Wire and cable online quality detection method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of Internet of Things, and provides a wire and cable online quality detection method and device, electronic equipment and a storage medium. The method comprises the following steps: performing multi-scale wavelet packet decomposition and signal entropy fusion on an original wire and cable signal matrix to obtain a feature tensor, and performing anomaly recognition extraction on the feature tensor through a graph attention network model to obtain a potential defect feature vector; and performing multi-physics coupling simulation inversion on the potential defect feature vector to obtain a defect quantization parameter set, performing dynamic reasoning according to the defect quantization parameter set through a Bayesian network model to obtain an online quality grade decision, and packaging the online quality grade decision according to a block chain intelligent protocol to obtain a quality traceability record. According to the invention, through organic coupling of multiple levels, multiple models and multiple technical means, real-time performance, accuracy, traceability and safety of online quality detection of wires and cables are realized.

Description

Technical Field

[0001] The present application relates to the technical field of the Internet of Things, and in particular, to a method, device, electronic device and storage medium for on-line quality detection of wire and cable. Background Art

[0002] As the basic carrier of modern energy and information transmission, in key fields such as high-voltage power grids, rail transit, and intelligent manufacturing, once there are defects such as concealed cracks, impurity cavities, or insulation deterioration inside the wire and cable, it will not only lead to local hot spots and dielectric breakdown, but may also cause large-scale power outages, equipment damage, and safety accidents.

[0003] The existing wire and cable quality detection methods mainly include partial discharge detection, ultrasonic detection, infrared thermal imaging, and DC dielectric loss testing. Partial discharge detection judges insulation defects by measuring discharge pulse signals; ultrasonic detection relies on the probe attachment method to capture echo signals in solid pipelines; infrared thermal imaging locates hot spots by comparing temperature difference maps when the cable is energized; DC dielectric loss testing evaluates the insulation condition by scanning the changes in capacitance and loss factor under voltage. These methods are widely used in standardized factory maintenance and periodic inspections, and can give early warnings for obvious external defects or thermal anomalies. However, they generally rely on a single physical quantity or single-field signal, often only able to detect surface-level faults, and it is difficult to balance the detection frequency and diagnostic depth at the same time. Summary of the Invention

[0004] In view of this, the present application provides a method, device, electronic device and storage medium for on-line quality detection of wire and cable to solve the problem of fusion analysis and detection of multi-scale and multi-physical field signals of wire and cable.

[0005] The first aspect of the present application provides a method for on-line quality detection of wire and cable, and the method includes: Performing multi-scale wavelet packet decomposition and signal entropy fusion on the synchronously collected original wire and cable signal matrix to obtain a feature tensor; Performing anomaly recognition and extraction on the feature tensor through a preset graph attention network model to obtain a potential defect feature vector; Performing multi-physical field coupling simulation inversion on the potential defect feature vector to obtain a set of defect quantification parameters; Performing dynamic reasoning according to the set of defect quantification parameters through a preset Bayesian network model to obtain an on-line quality grade decision; Encapsulating the on-line quality grade decision according to a preset blockchain intelligent protocol to obtain a quality traceability record.

[0006] In an alternative embodiment, the multi-scale wavelet packet decomposition and signal entropy fusion of the synchronously acquired original wire and cable signal matrix to obtain the feature tensor includes: Perform Daubechies wavelet decomposition on the synchronously acquired original wire and cable signal matrix to obtain a set of subband coefficients; Perform soft threshold denoising on the top-level high-frequency subband coefficients in the set of subband coefficients according to a preset SURE threshold function to obtain a denoised and optimized set of subband coefficients; Calculate the information entropy weight of each subband by performing weight calculation on the denoised and optimized set of subband coefficients through a preset Shannon entropy formula; Perform information entropy fusion on the denoised and optimized set of subband coefficients according to the information entropy weight to obtain the feature tensor.

[0007] In an alternative embodiment, the anomaly recognition and extraction of the feature tensor through a preset graph attention network model to obtain the potential defect feature vector includes: Perform high-order singular value decomposition on the feature tensor to obtain a core tensor, and calculate the mutual information measure for each element in the core tensor to obtain the edge weights between the elements; Connect each element in the core tensor according to the edge weights and a preset threshold to construct a spatio-temporal graph; Perform anomaly detection on the spatio-temporal graph through a preset graph attention network model to obtain the potential defect feature vector.

[0008] In an alternative embodiment, the multi-physical field coupling simulation inversion of the potential defect feature vector to obtain a set of defect quantification parameters includes: Perform multi-physical field coupling on the potential defect feature vector according to a preset Maxwell electromagnetic equation set, heat conduction equation, and structural mechanics equation to obtain a multi-physical field finite element model; Perform minimization simulation on the multi-physical field finite element model according to a preset inversion objective function to obtain an inversion problem; Perform mapping processing on the inversion problem through a preset Ising model to obtain an Ising Hamiltonian, and perform D-Wave quantum annealing solution processing on the Ising Hamiltonian to obtain the set of defect quantification parameters.

[0009] In an alternative embodiment, the dynamic inference through a preset Bayesian network model according to the set of defect quantification parameters to obtain an online quality grade decision includes: Perform decision prediction through a preset Bayesian network model according to the set of defect quantification parameters to obtain the conditional probability table of each node; Perform posterior distribution calculation on the defect quantization parameter set and the conditional probability table through a preset Monte Carlo sampling likelihood function to obtain quality level conditions and corresponding posterior probabilities; Perform MAP decision screening on the quality level conditions according to the posterior probabilities to obtain the online quality level decision.

[0010] In an alternative embodiment, the quality traceability record includes a decision record and a content identification record. Encapsulating the online quality level decision according to a preset blockchain intelligent protocol to obtain the quality traceability record includes: Perform hashing processing and combination on the original wire and cable signal matrix, the feature tensor, the potential defect feature vector, and the defect quantization parameter set according to a preset SHA3-256 algorithm to obtain the Merkle root hash value; Encapsulate and store the online quality level decision and the corresponding Merkle root hash value according to a preset blockchain intelligent protocol and timestamp to obtain the decision record; Perform distributed encrypted storage on the original wire and cable signal matrix, the feature tensor, the potential defect feature vector, and the defect quantization parameter set to obtain encrypted data shards and corresponding CID indexes, and encapsulate and store the CID indexes according to the blockchain intelligent protocol to obtain the content identification record.

[0011] In an alternative embodiment, the method further includes: Construct a loss function based on the original wire and cable signal matrix, the feature tensor, the potential defect feature vector, the defect quantization parameter set, and the online quality level decision; Perform automatic differentiation calculation on the loss function to obtain model parameter gradient values; Perform secure multi-party calculation and aggregation decryption on the model parameter gradient values through a preset federated learning model to obtain a global model parameter set; Perform point cloud registration on the surface point cloud data in the original wire and cable signal matrix and a preset virtual model through a preset ICP algorithm to obtain model mapping data; Perform AR terminal holographic rendering on the defect quantization parameter set according to the model mapping data to generate a human-machine collaborative decision-making interface.

[0012] The second aspect of the present application provides a wire and cable online quality detection device, and the device includes: A noise reduction and fusion module for performing multi-scale wavelet packet decomposition and signal entropy fusion on the synchronously collected original wire and cable signal matrix to obtain a feature tensor; A defect recognition module, configured to perform anomaly recognition and extraction on the feature tensor through a preset graph attention network model to obtain a potential defect feature vector; A coupled simulation module, configured to perform multi-physical-field coupled simulation inversion on the potential defect feature vector to obtain a set of defect quantification parameters; A dynamic reasoning module, configured to perform dynamic reasoning based on the set of defect quantification parameters through a preset Bayesian network model to obtain an online quality grade decision; A decision encapsulation module, configured to encapsulate the online quality grade decision according to a preset blockchain intelligent protocol to obtain a quality traceability record.

[0013] A third aspect of this application provides an electronic device, which includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the steps of the above-mentioned on-line quality detection method for wire and cable are implemented.

[0014] A fourth aspect of this application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned on-line quality detection method for wire and cable are implemented.

[0015] In summary, this application at least includes the following beneficial technical effects: 1. A multi-physical-field coupled finite element model based on Maxwell's equations, heat conduction equations, and structural mechanics equations can accurately reproduce defect behavior under multi-field coupling of electromagnetism, heat, and mechanics.

[0016] 2. Combining the Ising model with D-Wave quantum annealing for solution, mapping the complex inversion optimization problem to the solution of the quantum Hamiltonian, significantly improving the solution speed and global optimality, avoiding falling into local optimality, and thus obtaining highly reliable defect quantification parameters.

[0017] 3. Using multi-scale wavelet packet decomposition combined with SURE soft threshold denoising and Shannon entropy weighted fusion can fully retain tiny defect signals on different frequency bands while suppressing noise interference, thereby generating a more reliable feature tensor with a higher signal-to-noise ratio. Description of the Drawings

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or in the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0019] Figure 1 It is a flowchart of an on-line quality detection method for wire and cable provided by an embodiment of the present application; Figure 2 It is a functional module diagram of an on-line quality detection device for wire and cable provided by an embodiment of the present application; Figure 3 It is a schematic structural diagram of an electronic device provided by an embodiment of the present application. Specific embodiments

[0020] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of the present application.

[0021] As Figure 1 shown, it is a flowchart of an on-line quality detection method for wire and cable provided by an embodiment of the present application. The on-line quality detection method for wire and cable provided by an embodiment of the present application includes the following steps.

[0022] Step S1: Perform multi-scale wavelet packet decomposition and signal entropy fusion on the synchronously collected original wire and cable signal matrix to obtain a feature tensor.

[0023] It should be understood that the upper controller in the embodiment of the present application receives real-time data through a multi-type sensor network deployed on the cable production line. The sensors include but are not limited to an infrared thermal imager installed at key positions on the production line to capture the surface temperature distribution map of the wire and cable; a high-frequency eddy current sensor closely attached to the cable surface to measure the change in conductor resistivity; a distributed fiber optic strain sensor embedded along the cable length to monitor the strain value in real time, etc. All sensors in the embodiment of the present application align timestamps through the IEEE 1588 Precision Time Protocol to ensure the consistency of the data time axis. For example, when the infrared thermal imager takes a picture at time point t k is taken, other sensors record data within the same microsecond. And map the spatial data of different sensors to a unified reference system through coordinate transformation. Finally, the data of each sensor is integrated into a multi-dimensional original wire and cable signal matrix according to the time-space dimension, where each element contains 5 types of parameters (i.e., parameter, resistivity, strain, thickness deviation, acoustic emission energy).

[0024] After receiving the original wire and cable signal matrix collected synchronously from the wire and cable production line, first perform multi-scale decomposition on each time-series signal through Daubechies wavelet decomposition to generate sub-band coefficient sets at each level. Each sub-band represents the detailed or approximate components of the signal within a specific frequency band. For example, after performing 8-level decomposition on 1000 time sampling points collected from a 10-meter-long cable within 1 second, 256 sets of sub-band coefficients {C j,k} can be obtained, where the index j ∈ {1, ……, 8} represents the decomposition level, and k represents different frequency band nodes at the same level. Immediately afterwards, apply soft threshold denoising processing to the high-frequency sub-band coefficients of the 8th layer with the highest frequency. Specifically, first, for each sub-band coefficient set C j,k (the kth frequency band of the jth layer), the noise standard deviation is usually estimated and calculated using the median method. At the same time, the probability distribution p(x) of the band coefficients is statistically analyzed by amplitude segmentation, then the probability corresponding to each amplitude is multiplied by its own logarithm and summed, and finally the negative sign is taken to obtain the information entropy H(C j,k ). The larger the information entropy H(C j,k ), the richer and more irregular the information contained in this frequency band, and the more likely it is to contain real defect features. Finally, the entropy value is mapped to a weight factor through a monotonically decreasing function, and then multiplied by the noise standard deviation to obtain the denoising threshold λ j,k for each sub-band. For all coefficients in this sub-band, those with absolute values lower than the denoising threshold λ j,k are set to zero (i.e., noise is suppressed); the remaining coefficients are either retained at their original values or smoothly scaled. At the same time, through the SURE threshold function, when a certain sub-band has a high information entropy H(C j,k ) (i.e., rich in effective defect information), the denoising threshold λ j,k automatically decreases, thus retaining more details; conversely, when the entropy value is low, the threshold increases to suppress noise. For example, if the information entropy H(C 8,16 ) of the 16th sub-band of the 8th layer is significantly higher than the average level, the corresponding threshold λ 8,16 will drop to 60% of the original value, allowing the high-frequency oscillations generated by the crack to be retained.

[0025] After soft threshold denoising processing, all subband coefficient sets are reconstructed into a denoising-optimized subband coefficient set, and the information entropy weights of each subband are calculated based on the Shannon entropy formula. Specifically, the information entropy of each subband coefficient in the denoising-optimized subband coefficient set is calculated again to reflect the remaining effective information volume in the frequency band after denoising. Then, the information entropies of all subbands are summed up to obtain the total information entropy value. Thus, the information entropy weight of each subband is obtained by calculating the ratio of the information entropy of each subband to the total information entropy value. Finally, weighted superposition is performed on the denoising-optimized subband coefficient set according to the weights to generate a three-dimensional feature tensor, where the first dimension corresponds to time points, the second dimension corresponds to spatial sampling positions, and the third dimension represents the key frequency band features extracted by the entropy fusion method. For example, in an actual sample cable, when the acoustic emission sensor detects local spike signals in the 200 - 300 kHz frequency band, the entropy value of the corresponding subband is relatively high, and a prominent response area is formed in the feature tensor after fusion, thus providing high signal-to-noise ratio data support for subsequent defect location and quantification.

[0026] Step S2: Use a preset graph attention network model to perform anomaly recognition and extraction on the feature tensor to obtain a potential defect feature vector.

[0027] When performing high-order singular value decomposition on the denoising feature tensor, first consider the three-dimensional tensor as a tensor product in three modes (time, space, and feature), and realize the dimensionality reduction mapping by solving the orthogonal basis vectors and core tensors of each mode: Before mapping, the dimension of the feature tensor may be 1000×100×5, and the core tensor after mapping becomes 50×50×3. This mapping can not only retain more than 90% of the energy in the original data but also compress more than 90% of the redundant information, thus greatly reducing the complexity of subsequent correlation calculations. For example, when processing multi-physical field data of a 10m wire and cable in one second, the memory occupancy and computational amount can be reduced by dozens of times.

[0028] Then, regard each element in the core tensor as the potential node attribute of the graph structure, and calculate the correlation strength between each pair of nodes, that is, consider both the statistical correlation degree and the attribute difference between the two nodes. Specifically, first regard the attribute vectors of the two nodes as random variables, and evaluate their synchronization degree in multiple space-time and physical field measurement dimensions through the statistical method of mutual information; then calculate the Euclidean distance between the attribute vectors of this pair of nodes, and use an exponential decay function with the square of this distance as the independent variable to suppress the node pairs with larger distances; finally, multiply the mutual information value by this decay factor to obtain the edge weight that can reflect both synchronization and similarity.

[0029] Subsequently, all nodes are screened through a preset threshold, only the edges with edge weights greater than the threshold are retained, and a weighted undirected graph labeled as G=(V,E) is constructed, where the vertex set V comes from all elements of the core tensor, and the edge set E consists of highly correlated node pairs. This spatio-temporal graph structure effectively transforms the three-dimensional manifold into an adjacency relationship that can be processed by a graph neural network. A typical application is that when a thermal-acoustic synchronization anomaly occurs in a section of cable, a high-weight edge will be formed between the corresponding two spatio-temporal nodes, thus highlighting this anomaly pattern in the graph structure.

[0030] Then, based on the spatio-temporal graph structure, a preset graph attention network model is used for anomaly detection. In the graph attention network, the attention coefficients between each node and its neighbors are generated through a two-stage scoring and normalization process. First, each node multiplies its own vector and the vector of a certain neighbor by the same learnable linear mapping matrix respectively, then the mapping results are concatenated into a longer vector, the inner product of this concatenated vector and a group of learnable weight vectors is taken, and then it is processed by a leaky ReLU activation function to obtain an original scoring value. Next, the scoring values of all neighbors are first exponentiated to enhance the differences, then they are added up to obtain the normalized denominator, and the exponential score of each edge is divided by this denominator, so as to obtain an attention coefficient that falls between zero and one and the sum of all neighbors is one. This attention coefficient reflects the importance of neighbor features to the target node during the graph propagation process. After completing the multi-head attention aggregation, in order to quantify the anomaly degree of each node in its local graph, it is necessary to combine the attention coefficient with the edge weight calculated before. Specifically, for the target node, the attention coefficients corresponding to it and all its neighbors are multiplied by their respective edge weights in turn, then the sum of these products is obtained to get a weighted association sum, and finally the natural logarithm is taken after adding one to this sum. Taking the logarithm can not only suppress the influence of extreme values, but also retain the proportional relationship of the association strength between different nodes. Therefore, the obtained value can be regarded as the anomaly response score of this node. The larger the value, the more it is emphasized by both the attention mechanism and the statistical association measure, and thus it is more likely to correspond to a real defect. For example, when the resistance of a certain area suddenly increases and is accompanied by an acoustic emission spike at the same moment, a higher attention weight and mutual information will be generated, resulting in a significant increase in this value.

[0031] Finally, based on the anomaly response operator values of all nodes, the node features corresponding to the top 128 maximum values are selected as the potential defect feature vectors. This vector not only retains the most typical defect pattern response, but also significantly compresses the data dimension, facilitating subsequent physical parameter inversion and quality inference. Through the above whole process, from the multi-physical field tensor to the graph structure and then to the graph attention model, not only the in-depth correlation mining of high-dimensional data is realized, but also the effectiveness of accurately locating defect nodes in the cable thermal-acoustic spatio-temporal co-occurrence anomaly scenario is verified in the example.

[0032] Step S3: Perform multi-physical-field coupling simulation inversion on the potential defect feature vector to obtain a set of defect quantification parameters.

[0033] The mapping from the potential defect feature vector to the physical defect parameters first computationally couples the abstract high-dimensional features with the actual multi-physical behavior of the cable. First, by coupling the feature vector through the preset Maxwell's electromagnetic equations, heat conduction equation, and structural mechanics equations, a finite element model that can simultaneously describe the interaction of the electric field, temperature field, and stress field can be constructed. The necessity of this finite element model lies in that defects such as cracks, impurities, or eccentricity inside the cable often leave different signal traces in the three fields of electromagnetic response, heat diffusion, and mechanical deformation, and a single physical field cannot invert the complete defect parameters. For example, a conductor crack will cause a sudden increase in local resistivity (electromagnetics), heat accumulation (heat conduction), and stress concentration (structural mechanics). Only after the three are coupled can a simulation output matching the traceability feature vector be obtained, thus realizing a closed-loop mapping in the physical sense.

[0034] After completing the construction of the multi-physical-field finite element model, it is necessary to minimize the simulation of the model through the preset inversion objective function to clarify the mathematical expression of the quantification problem. This objective function not only measures the gap between the simulation output and the feature vector in the multi-physical domain but also considers the physical feasibility of the parameters. The inversion objective function can be expressed by the following formula: where, is the joint spatial domain of the cross-section and length of the wire and cable. represents the multi-physical-field response vector obtained by finite element solution under the given parameter set ={l, c, e} (i.e., crack length, impurity concentration, eccentricity). is the feature vector (including electromagnetic field strength, temperature field gradient, and strain field components) at the same spatial position extracted by the graph attention network. The weight function is adaptively allocated according to the spatial distribution density of the sensors, so that it has a higher weight in the key detection areas (for example, the paragraphs where fiber optic strain sensors are densely deployed). is used to prevent overfitting caused by over-reliance on a certain physical quantity in the inversion result, thus ensuring that the output parameters have good physical interpretability. The inversion objective function realizes the seamless connection between the feature space and the physical space by minimizing the weighted simulation residuals in the spatial domain. For example, when a micron-level crack appears at x0 in a certain section of the cable, the finite element simulation will generate an obvious peak in the temperature gradient and strain concentration at this position, making the optimal that minimizes the residual close to the true crack length.

[0035] Subsequently, in order to utilize the advantages of the quantum annealer in high-dimensional and non-convex optimization, the above continuous inversion problem needs to be mapped into a discrete Ising Hamiltonian problem. This mapping is achieved by designing a set of binary spin variables to approximate the discrete value range of each parameter and constructing a Hamiltonian function. Among them, the binary spin variable σ i ∈{-1, +1}. Specifically, first, each physical parameter to be inverted is mapped to a binary spin variable. By analyzing the gradient sign of the inversion objective function at this parameter, the local spin field corresponding to the parameter with a negative gradient is set to negative, and the local spin field with a positive gradient is set to positive. This process is equivalent to establishing a single-spin energy term that tends to increase or decrease for each parameter. Then, considering the mutual coupling characteristics of crack length, impurity concentration, and eccentricity that affect each other in physical field simulation, through statistical analysis of the coupling sensitivity of each parameter in the finite element model, the coupling coefficient between each pair of spins is determined. If two parameters show a common change trend in stress concentration or heat diffusion, a positive coupling energy is assigned between them to make these two spins tend to the same value; otherwise, a negative coupling energy is assigned to reflect the opposite change tendency. All local spin fields and coupling coefficients are combined to form an overall Ising Hamiltonian function, which characterizes the system energy level of each parameter spin under different value combinations. The Ising Hamiltonian design can organically express the multi-physical field coupling characteristics in the quantum annealing energy landscape.

[0036] Finally, the Ising Hamiltonian is input into the D-Wave quantum annealing processor, and the global lowest energy state is automatically searched through the quantum tunneling effect and the annealing process. The obtained optimal spin combination can be restored to a continuous parameter set (i.e., the defect quantification parameter set) θ={l*, c*, e*} through a preset mapping rule. This is the defect quantification parameter that best matches the eigenvector. For example, if the probability that the spin representing the crack length in the quantum annealing result biases towards +1 is higher than 90%, then the corresponding l* obtained after continuous mapping will be very close to the crack size implied by the high-entropy acoustic emission characteristics prompted by the aforementioned graph attention network, such as about 0.3 mm. The whole process realizes the closed-loop inversion from high-dimensional features to engineering quantification parameters, providing defect indicators that can be directly used for production decision-making for the online detection system.

[0037] Step S4: Perform dynamic reasoning based on the defect quantification parameter set through a preset Bayesian network model to obtain an online quality grade decision.

[0038] According to the defect quantification parameter set θ = {l*, c*, e*} obtained by multi-physics field simulation inversion, where l* represents the crack length, c* represents the impurity concentration, and e* represents the insulation eccentricity. First, input these parameters into the bottom-layer nodes in the Bayesian network model, and accordingly calculate the local prediction distributions of each fault mode node and quality grade node based on the pre-trained conditional probability table. The model structure adopts a three-layer directed acyclic graph, with the bottom-layer nodes being θ, the middle-layer nodes representing process fault modes such as wire drawing anomalies, insufficient cross-linking, and cooling defects, and the top-layer nodes corresponding to quality grade categories (e.g., excellent, qualified, rework, scrap). In this step, the "prediction" performed after mapping the feature vector to specific parameters is not simply querying the existing table, but by matching the input value of each bottom-layer node to its CPT row to obtain local conditional probabilities such as P(M1 = 1|l*, c*, e*) and P(Q = q|M1, M2, M3), thereby forming a preliminary probability assignment for all nodes within the network. This step is significant in mapping the abstract physical parameters to the probability of whether a fault occurs and the likelihood of the final quality grade in a probabilistic manner, laying the foundation for subsequent global inference. For example, if the crack length l* exceeds 0.2 mm and the impurity concentration c* exceeds 50 ppm, then P(M1 = 1|l*, c*, e*) may be above 0.9, indicating a high probability of wire drawing anomalies.

[0039] After completing the local prediction, it is necessary to integrate the conditional probabilities of these nodes into the posterior distribution calculation to achieve the global probability assessment of each quality grade under the given defect quantification parameter set θ. Traditional exact calculations are extremely time-consuming in high-dimensional parameter spaces, so Monte Carlo sampling is used to approximately solve the problem. Let the sampling points {θ (i)} N i=1 represent multiple groups of candidate parameters generated around the parameter space. First, calculate the joint likelihood of each group of parameters on each fault mode and quality grade node through the network. Specifically, it is the likelihood of observing this combination when a given sample parameter and fault mode combination are provided. By averaging the likelihood values of all samples, the overall support for different quality grades under the current parameter set can be obtained, thereby overcoming the problem that single-point estimation is vulnerable to noise. Then, accumulate for the same quality grade q over all sampling points to obtain an approximate likelihood function. During the accumulation process, the selection of the number of samples directly affects the confidence of the posterior estimate. Usually, at the 10 4 level can ensure a variance error of less than 1%. Detect the response stability of the network to the true parameter set through a large number of parameter perturbations, thereby resisting the bias that may be brought by single-point estimation.

[0040] In an alternative embodiment, after obtaining the posterior probability of each grade, in order to reflect different degrees of parameter uncertainty, an entropy-based temperature adjustment mechanism is further introduced: if the parameter exhibits a large entropy value in its sampling distribution (i.e., high uncertainty), the posterior probability is flattened so that the network does not overly favor a certain grade when highly uncertain; conversely, when the parameter stability is high, the sharp distribution of the posterior probability is retained to highlight the most likely grade.

[0041] Finally, according to the maximum a posteriori probability principle, the quality grade with the strongest support is selected as the online quality grade decision. This decision takes into account both the uncertainty of physical parameters and the prior knowledge of network inference. For example, when the parameter set exhibits a high level of cracks and impurities but a low eccentricity, the posterior distribution after temperature adjustment may result in probabilities of approximately 40% and 30% for "rework products" and "scrap products" respectively, while the probability of "qualified products" is pulled down to about 20%. Eventually, the MAP determines it as a "rework product" with a probability confidence of 40%, providing a quantitative and interpretable quality classification result for the production line.

[0042] Step S5: Package the online quality grade decision according to the preset blockchain intelligent protocol to obtain a quality traceability record.

[0043] After obtaining the online quality grade decision result, first, the core detection data and the decision result are fixed by hashing to ensure that they cannot be tampered with. At this time, the SHA3-256 hashing algorithm is applied to the original wire and cable signal matrix, the noise reduction feature tensor, the potential defect feature vector, and the defect quantification parameter set respectively to obtain four irreversible 256-bit digest values H1, H2, H3, H4. Then, the timestamp T accepted by the current blockchain network is also hashed in the same way to obtain H t =SHA3-256(T). These hash values are then combined according to the following self-designed hierarchical synthesis formula , by first hashing H1 and H2, and H3 and H4 pairwise respectively, and then concatenating with the timestamp digest and hashing the whole, not only can the data integrity and time be inseparably bound to a single root, but also provide multi-level verification for any single-piece data or timestamp forgery during verification. For example, if someone attempts to modify any byte of the feature tensor, its corresponding H2 will change, thereby causing the intermediate node hash and the final root hash to not match, immediately exposing the tampering attempt.

[0044] After obtaining the Merkle root hash value, next, the online quality grade decision, the confidence level, and the root hash value are packaged through the intelligent protocol and written into the blockchain. The deployed protocol interface will receive three parameters: the decision string, the root hash With the precise timestamps of the blockchain, the protocol generates a transaction log and stores these three elements together in the blockchain ledger. This transaction cannot be rolled back or tampered with and can be instantly retrieved through a block explorer, enabling any subsequent audit to verify the authenticity of data traceability by comparing the records on the chain with the local data hash. For example, when the test result of cable batch A is "rework product, confidence level 85% and corresponding root hash is 0xabc...123", anyone can query this transaction on the chain and confirm its integrity by repeating the above hash synthesis process.

[0045] Meanwhile, to alleviate the storage pressure on the blockchain network and protect the privacy of the original data, the core data is not directly uploaded to the chain but is encrypted in slices and stored in a distributed storage network. First, the original wire and cable signal matrix, noise reduction feature tensor, potential defect feature vector, and defect quantization parameter set are divided into several data slices according to a fixed byte size, denoted as {S j} n j=1 ; then, the symmetric encryption algorithm E k (for example, AES-256) is applied to each slice to obtain the encrypted slice C j = E k (S j ); the encryption process maintains data confidentiality, and only authorized devices holding the key k can restore it to the plaintext content. Subsequently, each encrypted slice is uploaded to IPFS, and IPFS returns a content identifier CID j for each slice. This CID is essentially the SHA3-256 hash of the encrypted slice. Finally, after concatenating all the CIDs, another round of SHA3-256 hash is performed to obtain the aggregated CID index, and this aggregated index is written into the blockchain through the same intelligent protocol interface. This approach not only ensures the verifiability of distributed data storage but also enables anyone to confirm that the slices have not been tampered with by comparing the aggregated CID index recorded on the blockchain with the individual CIDs, and at the same time, it can quickly retrieve and decrypt to restore the complete original data from the IPFS network when needed.

[0046] Through the above chain-down-chain design that organically combines Merkle tree hashing, intelligent protocol evidence storage, and distributed encrypted storage, it not only provides end-to-end traceability capabilities for online quality level decision-making but also achieves an optimal balance among data security, storage efficiency, and auditability, laying a solid foundation for the transparency and trustworthiness of the cable production and quality management processes.

[0047] In an alternative embodiment, to achieve continuous adaptive optimization of the detection system and intuitive human-machine interaction decision-making, the method further includes: First, construct a loss function that optimizes both comprehensive quality determination and system energy consumption, taking the original wire and cable signal matrix, noise reduction feature tensor, potential defect feature vector, defect quantization parameter set, and online quality grade decision as inputs. This loss function is continuous in both time and space and introduces a time-varying constraint on sensor power consumption. The loss function adopts a composite expression of time-weighted error and energy consumption accumulation, squares and integrates the deviation between the quality grade predicted by the model and the true decision weighted by the time decay coefficient, and at the same time integrates and linearly penalizes the power consumption of each processing unit during the entire detection cycle. Specifically, at each moment, first calculate the square of the Euclidean distance between the quality grade vector output by the model at this moment and the ground truth vector, which reflects the judgment accuracy; then multiply this error by an exponential decay weight with the difference between the current moment and the end of the detection cycle as the independent variable. The negative coefficient in the exponent makes the error at a later time receive more attention; subsequently, perform a definite integral on the weighted errors at all time points to obtain the total time-weighted error. In parallel, integrate the power consumption curves of all edge computing devices during the same period to obtain the total energy consumption, then multiply it by an energy consumption regularization coefficient and add it to the error integral to achieve a balance between accuracy and energy consumption.

[0048] Next, use automatic differentiation technology to accurately differentiate the above loss function and calculate the gradient values of all trainable parameters in the model (i.e., model parameter gradient values). The automatic differentiation algorithm will internally split the loss expression into basic operation units and automatically derive the partial derivatives of each parameter at each step according to the chain rule. For example, for the weighted error term, first calculate the derivative of the exponential decay weight with respect to the model output, and then multiply it by twice the difference between the output and the true value to obtain the gradient of this error term with respect to the output; then propagate backward layer by layer according to the operation graph of each layer of the model to accurately obtain the derivative value of the loss function with respect to each network weight or bias, realizing efficient and reliable gradient calculation.

[0049] In order to achieve collaborative self-optimization across multiple production lines or multi-location equipment, the embodiment of the present application adopts a federated learning model based on secure multi-party computing. Each local node first encrypts the model parameter gradient value calculated by each node, and then converges it to an aggregation server through an encrypted communication protocol. The server performs homomorphic addition while maintaining the gradient ciphertext to obtain the encrypted global gradient sum, and then the threshold decryption protocol in which all participants participate decrypts the true gradient. In this way, the global model parameter set is completed without leaking any local original data and model details. It should be understood that the global model parameter set includes the weight matrix and attention vector of the graph attention network, the conditional probability table of each node of the hierarchical Bayesian network, and the weights and biases of the optional edge computing network (such as a feedforward neural network or a convolutional network). Its role is to aggregate the gradient information fed back by the local models of different devices in the real operating environment and facing multi-source sensor data to a central perspective, thereby obtaining a new model that integrates multi-party experience and takes into account global performance. After the global model parameter set is distributed back to each node, the inspection system can maintain highly consistent defect recognition capabilities and quality grading standards between production lines, while taking into account the optimization of energy consumption and speed, achieving adaptive compensation for process drift and rapid response to new abnormal patterns, thereby greatly improving the robustness and accuracy of online inspection.

[0050] After the model update is completed, the Iterative Closest Point (ICP) algorithm is called to rigidly register the cable surface point cloud extracted from the original wire and cable signal matrix with the preset virtual 3D model to generate accurate model mapping data. The ICP algorithm is used to align the actual collected cable surface point cloud with the virtual 3D model. Its essence is a least squares optimization: in each iteration, the current point cloud and the model point pair are first matched by the nearest neighbor, and then the rigid transformation parameters for rotating and translating the entire point cloud to the model coordinates are calculated based on these correspondences. This transformation parameter is solved by minimizing the sum of the squared Euclidean distances between all point pairs, and is continuously updated in each iteration until it converges to the optimal position, ensuring that the virtual model is highly consistent with the real cable space position, providing an accurate geometric basis for subsequent holographic rendering. After the registration is completed, the obtained model mapping data contains both the real spatial geometry of the cable and the one-to-one coordinate relationship with the simulation model, which greatly improves the accuracy and stability of subsequent visualization rendering.

[0051] Finally, based on the model mapping data and the defect quantification parameter set, perform holographic rendering in the AR terminal to generate a human-machine collaborative decision-making interface. Holographic rendering utilizes the ray tracing and depth synthesis technologies of the GPU to superimpose the internal crack positions, lengths, and impurity distributions of the cable on the real view in the form of a semi-transparent red grid. At the same time, the contour is enhanced through edge filtering, making potential defects visible to the operator at a glance. The interface also supports eye tracking, and when the line of sight stays in the crack highlight area for more than two seconds, the parameter details of this position will automatically pop up; gesture recognition can adjust the transparency of the virtual layer or trigger data playback, and finally, the intuitive confirmation of the cable quality and decision input are completed under human-machine collaboration.

[0052] This application is applied to the field of Internet of Things technology. By performing multi-scale wavelet packet decomposition and signal entropy fusion on the original wire and cable signal matrix, a feature tensor is obtained. Through the graph attention network model, the feature tensor is abnormally identified and extracted to obtain a potential defect feature vector. The potential defect feature vector is subjected to multi-physical field coupling simulation inversion to obtain a defect quantification parameter set. Through the Bayesian network model, dynamic reasoning is performed based on the defect quantification parameter set to obtain an online quality grade decision. According to the blockchain intelligent protocol, the online quality grade decision is encapsulated to obtain a quality traceability record. This application realizes the real-time, accurate, traceable, and secure online quality detection of wire and cable through the organic coupling of multiple levels, multiple models, and multiple technical means.

[0053] As Figure 2 shown, it is a functional module diagram of an online quality detection device for wire and cable provided by an embodiment of this application.

[0054] In some embodiments, the online quality detection device 2 for wire and cable may include multiple functional modules composed of computer program segments. The computer programs of each program segment in the online quality detection device 2 for wire and cable can be stored in the memory of the server and executed by at least one processor to execute the functions of the online quality detection method for wire and cable (see the details in Figure 1 the description).

[0055] In this embodiment, the online quality detection device 2 for wire and cable can be divided into multiple functional modules according to the functions it performs. The functional modules may include: a noise reduction and fusion module 21, a defect identification module 22, a coupling simulation module 23, a dynamic reasoning module 24, a decision encapsulation module 25, and a human-computer interaction module 26. The module referred to in the present invention means a series of computer program segments that can be executed by at least one processor and can complete fixed functions, and are stored in the memory. In this embodiment, the functions of each module will be described in detail in subsequent embodiments.

[0056] The noise reduction and fusion module 21 is used to perform multi-scale wavelet packet decomposition and signal entropy fusion on the synchronously collected original wire and cable signal matrix to obtain a feature tensor.

[0057] In an optional embodiment, the noise reduction and fusion module 21 is specifically configured to: Perform Daubechies wavelet decomposition on the synchronously collected original wire and cable signal matrix to obtain a set of subband coefficients; Perform soft threshold denoising on the top-layer high-frequency subband coefficients in the set of subband coefficients according to a preset SURE threshold function to obtain a denoising optimized set of subband coefficients; Calculate the information entropy weight of each subband by using a preset Shannon entropy formula for the denoising optimized set of subband coefficients; Perform information entropy fusion on the denoising optimized set of subband coefficients according to the information entropy weight to obtain the feature tensor.

[0058] The defect identification module 22 is used to perform anomaly identification and extraction on the feature tensor through a preset graph attention network model to obtain a potential defect feature vector.

[0059] In an optional embodiment, the defect identification module 22 is specifically configured to: Perform higher-order singular value decomposition on the feature tensor to obtain a core tensor, and calculate the mutual information measure for each element in the core tensor to obtain the edge weights between each element; Connect each element in the core tensor according to the edge weights and a preset threshold to construct a spatio-temporal graph; Perform anomaly detection on the spatio-temporal graph through a preset graph attention network model to obtain the potential defect feature vector.

[0060] The coupled simulation module 23 is used to perform multi-physical field coupled simulation inversion on the potential defect feature vector to obtain a set of defect quantification parameters.

[0061] In an optional embodiment, the coupled simulation module 23 is specifically configured to: Perform multi-physical field coupling on the potential defect feature vector according to a preset Maxwell electromagnetic equation set, heat conduction equation, and structural mechanics equation to obtain a multi-physical field finite element model; Perform minimization simulation on the multi-physical field finite element model according to a preset inversion objective function to obtain an inversion problem; Perform mapping processing on the inversion problem through a preset Ising model to obtain an Ising Hamiltonian, and perform D-Wave quantum annealing solution processing on the Ising Hamiltonian to obtain the set of defect quantification parameters.

[0062] A dynamic inference module 24, configured to perform dynamic inference based on the defect quantization parameter set through a preset Bayesian network model to obtain an online quality level decision.

[0063] In an optional embodiment, the dynamic inference module 24 is specifically configured to: Perform decision prediction based on the defect quantization parameter set through a preset Bayesian network model to obtain a conditional probability table for each node; Perform posterior distribution calculation on the defect quantization parameter set and the conditional probability table through a preset Monte Carlo sampling likelihood function to obtain quality level conditions and corresponding posterior probabilities; Perform MAP decision screening on the quality level conditions according to the posterior probabilities to obtain the online quality level decision.

[0064] A decision encapsulation module 25, configured to encapsulate the online quality level decision according to a preset blockchain intelligent protocol to obtain a quality traceability record.

[0065] In an optional embodiment, the decision encapsulation module 25 is specifically configured to: Perform hashing processing and combination on the original wire and cable signal matrix, the feature tensor, the potential defect feature vector, and the defect quantization parameter set according to a preset SHA3-256 algorithm to obtain a Merkle root hash value; Encapsulate and store the online quality level decision and the corresponding Merkle root hash value according to a preset blockchain intelligent protocol and a timestamp to obtain the decision record; Perform distributed encrypted storage on the original wire and cable signal matrix, the feature tensor, the potential defect feature vector, and the defect quantization parameter set to obtain encrypted data shards and corresponding CID indexes, and encapsulate and store the CID indexes according to the blockchain intelligent protocol to obtain the content identification record.

[0066] In an optional embodiment, the wire and cable online quality detection device 2 further includes a human-computer interaction module 26, and the human-computer interaction module 26 is specifically configured to: Construct a loss function according to the original wire and cable signal matrix, the feature tensor, the potential defect feature vector, the defect quantization parameter set, and the online quality level decision; Perform automatic differentiation calculation on the loss function to obtain a model parameter gradient value; Perform secure multi-party calculation and aggregation decryption on the model parameter gradient value through a preset federated learning model to obtain a global model parameter set; Perform point cloud registration on the surface point cloud data in the original wire and cable signal matrix and a preset virtual model through a preset ICP algorithm to obtain model mapping data; Perform AR terminal holographic rendering on the defect quantization parameter set according to the model mapping data to generate a human-machine collaborative decision-making interface.

[0067] It should be understood that the various change methods and specific embodiments in the methods provided in the above embodiments are equally applicable to the wire and cable on-line quality detection device in this embodiment. Through the foregoing detailed description of the wire and cable on-line quality detection method, those skilled in the art can clearly know the implementation method of the wire and cable on-line quality detection device in this embodiment. For the sake of brevity of the specification, it will not be elaborated here.

[0068] As Figure 3 shown, it is a schematic structural diagram of an electronic device provided by an embodiment of the present application.

[0069] In a preferred embodiment of the present invention, the electronic device 3 may include, but is not limited to: a memory 31, at least one processor 32, and at least one communication bus 33.

[0070] Those skilled in the art should understand that Figure 3 the structure of the electronic device 3 shown does not constitute a limitation on the embodiments of the present invention. The electronic device 3 may further include more or fewer other hardware or software than shown, or different component arrangements.

[0071] In some embodiments, the electronic device 3 is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, application-specific integrated circuits, programmable gate arrays, digital processors, and embedded devices, etc.

[0072] It should be noted that the electronic device 3 is only an example, and other existing or future possible electronic products that can be adapted to the present application should also be included in the protection scope of the present application and are included herein by reference.

[0073] In some embodiments, a computer program is stored in the memory 31. When the computer program is executed by the at least one processor 32, all or part of the steps in the on-line quality detection method for the wire and cable as described above are implemented. The memory 31 includes a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), a one-time programmable read-only memory (OTPROM), an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM), or other optical disc memories, magnetic disk memories, tape memories, or any other computer-readable medium capable of carrying or storing data. Further, the computer-readable storage medium mainly includes a program storage area and a data storage area. Among them, the program storage area can store an operating system, application programs required for at least one function, and the like.

[0074] In some embodiments, the at least one processor 32 is the control core (Control Unit) of the electronic device 3, connecting various components of the entire electronic device 3 through various interfaces and circuits. By running or executing the programs or modules stored in the memory 31, and calling the data stored in the memory 31, various functions of the electronic device 3 are executed and data is processed. For example, when the at least one processor 32 executes the computer program stored in the memory 31, all or part of the steps in the on-line quality detection method for the wire and cable in the embodiments of the present application are implemented; or all or part of the functions of the on-line quality detection device for the wire and cable are implemented. The at least one processor 32 may be composed of integrated circuits. For example, it may be composed of a single packaged integrated circuit, or may be composed of multiple integrated circuits with the same or different functions packaged, including a combination of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips.

[0075] In some embodiments, the at least one communication bus 33 is configured to enable connection communication between the memory 31 and the at least one processor 32, etc. Although not shown, the electronic device 3 may further include a power source (such as a battery) for powering each component. Preferably, the power source can be logically connected to the at least one processor 32 through a power management device, so as to implement functions such as management of charging, discharging, and power consumption management through the power management device. The power source may further include any components such as one or more DC or AC power sources, a recharge device, a power failure detection circuit, a power converter or inverter, a power status indicator, etc. The electronic device 3 may further include a variety of sensors, a Bluetooth module, a Wi-Fi module, etc., which will not be elaborated here.

[0076] The integrated units implemented in the form of software function modules as described above can be stored in a computer-readable storage medium. The above software function modules are stored in a storage medium and include several instructions for causing an electronic device (which may be a personal computer, an electronic device, or a network device, etc.) or a processor to execute parts of the methods described in various embodiments of the present application.

[0077] In several embodiments provided by the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0078] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units. They can be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0079] The above are all preferred embodiments of the present application. Without limiting the protection scope of the present application accordingly, therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be covered within the protection scope of the present application.

Claims

1. An on-line quality inspection method for wire and cable, characterized in that, The method includes: Performing multi-scale wavelet packet decomposition and signal entropy fusion on the synchronously collected original wire and cable signal matrix to obtain a feature tensor; Performing anomaly recognition and extraction on the feature tensor through a preset graph attention network model to obtain a potential defect feature vector; Performing multi-physical field coupling simulation inversion on the potential defect feature vector to obtain a set of defect quantification parameters; Performing dynamic inference based on the set of defect quantification parameters through a preset Bayesian network model to obtain an online quality grade decision; Encapsulating the online quality grade decision according to a preset blockchain intelligent protocol to obtain a quality traceability record.

2. The on-line quality inspection method for wire and cable according to claim 1, characterized in that, The performing multi-scale wavelet packet decomposition and signal entropy fusion on the synchronously collected original wire and cable signal matrix to obtain a feature tensor includes: Performing Daubechies wavelet decomposition on the synchronously collected original wire and cable signal matrix to obtain a set of sub-band coefficients; Performing soft threshold denoising on the top-level high-frequency sub-band coefficients in the set of sub-band coefficients according to a preset SURE threshold function to obtain a denoising optimized set of sub-band coefficients; Calculating the information entropy weight of each sub-band by performing weight calculation on the denoising optimized set of sub-band coefficients through a preset Shannon entropy formula; Performing information entropy fusion on the denoising optimized set of sub-band coefficients according to the information entropy weight to obtain the feature tensor.

3. The on-line quality inspection method for wire and cable according to claim 1, characterized in that, The performing anomaly recognition and extraction on the feature tensor through a preset graph attention network model to obtain a potential defect feature vector includes: Performing high-order singular value decomposition on the feature tensor to obtain a core tensor, and calculating the mutual information measure for each element in the core tensor to obtain the edge weights between each element; Connecting each element in the core tensor according to the edge weights and a preset threshold to construct a spatio-temporal graph; Performing anomaly detection on the spatio-temporal graph through a preset graph attention network model to obtain the potential defect feature vector.

4. The on-line quality inspection method for wire and cable according to claim 1, characterized in that The performing multi-physical field coupling simulation inversion on the potential defect feature vector to obtain a set of defect quantification parameters includes: Performing multi-physical field coupling on the potential defect feature vector according to a preset Maxwell electromagnetic equation set, heat conduction equation, and structural mechanics equation to obtain a multi-physical field finite element model; Performing minimization simulation on the multi-physical field finite element model according to a preset inversion objective function to obtain an inversion problem; Performing mapping processing on the inversion problem through a preset Ising model to obtain an Ising Hamiltonian, and performing D-Wave quantum annealing solution processing on the Ising Hamiltonian to obtain the set of defect quantification parameters.

5. The on-line quality inspection method of wire and cable according to claim 1, characterized in that, The performing dynamic inference based on the set of defect quantification parameters through a preset Bayesian network model to obtain an online quality grade decision includes: Performing decision prediction based on the set of defect quantification parameters through a preset Bayesian network model to obtain a conditional probability table for each node; Performing posterior distribution calculation on the set of defect quantification parameters and the conditional probability table through a preset Monte Carlo sampling likelihood function to obtain quality grade conditions and corresponding posterior probabilities; Perform MAP decision screening on the quality level condition according to the posterior probability to obtain the online quality level decision.

6. The on-line quality inspection method for wire and cable according to claim 1, characterized in that, The quality traceability record includes a decision record and a content identification record. Encapsulating the online quality level decision according to a preset blockchain intelligent protocol to obtain the quality traceability record includes: Perform hashing processing and combination on the original wire and cable signal matrix, the feature tensor, the potential defect feature vector, and the defect quantization parameter set according to a preset SHA3-256 algorithm to obtain the Merkle root hash value; Encapsulate and store the online quality level decision and the corresponding Merkle root hash value according to a preset blockchain intelligent protocol and a time stamp to obtain the decision record; Perform distributed encrypted storage on the original wire and cable signal matrix, the feature tensor, the potential defect feature vector, and the defect quantization parameter set to obtain encrypted data shards and corresponding CID indexes, and encapsulate and store the CID indexes according to the blockchain intelligent protocol to obtain the content identification record.

7. The on-line quality inspection method for wire and cable according to claim 5, characterized in that, The method further includes: Construct a loss function according to the original wire and cable signal matrix, the feature tensor, the potential defect feature vector, the defect quantization parameter set, and the online quality level decision; Perform automatic differential calculation on the loss function to obtain the model parameter gradient value; Perform secure multi-party calculation and aggregation decryption on the model parameter gradient value through a preset federated learning model to obtain the global model parameter set; Perform point cloud registration on the surface point cloud data in the original wire and cable signal matrix and a preset virtual model through a preset ICP algorithm to obtain model mapping data; Perform AR terminal holographic rendering on the defect quantization parameter set according to the model mapping data to generate a human-machine collaborative decision-making interface.

8. An on-line quality inspection device for wire and cable, characterized in that, The device includes: A noise reduction and fusion module, configured to perform multi-scale wavelet packet decomposition and signal entropy fusion on the synchronously collected original wire and cable signal matrix to obtain a feature tensor; A defect identification module, configured to perform anomaly identification and extraction on the feature tensor through a preset graph attention network model to obtain a potential defect feature vector; A coupled simulation module, configured to perform multi-physical field coupled simulation inversion on the potential defect feature vector to obtain a defect quantization parameter set; A dynamic inference module, configured to perform dynamic inference according to the defect quantization parameter set through a preset Bayesian network model to obtain an online quality level decision; A decision encapsulation module, configured to encapsulate the online quality level decision according to a preset blockchain intelligent protocol to obtain a quality traceability record.

9. An electronic device, characterized in that, The electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the steps of the wire and cable online quality detection method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the wire and cable online quality detection method according to any one of claims 1 to 7.

Images