results by maximum pooling to generate region representative vectors; all region representative vectors are reorganized into a three-dimensional tensor in a spatial topological order, and the channel number of the three-dimensional tensor is compressed by principal component analysis to finally generate a composite fabric quality feature matrix;

[0062] It should be noted that the spatial topological order refers to a scanning order from left to right and from top to bottom in the conveying direction of the production line, which is defined based on the actual physical layout of the composite fabric production process.

[0063] S2, the composite fabric quality feature matrix is input into a hot melt infiltration prediction model, the physical coupling layer is subjected to multi-physical field coupling solution, and the residual error correction layer is subjected to feature difference transmission to obtain a composite fabric infiltration depth distribution map.

[0064] The specific operation steps are as follows,

[0065] S2.1, construct a hot melt infiltration prediction model and train, in the specific operation, in the PyTorch framework, utilize nn.Module Parameter calls U-Net network (U-shaped network) architecture, and embed heat conduction-convection differential equation for U-Net network architecture, simultaneously, set the input channel number to 3, the convolution kernel size to 7x7, and the down-sampling layer number to 4 levels;After U-Net network architecture, gradient feature extraction is carried out by Sobel operator to enhance the physical field coupling effect, and nonlinear activation is carried out by LeakyReLU function, to complete the construction of the physical coupling layer;Use nn.Sequential function to call ResNet network (deep residual network), and use dilated convolution to perform multi-scale feature fusion, while setting the input channel number to 64, the basic dilatation rate to [1,2,4], and the residual block stacking number to 3 times;After ResNet network, the attention mechanism is connected to weight the features to improve the accuracy of the penetration depth prediction, and the feature scaling is carried out by batch normalization to complete the construction of the residual correction layer;

[0066] It should be noted that Sobel operator is directly called by cv2.Sobel function of OpenCV, and the value range is [-255, 255];

[0067] The residual connection is used to splice the cross-layer features of the physical field coupling layer and the residual correction layer to obtain the fused feature tensor, and the gradient optimization is carried out on the fused feature tensor to generate the optimized feature representation;The depth separable convolution is used to transform the dimension of the optimized feature representation to obtain the cross-channel spatial feature;Xavier initialization is applied to the cross-channel spatial feature to allocate weights and form the initialized weights;According to the initialized weights, the physical coupling layer and the residual correction layer are parameterized and stacked through the gated recurrent unit, and the dynamic learning rate is adjusted through the recurrent learning rate scheduler to complete the construction of the hot melt infiltration prediction model;

[0068] Next, the hot melt infiltration prediction model is trained, and further, the composite cloth quality feature matrix is divided into sample set, training set and validation set;On the sample set, random rotation is carried out by random affine transformation, and data normalization is carried out by Z-score standardization to form preprocessed samples;On the training set, the preprocessed samples are propagated backward by using Adam optimizer, and the gradient clipping is applied simultaneously to limit the amplitude, to obtain the updated hot melt infiltration prediction model parameters;On the validation set, the updated hot melt infiltration prediction model parameters are quantified by using smooth L1 loss function Error, get the validation error;When the validation error exceeds the convergence threshold for 5 consecutive rounds, the training is terminated, and the trained hot melt infiltration prediction model is output simultaneously;

[0069] It should be noted that the convergence threshold is defined based on the rate of decrease of the smoothed L1 loss function, and the value range is [0.001, 0.01].

[0070] S2.2, generate the penetration depth prediction value by using the hot melt infiltration prediction model, in the specific operation, input the composite cloth quality feature matrix into the hot melt infiltration prediction model through the input interface, the physical coupling layer extracts the local features of the composite cloth quality feature matrix through 3*3 convolution to obtain the temperature-pressure field characteristics; simultaneously, the temperature-pressure field characteristics are mapped to the hidden space through the gate recurrent unit to extract the deep features and obtain the 128-dimensional hidden variables; the 128-dimensional hidden variables are spatially discretized by the voxelization algorithm to form voxel grid units, in each voxel grid unit, the local gradient calculation is performed by using the automatic differentiation mechanism to obtain the heat conduction matrix and the convection matrix; the heat conduction matrix is discretized by the second-order central difference format through the heat conduction-convection differential equation, and the convection matrix is discretized by the first-order upwind format at the same time to obtain the discrete algebraic equation group, and the multi-physical field coupling solution is performed on the discrete algebraic equation group to generate the steady-state field solution value, and the specific mathematical formula is as follows,

[0071]

[0072] Wherein, h represents the steady-state field solution value, T represents the temperature field distribution value, P represents the pressure field distribution value, L represents the heat conduction matrix, C represents the convection matrix, and b represents the source term vector.

[0073] It should be noted that the temperature field distribution value is obtained by sparse Cholesky decomposition of the heat conduction matrix; the pressure field distribution value is obtained by Krylov subspace solution of the convection matrix; the source term vector refers to the joint discrete term of the heat conduction matrix and the convection matrix, which is obtained by regional integration of the heat conduction-convection differential equation.

[0074] The steady-state field solution value is subjected to gradient amplitude normalization by using the Sobel operator, which avoids the oscillation of the steady-state field solution value while enhancing the physical field coupling effect, and outputs the smooth field gradient distribution, and the smooth field gradient distribution is subjected to multi-scale aggregation to obtain enhanced physical field characteristics; the enhanced physical field characteristics are subjected to nonlinear activation and feature scaling by using the LeakyReLU function to obtain normalized physical characteristics; the normalized physical characteristics are subjected to spatial upsampling and linear fitting by using deconvolution to obtain the penetration depth distribution.

[0075] The residual correction layer performs feature difference transmission through a three-layer deep residual network. The first layer of the deep residual network uses a 3x3 convolution to perform multi-scale feature extraction on the composite fabric quality feature matrix to obtain 64-dimensional basic features. The second layer uses a 1x1 convolution to compress the channel and align the dimensions of the penetration depth distribution to generate 32-dimensional intermediate features. The third layer uses three groups of cascaded dilated convolutions to perform feature difference transmission and multi-scale fusion on the 64-dimensional basic features and the 32-dimensional intermediate features to form cross-scale associated features. The dilated convolution is used to perform cavity sampling on the cross-scale associated features to gradually expand the receptive field and capture feature differences at different scales, outputting 256-dimensional fusion features.

[0076] Batch normalization (BN) is applied to normalize the 256-dimensional fusion features, and a LeakyReLU activation function is simultaneously applied for nonlinear transformation to enhance the feature expression capability and obtain enhanced high-order features. Then, the composite fabric quality feature matrix and the enhanced high-order features are added element by element through a jump connection to avoid gradient disappearance, and residual enhanced features are output. The residual enhanced features are compressed to a single channel by a 1x1 convolution, and the numerical range is constrained by a Sigmoid function to obtain a penetration depth correction amount. Finally, the penetration depth correction amount and the penetration depth distribution are superimposed by element-wise addition, and gradient amplitude limitation is simultaneously performed by gradient clipping to obtain a penetration depth prediction value.

[0077] S2.3, the penetration depth prediction value is projected in the spatial domain by bilinear interpolation to obtain a composite fabric penetration depth distribution map. In specific operations, the penetration depth prediction value is upsampled by bilinear interpolation to obtain a target pixel point, and a nearest neighbor search is performed on the target pixel point to obtain four nearest neighbor points. The four nearest neighbor points are weighted and averaged to obtain a preliminary interpolation result. The Euclidean distance is used to measure the spatial distance of the preliminary interpolation result to obtain the relative distance between adjacent points. According to the relative distance between adjacent points, the target pixel point is assigned a weight, and the weighted fusion is simultaneously performed by weight multiplication to output the interpolated depth value. The interpolated depth value is edge filled by zero value expansion to complete the boundary missing data and form continuous depth values;

[0078] The continuous depth values are converted in the coordinate system to obtain projection parameters. The projection parameters are projected in the spatial domain by perspective transformation to map the target pixel point to the physical coordinate system of the actual production line of the composite fabric to obtain the depth distribution under the physical coordinates. The depth distribution under the physical coordinates is edge smoothed and optimized by bicubic interpolation to obtain an optimized depth distribution. Finally, the optimized depth distribution is denoised by Gaussian filtering to remove high-frequency noise introduced by interpolation and coordinate transformation, and a composite fabric penetration depth distribution map is output.

[0079] S3, according to the composite cloth penetration depth distribution map, the abnormal area detection and risk level quantification are carried out through the improved gradient direction clustering method, the leakage risk heat map is obtained, the temperature difference comparison analysis is carried out on the leakage risk heat map, and the online monitoring scheme is formed.

[0080] The specific operation steps are as follows,

[0081] S3.1, the improved gradient direction clustering method is used for abnormal detection and clustering purification of the composite cloth penetration depth distribution map, and an abnormal area set is generated. In the specific operation, a sliding window is used to scan the composite cloth penetration depth distribution map according to a fixed size (such as 16*16 pixels), a depth data block is obtained, and local area block division is performed on the depth data block to obtain a grid analysis unit.

[0082] The Scharr operator is used to calculate the gradient amplitude of each grid analysis unit to obtain the depth gradient amplitude. The depth gradient amplitude is direction histogram statistic to obtain the gradient direction distribution. The gradient direction distribution is divided into 8 intervals according to equal interval, and the sum of the depth gradient amplitude in each interval is calculated to form an 8-dimensional feature vector. In the abnormal detection stage, the improved gradient direction clustering algorithm is applied to the 8-dimensional feature vector to perform similarity clustering to obtain an initial clustering result, and the initial clustering result is probability density weighted to generate an initialization clustering center. The cosine similarity of the 8-dimensional feature vector to the initialization clustering center is quantified by using the vector angle formula, and the specific mathematical formula is as follows,

[0083]

[0084] Wherein, s represents the cosine similarity, v represents the 8-dimensional feature vector, and c represents the clustering center vector.

[0085] It should be noted that the Scharr operator is defined based on the gray change rate of the composite cloth penetration depth distribution map, and the value range is [0, 255]. The abnormal threshold is defined based on the distribution quantile of the cosine similarity, and the value range is [0.5, 0.8]. The clustering center vector is obtained by probability density weighting of the initialization clustering center.

[0086] The grid analysis unit with a cosine similarity lower than the abnormal threshold is marked as an abnormal point, and the candidate abnormal point set is obtained by weighted fusion.

[0087] In the clustering purification stage, the K nearest neighbor search is used to perform neighborhood expansion on the candidate abnormal point set to obtain a connected region, and the spatial neighborhood aggregation is performed on the connected region to obtain a preliminary abnormal area. At the same time, the Gaussian filter is applied to the preliminary abnormal area to remove noise and generate an optimized abnormal area. Finally, the bilateral filter is used to smooth the edges of the optimized abnormal area to output the abnormal area set.

[0088] It should be explained that the improved gradient direction clustering algorithm makes the following improvements on the existing gradient direction clustering algorithm. On the one hand, probability density weighting is applied for cluster center initialization, realizing sensitive capture of abnormal patterns. On the other hand, dynamic neighborhood expansion is performed through K-neighbor search, improving the spatial aggregation accuracy of the candidate abnormal point set.

[0089] S3.2, applying the analytic hierarchy process to quantize the risk level of the abnormal region set, outputting a risk classification vector, performing kernel density estimation fitting on the risk classification vector, obtaining a leakage risk heat map, in the specific operation, the feature extraction and weight distribution of the abnormal region set are performed through the analytic hierarchy process to obtain weighted abnormal features; the weighted abnormal features are dimensionally compressed by principal component analysis to obtain a low-dimensional abnormal matrix; the low-dimensional abnormal matrix is discretized by spatial grid to establish a hierarchical evaluation network, and the hierarchical evaluation network is grid interpolated and spatial continuity enhanced by bilinear interpolation to generate a smooth evaluation surface;

[0090] The smooth evaluation surface is quantized by a three-level threshold, for example, the range of the first-level threshold is set to [0, 0.3), the range of the second-level threshold is set to [0.3, 0.7), and the range of the third-level threshold is set to [0.7, 1.0]; the smooth evaluation surface is fitted by probability density estimation to obtain a risk quantization value; when the risk quantization value is within the range of the first-level threshold, it is identified as a risk and needs to be monitored regularly; when the risk quantization value is within the range of the second-level threshold, it is identified as a medium risk and needs to be warned and trigger manual review; when the risk quantization value is within the range of the third-level threshold, it is identified as a high risk and needs to be immediately shut down for maintenance; the divided risk level is resampled in space by Kriging interpolation and smoothed by Gaussian filtering to output a risk classification vector;

[0091] The risk classification vector is convolved by a Gaussian kernel function to obtain a risk probability density field, the risk probability density field is kernel density estimated to obtain a risk probability distribution; the obtained risk probability distribution is nonlinearly regressed and fitted by a surface fitting algorithm to generate a risk probability distribution surface, the risk probability distribution surface is three-dimensionally rendered by an OpenGL engine to generate a leakage risk heat map.

[0092] S3.3, perform discrete sampling on the leakage risk heat map using Gaussian interpolation to extract temperature distribution parameters; use temperature difference gradient analysis to compare and analyze the temperature distribution parameters, and obtain a temperature anomaly vector; in the specific operation, sample points are generated on the leakage risk heat map at a grid spacing of 10x10 pixels, and discrete sampling is performed through Gaussian interpolation to obtain discrete temperature sampling values; 8-neighborhood search is performed on the discrete temperature sampling values to obtain a neighborhood temperature set; weighted average and smooth transition are performed on the discrete temperature sampling values and the neighborhood temperature set through a radial basis function (RBF) to form a continuous temperature distribution parameter; median filtering is performed on the formed continuous temperature distribution parameter to obtain a temperature distribution parameter;

[0093] Then, using the temperature difference gradient analysis method, the temperature distribution parameters are compared with the reference temperature distribution parameters point by point to obtain the temperature difference change; the spatial gradient of the temperature distribution parameters is extracted through the Prewitt operator to obtain the gradient amplitude and direction of the temperature distribution; the temperature difference change and the gradient amplitude and direction of the temperature distribution are spliced into a vector, and normalized using Min-Max standardization to generate a temperature anomaly vector containing the temperature difference size and direction;

[0094] It should be noted that the reference temperature distribution parameter refers to the historical average temperature distribution under normal operating conditions of the production line, which is defined based on the average temperature distribution of the same process parameters in the past 30 days; the Prewitt operator is directly called through the filter2D function of the OpenCV library.

[0095] S3.4, perform time series analysis and weighted fusion on the temperature anomaly vector and the leakage risk heat map to form an online monitoring scheme; in the specific operation, the temperature anomaly vector is time series sliced through a sliding time window to obtain a temperature difference change time series; moving average filtering is performed on the temperature difference change time series to form a smooth temperature trend; at the same time, the smooth temperature trend is frequency domain decomposed through fast Fourier transform to extract the main fluctuation frequency and output the temperature frequency domain feature; the deformation tracking of the leakage risk heat map is performed using the optical flow method to obtain the risk area motion vector, and the Kalman filter is used to perform time series analysis and noise suppression on the risk area motion vector to obtain an optimized risk trajectory; the dynamic time warping is used to match the similarity of the optimized risk trajectory to generate a real-time risk indicator; then, the dynamic weights are assigned to the temperature frequency domain feature and the real-time risk indicator, and the linear weighted fusion method is used for weighted summation to generate a comprehensive monitoring indicator; the comprehensive monitoring indicator is integrated to obtain an online monitoring scheme.

[0096] S4, convert the online monitoring scheme into a comprehensive control instruction to drive the PLC control unit to monitor the real-time leakage risk and abnormal temperature distribution of the composite cloth production line.

[0097] S4.1, encapsulate the online monitoring scheme by using the OPC protocol to form a comprehensive control instruction, in specific operation, extract the multi-source monitoring parameters in the online monitoring scheme, register the multi-source monitoring parameters through the OPC protocol to obtain readable and writable OPC tags, each OPC tag represents a different control variable (such as temperature set value, risk level, switch state); then execute data subscription on the OPC tags through the OPC Group object to obtain real-time updated data stream; then call the OPCWrite method to encapsulate the real-time updated data stream into instructions and write them into the PLC control unit register in batches to obtain register state feedback; simultaneously use the ST structured language to perform signal conversion on the register state feedback to output a comprehensive control instruction.

[0098] S4.2, according to the comprehensive control instruction, the PLC control unit performs scan cycle analysis and register writing to output a high-low level signal, in specific operation, use the PLC control unit to perform scan cycle analysis and instruction decoding on the comprehensive control instruction to obtain an output register state; use the backplane bus to transmit the output register state to generate a physical output signal; perform register writing on the physical output signal through the relay driving circuit to output a 24V high level or 0V low level, and simultaneously apply a magnetic coupling isolator for electrical isolation and filtering to ensure signal stability and anti-interference ability, output a high-low level signal.

[0099] S4.3, based on the high-low level signal, real-time monitor the infiltration risk of the composite cloth production line, and simultaneously use a PID controller to dynamically adjust the abnormal temperature distribution, in specific operation, use Canny edge detection to perform region segmentation on the leakage risk heat map to obtain a risk area contour, and apply local binary pattern (LBP) to perform feature extraction and neighborhood aggregation on the risk area contour to form the infiltration risk of the composite cloth production line; according to the high-low level signal, prioritize the infiltration risk of the composite cloth production line to obtain a temperature risk sequence; perform real-time monitoring on the temperature risk sequence through the OPC UA data bus to realize risk closed-loop control;

[0100] perform window analysis on the temperature anomaly vector to obtain a temperature change trend; perform frequency domain decomposition on the temperature change trend through fast Fourier transform to form an abnormal temperature distribution; according to the high-low level signal, perform region feature labeling on the abnormal temperature distribution to obtain a temperature anomaly core area; perform parameter extraction and optimization on the temperature anomaly core area through the PID controller to obtain temperature control parameters; according to the temperature control parameters, dynamically adjust and real-time feedback the abnormal temperature distribution to ensure the temperature stability of the composite cloth production line.

[0101] The embodiment also provides an online monitoring system of the composite cloth production line, comprising: a registration aggregation module, a coupling correction module, a scheme generation module, and a monitoring control module.

[0102] The registration aggregation module performs space-time registration on the composite cloth production line dataset by using an iterative closest point algorithm, acquires spatial registration data, performs multi-source feature aggregation on the spatial registration data, and outputs a composite cloth quality feature matrix.

[0103] The coupling correction module inputs the composite cloth quality feature matrix into a hot melt infiltration prediction model, performs multi-physical field coupling solution by using a physical coupling layer, and performs feature difference transmission by using a residual error correction layer, to acquire a composite cloth infiltration depth distribution map.

[0104] The scheme generation module performs abnormal area detection and risk level quantification on the composite cloth infiltration depth distribution map by using an improved gradient direction clustering method, acquires a leakage risk heat map, performs temperature difference comparison analysis on the leakage risk heat map, and forms an online monitoring scheme.

[0105] The monitoring control module converts the online monitoring scheme into comprehensive control instructions, and drives a PLC control unit to perform real-time monitoring on the infiltration risk and abnormal temperature distribution of the composite cloth production line.

[0106] The embodiment also provides a computer device suitable for the online monitoring method of the composite cloth production line, comprising: 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 realize the online monitoring method of the composite cloth production line proposed in the above embodiment.

[0107] The computer device can be a terminal, and the computer device comprises a processor, a memory, a communication interface, a display screen and an input device connected through a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises 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 operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to perform wired or wireless communication with external terminals. The wireless communication can be achieved by WIFI, operator network, NFC (near field communication) or other technologies. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of the computer device can be a touch layer overlaid on the display screen, or a key, trackball or touchpad arranged on the shell of the computer device. In addition, the input device can be an external keyboard, touchpad or mouse, etc.

[0108] The embodiment also provides a storage medium on which a computer program is stored, the computer program being executed by a processor to implement the online monitoring method for realizing the composite cloth production line as proposed in the above embodiment; the storage medium can be realized by any type of volatile or non-volatile storage device or a combination thereof, such as a static random access memory (SRAM), an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk, or an optical disk.

[0109] To sum up, the present application improves the spatiotemporal registration accuracy of multi-source heterogeneous data in the composite cloth production line by the iterative closest point algorithm and the multi-scale feature fusion method, and realizes efficient alignment and integration of multi-source data. At the same time, a hot melt penetration prediction model is constructed, fully considering the interaction between multiple physical fields such as heat conduction and convection, enhancing the prediction ability of the penetration depth distribution, effectively improving the accuracy and real-time performance of the online monitoring scheme, and further improving the stability of the composite cloth production process and the consistency of the product quality.

[0110] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. An online monitoring method for a composite fabric production line, characterized by: include, The iterative closest point algorithm is used to perform spatiotemporal registration on the composite fabric production line dataset to obtain spatial registration data. Multi-source feature aggregation is then performed on the spatial registration data to output the composite fabric quality feature matrix. The composite fabric quality feature matrix is ​​input into the hot melt penetration prediction model. The physical coupling layer performs multi-physics field coupling solution, and the residual correction layer performs feature difference transfer to obtain the composite fabric penetration depth distribution map. Based on the penetration depth distribution map of the composite fabric, an improved gradient directional clustering method is used to detect abnormal areas and quantify risk levels. This generates a leakage risk heat map, which is then subjected to temperature difference comparative analysis to form an online monitoring plan. The online monitoring scheme is converted into comprehensive control instructions to drive the PLC control unit to conduct real-time monitoring of the composite fabric production line for penetration risk and abnormal temperature distribution.

2. The online monitoring method for a composite fabric production line according to claim 1, wherein: The composite fabric production line data set includes production line equipment operating parameters, process control parameters, and composite fabric product quality data.

3. The online monitoring method for a composite fabric production line according to claim 1, wherein: The outputting of the composite fabric quality characteristic matrix specifically includes the following steps: The iterative closest point algorithm is used to perform multi-scale downsampling and nearest neighbor search on the composite fabric production line dataset to obtain a set of matching points. The matching point set is then spatially and temporally aligned to output spatially aligned data. The multi-scale feature fusion method is applied to aggregate multi-source features of spatial registration data to generate a multi-dimensional joint tensor. The multi-dimensional joint tensor is sliced ​​and reorganized to output the composite fabric quality feature matrix.

4. The online monitoring method for a composite fabric production line according to claim 2, wherein: The method of obtaining the composite fabric penetration depth distribution map specifically includes the following steps: Build a physical coupling layer and residual correction layer, and use residual connections to perform gradient optimization and parameter stacking to construct a hot melt penetration prediction model; The composite fabric quality characteristic matrix is ​​input into the hot melt penetration prediction model, and the physical coupling layer solves the multi-physics field coupling through the heat conduction-convection differential equation to form the penetration depth distribution; The residual correction layer uses dilated convolution to perform feature difference transfer to obtain the penetration depth correction; The penetration depth distribution and the penetration depth correction amount are weightedly spliced ​​in the characteristic channel to generate the penetration depth prediction value; Bilinear interpolation is used to perform spatial domain projection on the penetration depth prediction value to obtain the penetration depth distribution map of the composite fabric.

5. The online monitoring method for a composite fabric production line according to claim 1, wherein: The method of obtaining the leakage risk heat map specifically includes the following steps: Divide the composite fabric penetration depth distribution map into local area blocks to form grid analysis units; The improved gradient direction clustering method is used to perform anomaly detection and cluster purification on the grid analysis unit to obtain a set of candidate anomaly points. The spatial neighborhood aggregation is then performed on the candidate anomaly point set to generate a set of anomaly regions. The analytic hierarchy process is applied to quantify the risk level of the abnormal area set, output the risk classification vector, and perform kernel density estimation fitting on the risk classification vector to obtain the leakage risk heat map.

6. The online monitoring method for a composite fabric production line according to claim 5, wherein: The online monitoring scheme is formed, specifically including the following steps: Gaussian interpolation is used to perform discretized sampling on the leakage risk heat map to extract temperature distribution parameters; Use the temperature difference gradient analysis method to perform temperature difference comparison analysis on the temperature distribution parameters and obtain the temperature anomaly vector; Time series analysis and weighted fusion are performed on temperature anomaly vectors and leakage risk heat maps to form an online monitoring solution.

7. The online monitoring method for a composite fabric production line according to claim 1, wherein: The driving PLC control unit performs real-time monitoring of the penetration risk and abnormal temperature distribution of the composite fabric production line, specifically including the following operating steps: Use the OPC protocol to encapsulate the online monitoring solution into instructions to form comprehensive control instructions; According to the comprehensive control instructions, the PLC control unit performs scan cycle analysis and register writing, and outputs high and low level signals; Based on high and low level signals, the penetration risk of the composite fabric production line is monitored in real time, and the PID controller is used to dynamically adjust the abnormal temperature distribution.

8. An online monitoring system for a composite fabric production line, based on the online monitoring method for a composite fabric production line according to any one of claims 1 to 7, characterized in that: Including registration aggregation module, coupling correction module, solution generation module, monitoring and control module: The registration and aggregation module uses the iterative closest point algorithm to perform spatiotemporal registration on the composite fabric production line dataset, obtains spatial registration data, performs multi-source feature aggregation on the spatial registration data, and outputs the composite fabric quality feature matrix; The coupling correction module inputs the composite fabric quality feature matrix into the hot melt penetration prediction model, the physical coupling layer performs multi-physics field coupling solution, and the residual correction layer performs feature difference transfer to obtain the composite fabric penetration depth distribution map; The solution generation module uses an improved gradient directional clustering method to detect abnormal areas and quantify risk levels based on the composite fabric penetration depth distribution map, obtain a leakage risk heat map, perform temperature difference comparison analysis on the leakage risk heat map, and form an online monitoring solution. The monitoring and control module converts the online monitoring scheme into comprehensive control instructions, driving the PLC control unit to conduct real-time monitoring of the composite fabric production line for penetration risks and abnormal temperature distribution.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the online monitoring method of the composite fabric production line according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the online monitoring method of the composite fabric production line according to any one of claims 1 to 7 are implemented.

Images