A multi-dimensional detection method for quality defects of wire and cable production

Abstract

This invention discloses a multi-dimensional detection method for quality defects in wire and cable manufacturing, belonging to the field of wire and cable manufacturing and testing technology. The method involves real-time acquisition of process data using a high-precision multi-modal sensor array deployed on the production line; calculation of microscopic state parameters along the cable length using an online material micro-state deduction module; simultaneous acquisition of apparent, electrical, thermal, and mechanical performance parameters of the cable through a multi-dimensional macroscopic performance detection unit; and construction of feature vectors by spatiotemporal fusion of multi-source data. This invention is particularly applicable to the intelligent production of cross-linked polyethylene insulated power cables, providing key material guarantees for intelligent large-scale power equipment (such as DC converter transformers and intelligent reactors) and intelligent power distribution systems, realizing a leap from finished product inspection to process early warning and control in quality management. It enables early warning of defects, accurate tracing of root causes, and process self-optimization, significantly improving production quality and efficiency.

Description

Technical Field

[0001] This invention belongs to the field of wire and cable manufacturing and testing technology, specifically relating to a multi-dimensional detection method for quality defects in wire and cable production. Background Technology

[0002] Electric wires and cables are key basic components for power transmission and information communication, and their quality directly affects power grid safety, equipment reliability, and system lifespan. Currently, with the increasing demands on cable performance in fields such as ultra-high voltage power transmission, new energy equipment, and high-speed rail transportation, traditional quality control methods that rely on manual visual inspection, spark testing, or single-point parameter sampling are no longer sufficient to meet the manufacturing requirements of high consistency and high reliability.

[0003] Chinese patent application CN114923525A proposes a multi-dimensional inspection scheme. By integrating appearance monitoring and performance monitoring modules, it comprehensively inspects the cable's dimensions, weight, surface defects, electrical and mechanical properties, and uses artificial intelligence algorithms to classify defects. This scheme represents an advancement from single-parameter detection to systematic evaluation, and is particularly suitable for the final inspection stage before finished products leave the factory. However, the aforementioned existing technology still has several shortcomings: Existing methods are essentially static snapshot-style inspections of the "finished product state," with data acquisition and analysis occurring after the manufacturing process. However, in critical processes such as extrusion, cross-linking, and cooling, the evolution of the material's microstructure (e.g., crystallinity, degree of cross-linking), process fluctuations (e.g., temperature gradients, tension abrupt changes), and environmental disturbances often occur before macroscopic defects appear. Because existing technologies cannot perceive these process dynamics in real time, they struggle to identify early-stage defects, missing the optimal intervention window.

[0004] The existing systems typically operate their detection modules independently, lacking deep integration and correlation analysis between the collected macroscopic performance parameters (such as outer diameter and conductivity) and the process root causes of defects (such as temperature settings and traction speed). When complex operating conditions such as raw material batch fluctuations or equipment status drift occur, the system struggles to accurately trace the root cause of defects, leading to increased false alarms and missed detection rates.

[0005] Existing solutions focus on a "monitoring-alarm" model, lacking a real-time, automatic feedback path between detection results and the upstream process control system. Even when potential defects are identified, process parameters are typically adjusted based on manual experience, resulting in slow response, low accuracy, and an inability to achieve online self-optimization of the production process.

[0006] Therefore, the industry urgently needs an intelligent testing method that can deeply integrate process dynamic perception, real-time material state simulation, multi-dimensional performance online detection, and closed-loop adaptive control, in order to promote the fundamental transformation of wire and cable quality control from "post-inspection" to "process prevention". Summary of the Invention

[0007] The purpose of this invention is to provide a multi-dimensional detection method for quality defects in wire and cable production, which solves the technical problem that existing wire and cable quality inspection technologies rely solely on static inspection at the finished product stage, thus failing to achieve early warning of defects during production, real-time intervention in the process, and adaptive optimization of process parameters.

[0008] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: A multi-dimensional detection method for quality defects in wire and cable manufacturing, applied to a continuous wire and cable production line, includes the following steps: S1: Real-time acquisition of raw data streams reflecting the dynamics of the production process through high-precision multimodal sensor arrays deployed at key process nodes of the production line; S2: Input the raw data stream into the online material microstate deduction module, and deduce the material microstate parameters distributed along the cable length direction online based on the preset physicochemical kinetic model; S3: Through the multi-dimensional macroscopic performance testing unit, the cable semi-finished products on the production line are continuously tested online to obtain multi-dimensional macroscopic performance parameters, including appearance morphology, electrical characteristics, thermal behavior and mechanical properties. S4: Based on a unified time reference, the microscopic state parameters of the material and the multidimensional macroscopic performance parameters are spatiotemporally aligned and fused to construct a high-dimensional defect feature vector; S5: Input the high-dimensional defect feature vector into a pre-trained defect recognition model, which is a graph neural network embedded with material physics constraints, and output the defect type and probability of the cable. S6: Through the closed-loop feedback control interface, the defect type and probability, as well as the related abnormal process parameters, are fed back to the process control system of the production line in real time to trigger dynamic adjustment of key process parameters.

[0009] Furthermore, in step S1, the high-precision multimodal sensing array includes at least: a distributed fiber optic temperature sensor deployed at the extruder or cross-linking pipe to measure the material temperature field; a laser displacement sensor group deployed at the extrusion die outlet to measure the insulation layer size and concentricity; a tension-velocity composite sensor deployed in the traction section to measure instantaneous tension and linear velocity; and a composite probe deployed inside the cross-linking pipe to measure ambient humidity and oxygen concentration.

[0010] Furthermore, in step S2, the online material microstate deduction module calls the corresponding physicochemical kinetic model to perform deduction based on the type of cable material currently being produced on the production line. When the material is cross-linked polyethylene, the cross-linking kinetic model based on the Arrhenius equation is called, and the cross-linking degree distribution of the cable insulation layer along the length direction is deduced based on the temperature field measured by the distributed optical fiber temperature sensor and the historical linear velocity data measured by the tension-velocity composite sensor. When the material is polyvinyl chloride, the crystallization kinetic model based on the Avrami equation is called, and the crystallinity distribution of the cable sheath layer along the length direction is deduced based on the cooling rate measured by the thermal behavior detection subsystem.

[0011] Furthermore, in step S3, the multi-dimensional macroscopic performance testing unit includes: an appearance defect detection subsystem, which, based on a linear CCD camera and machine vision algorithms, identifies and quantifies scratches, bubbles, impurities, and color difference defects on the cable surface; an electrical performance testing subsystem, based on a non-contact eddy current probe and a dielectric strength testing module, measures the conductor conductivity and dielectric loss factor of the cable's insulation layer; a thermal behavior testing subsystem, based on an infrared thermal imager and a flash thermal diffusivity measurement device, measures the temperature field on the cable surface and the thermal diffusivity of the insulation layer; and a mechanical performance testing subsystem, based on an electromagnetic excitation and laser vibration measurement device, measures the mechanical loss factor or dynamic modulus of the cable.

[0012] Furthermore, step S4 specifically includes: S41: Synchronize the data streams output by the high-precision multimodal sensing array, the online material microstate inference module, and the multidimensional macroscopic performance detection unit to a unified time reference; S42: Divide the synchronized time series data into preset time windows and sliding steps; S43: Within each time window, extract core parameters from the material's microscopic state parameters and the multidimensional macroscopic performance parameters, and calculate dynamic indicators reflecting process disturbances; S44: Extract and combine time-domain statistical features, frequency-domain features, and spatiotemporal correlation features from the core parameters and the dynamic indicators to generate the high-dimensional defect feature vector.

[0013] Furthermore, in step S5, the defect identification model is a graph neural network, and its architecture is configured as follows: the cable is discretized into a series of spatial nodes, and the initial feature vector of each node contains its corresponding local material micro-state parameters and multi-dimensional macro-performance parameters; a graph structure is constructed according to the spatial adjacency relationship of the nodes, and the feature information of the nodes and their neighborhoods is aggregated using graph convolution operations; in the training loss function of the model, a material physical constraint term based on the physical laws of materials is introduced to penalize the model output prediction results that violate the physical laws.

[0014] Furthermore, the physical constraints of the material include: when the degree of crosslinking is higher than a preset threshold, the corresponding dielectric loss factor should be lower than a preset upper limit.

[0015] Furthermore, in step S6, the closed-loop feedback control interface executes the following logic: when the probability of a specific defect output by the defect identification model exceeds a preset threshold, and the root cause analysis confirms that the defect has a stable and strong correlation with the abnormality of one or more process parameters, a control command is generated; the control command is sent to the corresponding process control system through an industrial standard communication protocol, and the control command includes adjusting the set temperature of the extruder heating zone, adjusting the traction line speed, or triggering control of the auxiliary actuator.

[0016] Furthermore, the method also includes an anti-oscillation mechanism: during a preset latching time after a control command is executed, the initiation of new control commands for the same controlled process parameter is suspended.

[0017] Furthermore, the method also includes calibration and update steps: periodically acquiring standard samples of the cable and conducting offline laboratory tests to obtain benchmark values ​​for the material's microscopic state parameters and multidimensional macroscopic performance parameters; calculating the deviation between the online simulation and test values ​​and the benchmark values; using the deviation to fine-tune and calibrate the model parameters of the online simulation module for the material's microscopic state; and incrementally learning and updating the defect identification model.

[0018] Furthermore, the method is supported by an integrated materials knowledge base, which pre-stores a set of physical property parameters for various cable materials. This knowledge base is used to configure the dynamic model of the online simulation module of the material microstate, the material physical constraints of the defect identification model, and the control template of the closed-loop feedback control interface.

[0019] Furthermore, the raw data stream collected in step S1, the material micro-state parameters deduced in step S2, the multi-dimensional macroscopic performance parameters obtained in step S3, the defect identification results output in step S5, and the control commands and process parameters triggered in step S6 are all synchronously recorded and stored together to form a production holographic data chain that can be used for quality traceability and process optimization.

[0020] Furthermore, the closed-loop feedback control interface integrates a multi-objective adaptive optimization module; the multi-objective adaptive optimization module is configured to perform the following steps: Receive the current material microstate parameters from the online material microstate simulation module, the current defect probability from the defect identification model, and the current process parameter settings; The multi-objective optimization problem is constructed with multiple optimization objectives, namely minimizing the predicted defect probability, minimizing the increase in production energy consumption, and maximizing the traction line speed, and with the safe range of process parameters, the safe range of material state, and the threshold of key performance indicators as constraints. The predicted defect probability is obtained by rapidly predicting the candidate process parameter adjustment scheme by inputting it into the simplified model of the online material microstate simulation module and the defect identification model. A multi-objective evolutionary algorithm is used to solve the multi-objective optimization problem online to obtain a Pareto optimal solution set. According to a preset decision-making strategy, the final process parameter adjustment command is selected from the Pareto optimal solution set and sent to the process control system.

[0021] Furthermore, the multi-objective evolutionary algorithm is a non-dominated sorting genetic algorithm based on reference points (NSGA-III). The optimization objectives include: First objective function , is the predicted defect probability ; Second objective function , which is the increase in production energy consumption Its calculation depends on the energy consumption coefficient of each process unit obtained from the rated power and efficiency calibration of the equipment; Third objective function , is the adjusted traction line speed .

[0022] The decision-making strategy includes at least one of the following modes: In the quality-first mode, the solution with the lowest predicted defect probability is selected; In the energy efficiency priority mode, the solution with the smallest increase in production energy consumption is selected; In the capacity-first mode, the solution with the highest adjusted traction line speed is selected; in the automatic balancing mode, multiple optimization objectives are weighted and summed according to preset weights, and the solution with the highest comprehensive utility value is selected.

[0023] Furthermore, the solution process of the multi-objective adaptive optimization module is equipped with a maximum computation time threshold; when the optimization calculation reaches the maximum computation time threshold, the currently obtained optimal solution set is output to ensure the real-time performance of the control command.

[0024] Compared with the prior art, the present invention has the following beneficial effects: Traditional methods can only perform static inspections after production is completed, making it difficult to capture dynamic defects during the production process. This invention, by collecting process data in real time on the production line and extrapolating the microscopic state online based on a material dynamics model, enables the system to provide early warnings before defects manifest as macroscopic anomalies. This early warning mechanism based on process perception and state extrapolation overcomes the limitations of post-production inspection, truly achieving upstream quality control.

[0025] Traditional detection systems often have relatively independent modules with weak data correlation. This invention spatiotemporally aligns and fuses the microscopic state parameters of materials with multi-dimensional macroscopic performance parameters (apparent, electrical, thermal, mechanical, etc.) to form a high-dimensional feature vector, which is then input into a physically constrained graph neural network for analysis. This model not only possesses powerful pattern recognition capabilities but also incorporates materials science principles as constraints, giving defect identification results both data-driven and physically reliable advantages, significantly improving the accuracy and reliability of detection.

[0026] Traditional methods typically only monitor without controlling, or rely on manual intervention. This invention, through a closed-loop feedback control interface, automatically correlates defect identification results with abnormal process parameters, and drives real-time dynamic adjustments to key process parameters such as extrusion temperature and traction speed. It also introduces a multi-objective adaptive optimization mechanism, which intelligently balances multiple objectives such as quality, energy consumption, and capacity to achieve optimal overall benefits, driving the evolution of production lines from automation to intelligence.

[0027] This invention integrates a materials knowledge base, supporting rapid switching and configuration of various cable materials; through regular standard sample calibration and incremental learning mechanisms, it continuously optimizes model parameters to adapt to raw material fluctuations and equipment status changes; all hardware uses industrial-grade components, and the software architecture supports real-time communication and high-precision synchronization, ensuring stable operation and long-term reliability of the system in industrial environments.

[0028] This invention, through the collaborative innovation of process dynamic perception, micro-state inference, multi-dimensional detection fusion and closed-loop adaptive control, not only significantly improves the early defect detection rate and first-pass yield, but also realizes real-time optimization of the production process and efficient utilization of resources, providing the wire and cable industry with a full-process intelligent quality control solution from passive inspection to proactive prevention. Attached Figure Description

[0029] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained from these drawings without creative effort.

[0030] Figure 1 This is an overall flowchart of the method described in this invention.

[0031] Figure 2 This is a flowchart of the real-time acquisition process data from multiple sources according to the present invention.

[0032] Figure 3 This is a flowchart of the microscopic state deduction and macroscopic performance testing process of the present invention.

[0033] Figure 4 This is a flowchart of the defect identification and closed-loop adaptive control process of the present invention. Detailed Implementation

[0034] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of the embodiments of the invention. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0035] The following is in conjunction with the appendix Figures 1-4 The embodiments of the present invention will be described in detail below.

[0036] Example 1: This example discloses a multi-dimensional detection method for quality defects in wire and cable production. The method is applied to a continuous production line for wires and cables and includes the following steps: S1: Real-time acquisition of raw data streams reflecting the dynamics of the production process through high-precision multimodal sensor arrays deployed at key process nodes of the production line; S2: Input the raw data stream into the online material microstate deduction module, and deduce the material microstate parameters distributed along the cable length direction online based on the preset physicochemical kinetic model; S3: Through the multi-dimensional macroscopic performance testing unit, the cable semi-finished products on the production line are continuously tested online to obtain multi-dimensional macroscopic performance parameters, including appearance morphology, electrical characteristics, thermal behavior and mechanical properties. S4: Based on a unified time reference, the microscopic state parameters of the material and the multidimensional macroscopic performance parameters are spatiotemporally aligned and fused to construct a high-dimensional defect feature vector; S5: Input the high-dimensional defect feature vector into a pre-trained defect recognition model, which is a graph neural network embedded with material physics constraints, and output the defect type and probability of the cable. S6: Through the closed-loop feedback control interface, the defect type and probability, as well as the related abnormal process parameters, are fed back to the process control system of the production line in real time to trigger dynamic adjustment of key process parameters.

[0037] In a typical cross-linked polyethylene (XLPE) insulated power cable production line, a distributed optical fiber temperature sensor is first wound along the screw axis on the outer wall of the extruder barrel. This sensor has a spatial resolution of 10 mm, a temperature measurement range covering 20℃ to 400℃, and a sampling frequency set to 12 Hz. It is used to acquire the temperature field distribution of the molten polymer as it flows through each heating zone in real time. The distributed optical fiber temperature sensor uses the Raman scattering principle, achieving temperature inversion by demodulating the Stokes and anti-Stokes light intensity ratio. Its calibration curve has been calibrated in a blackbody furnace before leaving the factory, with a maximum nonlinear error not exceeding ±0.8℃.

[0038] At the extrusion die exit, a triangulation group consisting of three laser displacement sensors is installed, evenly distributed at 120° around the cable axis. The measurement reference plane is 5mm from the die exit. The measurement accuracy of a single sensor is ±1μm, and the sampling rate is synchronized to 20kHz. This is used to calculate the outer diameter D and concentricity C of the insulation layer. , , The insulation layer thickness was measured in three directions.

[0039] The traction section is equipped with a tension-velocity composite sensor, which consists of a pair of high-stiffness strain gauge tension sensing elements and an incremental photoelectric encoder. The tension measurement range is 0-500N with a resolution of 0.1N. Velocity measurement is based on a 2048-line pulse signal per encoder revolution, which, after quadrupling the frequency, achieves an equivalent resolution of 8192 PPR, corresponding to a linear velocity measurement accuracy better than ±0.01%. This sensor is installed between the front and rear traction rollers, providing real-time output of instantaneous tension. With linear velocity It is used to capture dynamic disturbances caused by line slippage or traction fluctuations.

[0040] Inside the subsequent dry cross-linking tube, a set of humidity-oxygen concentration composite probes is arranged every 50cm along the axial direction. The humidity sensing unit is based on a capacitive polymer film, and the oxygen concentration detection adopts the principle of electrochemical fuel cell. The humidity measurement range is 0-100%RH with an accuracy of ±1.5%RH, the oxygen concentration range is 0-25% with a resolution of 0.1%, and the response time is less than 10s. All probe data are collected to the central acquisition unit via RS-485 bus, with an update cycle of 100ms.

[0041] The raw data stream collected by the aforementioned high-precision multimodal sensing array is transmitted in real time to the online simulation module of the material's microstructure.

[0042] For XLPE insulation materials, the online material microstate simulation module has a built-in crosslinking degree distribution simulation unit, the core of which is a first-order Arrhenius type kinetic equation as follows: ; Where the reaction rate constant Determined by the following formula: In the formula: Degree of cross-linking, dimensionless, range of values ; Reaction time, in units ; : Reaction rate constant, in units ; Pre-exponential factor, taking values ; Activation energy of cross-linking reaction, value ; Gas constant, values ; Real-time temperature of the material, in units ; : Reaction order, values , dimensionless.

[0043] For any given moment on the production line The system uses the axial temperature sequence provided by distributed fiber optic temperature sensors. ( , (Position coordinates), combined with historical traction speed data Calculate the residence time of the material at various points within the crosslinking tube. The material enters from the inlet of the crosslinking tube (…). ) Flow to location Time required By integrating the reciprocal of the velocity, we obtain: ; In the formula: The material flows from the inlet of the cross-linking pipe to the first... The actual time required to stay at the point, in units ; : The first cross-linked tube Axial distance from the point to the entrance, in units ; Material in position The instantaneous linear velocity at a given point is obtained by spatiotemporal mapping from the measurement value of the traction speed sensor, and the unit is... .

[0044] In practical implementation, using discrete time steps Recording velocity sequences And by accumulating, an approximate integral is calculated: ,in For time step The distance moved within the body.

[0045] The fourth-order Runge-Kutta method was used to numerically integrate the differential equation, with a time step of 0.01 s and initial conditions. The final output is the crosslinking degree distribution curve along the cable length direction. .

[0046] When any oxygen concentration probe reading exceeds the 5% threshold, the system automatically introduces a correction term. The local reaction rate is multiplied by this coefficient to reflect the oxidation inhibition effect.

[0047] For production lines using polyvinyl chloride (PVC) as the sheath material, the system switches to the crystallinity gradient deduction unit (see attached figure, label 8). This unit is based on the Avrami equation as follows: ; Crystallization rate constant With cooling rate The relation is: ; in: : Crystallinity at time t, dimensionless, range of values ; Crystallization rate constant, in units ; Avrami index, values Dimensionless; : Standard value of crystallization rate constant, with various values ; Cooling rate influence coefficient, value ; Cooling rate, in units ; The infrared thermal imager has a frame rate of 32Hz and a spatial resolution of 5mm.

[0048] Cooling rate It is obtained directly by calculating the rate of temperature change at the same spatial location between consecutive frames: ; In the formula: : in position ,time Cooling rate, unit ; Infrared thermal imager at location ,time The measured surface temperature of the cable, in units of ; The sampling time interval of the infrared thermal imager ( ),unit .

[0049] The calculated Substitute the relation The crystallization rate constant can then be determined. The degree of crystallinity can then be obtained by integrating the Avrami equation. Distribution along the axial direction.

[0050] Simultaneously, the multi-dimensional macroscopic performance testing unit performs continuous online measurements on the semi-finished cable products. The appearance defect detection subsystem consists of a linear CCD camera, a coaxial ring LED light source, and an embedded image processing unit. The CCD camera is a Baslerra L4096-48gm with a pixel size of 5.5μm and a line frequency of 20kHz. It is equipped with a 50mm telecentric lens (depth of field ±2mm) to ensure that the imaging of cables with diameters of 1-50mm is free of perspective distortion. The LED light source has a center wavelength of 630nm and a peak illuminance of 5000lux. It achieves microsecond-level pulse illumination that is strictly synchronized with the camera's line frequency through PWM modulation, eliminating motion blur.

[0051] The image processing unit runs the U-Net semantic segmentation network, with a ResNet-34 backbone encoder and transposed convolutional upsampling decoder. The input image size is 4096×128 pixels, and the output is a binary mask of the same size. It can identify five types of defects: scratches (width ≥ 20 μm), bubbles (diameter ≥ 30 μm), impurities (area ≥ 500 μm²), color difference (ΔE ≥ 3.0), and abnormal surface roughness (Ra ≥ 1.6 μm). Defect density... Defined as the percentage of defective pixels per unit area, with units of defects / mm².

[0052] The electrical performance testing subsystem comprises two non-contact modules. The conductivity measurement module uses a differential eddy current probe. A 1MHz sinusoidal current is applied to the excitation coil, and the receiving coil induces a secondary magnetic field. The amplitude and phase are extracted by a lock-in amplifier and converted into conductivity values ​​via a calibration curve. Measurement range The relative error is ±0.45%. The dielectric strength testing module applies a 5kV, 100kHz AC voltage to the surface of the insulating layer, and measures the leakage current through a high-impedance buffer. Its amplitude and phase angle Used to calculate dielectric loss factor The detection sensitivity reaches 0.001, typical XLPE qualified product. .

[0053] In addition to the aforementioned infrared thermal imager, the thermal behavior detection subsystem also integrates a flash thermal diffusivity measurement device. A short-pulse (pulse width < 1ms) high-energy light source (such as a xenon lamp) is placed at the end of the cooling section to instantaneously and uniformly heat the surface of the cable insulation layer. On the opposite side of the heated surface, a high-speed infrared thermal imager (frame rate > 100Hz) records the temperature rise curve of the back side over time. According to the flash method theory, the thermal diffusivity... Due to the thickness of the insulation layer and the time required for the back temperature to reach half of its maximum value. calculate: ; The characters in the formula are defined as follows: Thermal diffusivity, unit ; The thickness of the sample (insulating layer), measured by a laser displacement sensor. The average value is provided in units of ; The time corresponding to when the temperature rise curve on the back of the sample reaches half of the maximum temperature rise, in units of... This coefficient is extremely sensitive to internal micropores (porosity > 0.5%) and interfacial debonding (bonding strength < 0.8 MPa), and deviations... This is considered abnormal.

[0054] The mechanical performance testing subsystem uses a miniature electromagnetic vibrator (resonant frequency 200Hz) to apply a 100Hz, 10μm amplitude transverse harmonic excitation to the cable, and a laser Doppler vibrometer (Polytec OFV-505) measures the response displacement. The amplitude was obtained by FFT analysis. Phase lag Mechanical loss factor It can be obtained directly by measuring the phase difference between the response signal and the excitation signal: .

[0055] If you need to obtain the energy storage modulus The absolute value of the value needs to be determined through calibration to establish the system's equivalent stiffness. The calibration method is as follows: Using known dynamic modulus A standard sample (material and geometry known) was used to replace the cable under test, and its response amplitude was measured under the same excitation conditions. Phase lag Through formula Deducing the equivalent stiffness of the system Subsequently, for the tested cable, its energy storage modulus... and loss modulus They can be calculated separately as follows: ; in: The equivalent stiffness of the system obtained through calibration using standard samples, in units of... For normal XLPE If the cross-linking is uneven or microcracks are present, this value can rise to above 0.05.

[0056] The defect feature vector construction module receives heterogeneous data streams from microscopic deduction and macroscopic detection. All sensor data are time-synchronized via hardware triggering or based on a precise network time protocol, and uniformly interpolated and resampled to a common time series of 100Hz. The system divides the time series with a window length of 1 second (i.e., 100 data points) and a sliding step size of 0.1 seconds (i.e., 10 data points). For each window, the standard deviation of the crosslinking degree distribution is extracted. Crystallinity gradient extremes (defined as) Mean dielectric loss factor Surface defect density thermal diffusivity deviation Mechanical loss factor Core parameters, etc.

[0057] Simultaneously calculate the process disturbance index: Temperature fluctuation rate ( (Number of data points within the window, fixed at 100), tension change amplitude. .

[0058] For the above 8 basic variables, their mean, variance, skewness, kurtosis, and first-order autocorrelation coefficient (5 statistics × 8 variables) are further calculated, generating a total of 40 statistical features; FFT is performed on the time series of the 8 core parameters to extract the dominant frequency amplitude, dominant frequency phase, and spectral energy (or similar 3 features) (3 features × 8 variables), generating a total of 24 frequency domain features; temperature fluctuation rate is calculated. Tension change amplitude The first and second derivatives (2 types of derivatives × 2 variables) are used to generate 4-dimensional process disturbance extension features; material one-hot codes (e.g., 4 types of materials), equipment status codes (e.g., 2 types of statuses), and normalized values ​​of equipment runtime and maintenance cycle (2 items) are integrated to generate 8-dimensional (4+2+2) material and equipment features; the mean and variance of the differences between each core parameter and the corresponding parameters of its 10 adjacent nodes (2 types of statistics × 8 variables) are calculated to generate 16-dimensional spatiotemporal correlation features. Finally, all features are concatenated, for example (40+24+4+8+16=92 dimensions), and according to the model input requirements, they can be normalized to 128 dimensions using methods such as principal component analysis (PCA) or autoencoder to form a high-dimensional defect feature vector. .

[0059] The defect identification model employs a graph neural network architecture in conjunction with physical constraint embedding units. The cable is discretized into... Number of nodes (node ​​spacing 10cm), nodes initial features Includes local microstates ( ) and macroscopic performance ( (etc.) A total of 16 dimensions.

[0060] Neighborhood Defined as 5 nodes before and 5 nodes after (11 nodes in total), edge features Including Euclidean distance And a material continuity indicator (1 for the same material, 0 for dissimilar materials). The AGGREGATE function uses a distance-weighted average: , , The network has three layers, each... for The learnable matrix has ReLU as its activation function.

[0061] Global pooling employs an attention mechanism: , The classification head is a three-layer fully connected network. The output shows the probability distribution of five types of defects: scratches, insufficient cross-linking, bubbles, abnormal crystallization, and comprehensive aging.

[0062] Physically constrained embedding units introduce consistency penalties during the training phase. For example, when extrapolating crosslinking degree... At that time, the dielectric loss is required. Otherwise, the loss function will add a term: ; in: Physical constraint loss term, dimensionless; : No. The dielectric loss factor of a node is dimensionless. : Indicator function, takes the value 1 when the condition in parentheses is true, otherwise takes the value 0; : No. The degree of cross-linking at nodes is dimensionless. Total loss ,in The standard cross-entropy is used. The model converges on the training set containing 100,000 meters of cable data, and achieves an F1-score of 0.96 on the validation set.

[0063] The closed-loop feedback control interface interacts with the PLC control system via the OPCUA communication module. The system maintains a continuously updated process parameter-defect correlation graph. When the defect identification model outputs the probability of a certain type of defect... Exceeding the first threshold (e.g.) When the defect is marked in the correlation diagram as having a stable and strong correlation with a certain process parameter (such as temperature in a certain area or traction speed) within a recent period (e.g., the absolute value of the sliding Pearson correlation coefficient is consistently >0.8), the root cause analysis submodule is triggered.

[0064] After the root cause analysis submodule, in conjunction with the state of the material model, confirms that the abnormal process parameters are a credible cause of the defect, the system generates control commands. Furthermore, the system has a control latch mechanism that suspends further control of the same parameter for at least 60 seconds after a control action is executed, to prevent oscillations.

[0065] Temperature adjustment range The basis is the minimum adjustment step size of the extruder heating zone. and process safety range The optimal adjustment range is set at 2°C to avoid material degradation caused by sudden temperature changes; the traction speed adjustment range... The basis is the production line speed stability threshold. A 5% adjustment ensures that the concentricity of the cable is not affected, while extending the residence time of the material in the cross-linking tube.

[0066] For example, if and Then, a command is sent to the extruder temperature control PLC to adjust the set temperature of the corresponding heating zone (e.g., Zone 5). At the same time, the traction speed is reduced by 5%. ).

[0067] like And with tension fluctuations If the correlation coefficient is >0.85, the mold cleaning solenoid valve is triggered (opening for 30 seconds) and the traction roller air pressure is adjusted by ±0.1MPa. All commands are transmitted through the OPCUA safety channel with a response delay of <100ms.

[0068] The materials knowledge base stores sets of physical property parameters for materials such as XLPE, PVC, LSZH, and EPR, including... The HMI (Human-Machine Interface) provides a material selection drop-down menu. After switching, the system automatically loads the corresponding simulation model, feature weights, and control templates. For example, when switching to LSZH, the crosslinking degree simulation unit is disabled, the crystallinity unit is enabled, and the defect categories are adjusted to "excessive smoke density" and "insufficient mechanical strength," etc.

[0069] The data calibration and model update module automatically extracts a 1m standard sample and sends it to the laboratory every 24 hours or after producing 100km of cable. Metallographic analysis is then used to determine the actual degree of cross-linking. DSC measurement of crystallinity Full performance test results wait.

[0070] Online simulation deviation Used to calibrate the inference model. An optimization method based on gradient descent is employed to fine-tune the parameters in the dynamic equations (such as...). and ), to make the inferred value Approaching laboratory measurements .

[0071] Specifically, constructing the loss function And calculate its relative to the parameter and The gradient, with a small learning rate (e.g. The system performs iterative updates. Each calibration involves only minor adjustments to the parameters, and the adjustment history is recorded. If multiple consecutive calibrations require significant parameter adjustments in the same direction, an alarm for sensor or raw material abnormalities will be triggered.

[0072] Incremental learning employs the Elastic Weight Fixation (EWC) algorithm, setting the learning rate to [missing information]. After each calibration, iterate 100 times, only fine-tuning the weights of the last two layers of the classification head, with the weight solidification coefficient set to 0.9, to retain the ability to extract features from the bottom layer.

[0073] In practical implementation, the method of this invention was applied to a production line for 6 / 10kV XLPE insulated cables. The production line speed was 15m / min, the extrusion temperature was set at 220℃, the cross-linking tube temperature was 390℃, and the oxygen concentration under nitrogen protection was <1%. The system ran continuously for 72 hours, and a total of 130km of cables were tested.

[0074] Three latent defects were identified during the period: Crosslinking degree distribution at 12 hours (Threshold 0.05), the simulation showed that the Zone4 temperature was too low (actual measured 215℃ vs. set 220℃). The system automatically increased the temperature by 2℃ and reduced the speed to 14.25m / min. After 20 minutes... It dropped to 0.041; 45 hours It suddenly increased to 0.058, combined with tension fluctuations. (Normal <20N) indicates wear of the traction roller bearing, triggering a maintenance alarm; Dielectric loss factor at 68 hours Abnormal fluctuations occurred (rising to 0.0045) and the mechanical loss factor... The synchronous increase in crosslinking was not detected by the microscopic simulation module, but the defect identification model determined that the product was prone to "comprehensive aging" based on multi-source information. It was found that the raw material batch contained trace amounts of moisture. The system recorded and isolated the product segment.

[0075] The final product, after third-party testing, has a breakdown field strength ≥20kV / mm. All standards were met.

[0076] Comparative Example 1: On the same production line, the system of this invention was shut down, relying solely on final inspection (spark test plus visual inspection). Within the same 72 hours, the final inspection removed 1.2km of products with obvious defects (mainly surface scratches and out-of-tolerance outer diameter), but after delivery, the customer reported 3 batches of early breakdown failures. Dissection revealed uneven cross-linking. The production logs showed that there were large temperature fluctuations at the time, but no intervention was taken.

[0077] Example 2: This example is basically the same as Example 1, except that in this example, the production line is switched to PVC sheathed control cable production, with a wire diameter of 8mm and a production line speed of 25m / min. The system loads the PVC parameter set, and the crystallinity deduction unit is activated. At the 8th hour, the infrared thermal imager detected an abnormal temperature gradient in the cooling section. (vs. normal 8℃ / s), extrapolating crystallinity (Target 0.75), the system automatically reduces the cooling fan speed by 10%, and after 15 minutes... It rebounded to 0.73. Appearance testing simultaneously revealed a decrease in surface gloss (positively correlated with crystallinity), verifying the accuracy of the inference.

[0078] Comparative Example 2: Microscopic deduction module is turned off, and only macroscopic detection is used. When crystallinity is insufficient, apparent defects have not yet appeared. The system failed to issue a warning, resulting in subsequent batches failing the bending performance test (elongation at break <120% vs. standard ≥150%).

[0079] The following table summarizes the comparison of key indicators, as shown in Table 1: Table 1:

[0080] The method described in this invention utilizes all industrial-grade components at the hardware level: the distributed fiber optic temperature sensor is the OmniSens DTSX series, the laser displacement sensor is the Keyence LK-G5000, the tension-velocity sensor is the HBMSLB700, and the OPCUA server is a built-in module of the Siemens S7-1500. The software system runs on a Linux real-time kernel (with the PREEMPT_RT patch), with a task scheduling cycle of 1ms to ensure data synchronization accuracy. All algorithm modules are encapsulated in Docker containers, and inter-node communication is achieved through ROS2 (Foxy version), with message latency <5ms.

[0081] The installation of high-precision multimodal sensor arrays must meet mechanical tolerance requirements: the coaxiality of the laser displacement sensor group axis and the cable centerline ≤ 0.05 mm, the helix angle of the distributed optical fiber along the extruder barrel ≤ 5°, and the horizontality of the tension sensor mounting plane ≤ 0.1°. The calibration process includes: temperature sensors are calibrated at three points (100℃, 250℃, 400℃), displacement sensors are calibrated using standard gauge blocks (10.000±0.001 mm), and electrical modules are verified using standard resistors (1mΩ–100Ω) and capacitors (10pF–1nF).

[0082] This invention achieves a leap from "post-inspection" to "process immunity" in quality management by deeply integrating process sensing, microscopic simulation, multi-dimensional detection, and closed-loop control. Its technical solution possesses complete engineering implementation details; all parameters, algorithms, and interfaces are clearly defined and reproducible, allowing those skilled in the art to implement it without creative effort.

[0083] Example 3: This embodiment, based on the system architecture and data flow described in Embodiment 1, elaborates in detail on the construction, integration, and application process of the multi-objective adaptive optimization module. The multi-objective adaptive optimization module operates as the core decision-making unit of the closed-loop feedback control interface.

[0084] Construction and integration of multi-objective adaptive optimization modules: The module's input consists of three parts: Real-time status input: Current material microstate parameters (such as crosslinking degree distribution curve) from the online material microstate simulation module. ), and the current defect probability from the defect identification model. .

[0085] Process parameter input: Current process parameter setpoints from a high-precision multimodal sensor array and process control system, including the temperatures of each heating zone in the extruder. ( ), Traction line speed Cooling fan power wait.

[0086] Optimization goals and constraints configuration: Optimization goal weights, process safety boundaries, and key quality indicator thresholds are preset through the HMI human-machine interface or issued by the upper-level production management system.

[0087] The module outputs a set of Pareto optimal process parameter adjustment schemes and their corresponding predicted performance indicators, which are then selected by the decision-making system and sent to the process control system for execution.

[0088] Mathematical modeling of multi-objective optimization problems: The process adjustment decision problem is formalized as a three-objective optimization problem. The decision variables are the adjustment vectors of process parameters. : ; In the formula: Extruder No. Temperature adjustment amount for the heating zone, in °C; : Adjustment amount of traction line speed, expressed as a percentage (%); Cooling fan power adjustment amount, expressed as a percentage (%).

[0089] The three objective functions are as follows: Quality objective (minimize defect risk): ; Indicates implementation of the adjustment plan Then, the system predicts the probability of defects.

[0090] This is the output of a fast prediction model. The model takes the current state and adjustment plan as input. For input, quickly calculate using the following steps: Process parameter mapping: Calculate the adjusted virtual process parameters: , , .

[0091] Microscopic state deduction: , Substituting the simplified model (such as using linearized kinetic equations) into the online modeling module for material microstates allows for rapid estimation of the adjusted material microstate parameters. , .

[0092] Defect probability prediction: the estimated , The predicted changes in other macroscopic performance parameters are input into the defect identification model (or a lightweight version thereof), and the predicted defect probability is output. .

[0093] The model is built based on offline historical data training, which ensures the real-time nature of online evaluation (single evaluation <50ms).

[0094] Energy consumption target (minimize energy consumption increment): ; Indicates implementation of the adjustment plan The resulting increase in production energy consumption per unit time, expressed in kWh / km.

[0095] Energy consumption increment Calculated from the equipment physical model: ; In the formula: : No. The effective temperature control energy consumption coefficient of the heating zone, expressed in kWh / (km·℃). Its physical meaning is the energy consumption coefficient for the heating zone per unit length of cable. The energy required to raise the temperature of a zone by 1°C. This coefficient is obtained through calibration using the rated power on the equipment nameplate, heating efficiency, and thermodynamic model. For example, for the heating power of a certain zone... heating efficiency Production line speed ,but .

[0096] Traction speed energy consumption coefficient, unit kWh / (km·%). It represents the change in traction energy consumption per unit length caused by a 1% change in speed, and is determined by the traction motor power-speed characteristic curve.

[0097] Cooling power energy consumption coefficient, in kWh / (km·%). Represents the change in cooling energy consumption per unit length resulting from a 1% change in cooling power.

[0098] Capacity target (maximizing production efficiency): ; That is, the adjusted traction line speed The unit is m / min. The objective is to maximize... .

[0099] The constraints are as follows: Process safety constraints: ; ; Boundary values ​​are derived from the equipment manual and process specifications.

[0100] Material condition safety constraints: ; Ensure that the material does not degrade or deteriorate in performance.

[0101] Key quality indicator constraints: ; Ensure that the predicted macroeconomic performance remains within acceptable limits.

[0102] Online solution and decision-making based on NSGA-III algorithm: To effectively handle the three objectives and obtain a uniformly distributed Pareto front, a non-dominated sorting genetic algorithm based on reference points (NSGA-III) is adopted.

[0103] Algorithm initialization: Population size This matches the number of reference points provided.

[0104] Decision variable range: Determine each based on the constraints. upper and lower boundaries .

[0105] Randomly generate the initial population Each individual represents a complete adjustment plan. .

[0106] Iterative optimization (the first) generation): Evaluation: On the population For each individual, calculate three objective function values. .

[0107] Non-dominated sorting: sorting the population Divided into multiple non-dominated layers ( ).

[0108] Reference point association: Adaptively normalize the target value and associate the individual with a predefined reference point.

[0109] niche selection: Based on the niche count at the reference point, select from the last acceptable dominance layer. Individuals are selected to maintain population diversity.

[0110] Genetic manipulation: The selected individuals undergo simulated binary crossover (SBX) and polynomial mutation to generate a progeny population. Cross-distribution index Variation distribution index .

[0111] Merge and Selection: Merge and A new round of non-dominated sorting and reference point association selection will be carried out to form the next generation of population. .

[0112] Termination and Output: Set the maximum number of iterations Or calculate the upper limit of time. After the termination condition is met, output the first non-dominated layer in the final population. As the Pareto optimal solution set .

[0113] Online decision-making: The system makes decisions based on real-time production strategies. Choose a final solution from the options. : Quality-first mode: Select The smallest solution.

[0114] Energy Efficiency Priority Mode: Select The smallest solution.

[0115] Capacity-first model: Select The largest solution.

[0116] Automatic balancing mode (default): Calculates the overall utility value of each solution. ,choose The largest solution. Weights It can be set by the operator or optimized by back-engineering from historical benefit data.

[0117] In practical use, it is linked with the scenario in Example 1: In the 12-hour scenario described in Example 1, the system detected the standard deviation of the crosslinking degree distribution. (Threshold 0.05), the comprehensive defect probability output by the defect identification model. At this point, the system enters a multi-objective optimization and control process.

[0118] Current parameters: , , .

[0119] Optimization computation: The multi-objective adaptive optimization module completes 100 iterations within 45 seconds, obtaining a Pareto optimal solution set containing 28 non-dominated solutions. .

[0120] Decision-making and execution: Running in "Automatic Balancing Mode", the weights are preset to 1. (quality), (Energy consumption) (Production capacity). After calculation, the optimal solution is selected as: , (Auxiliary adjustment to reduce temperature gradient) , (Enhance cooling to stabilize crystallization).

[0121] Prediction Performance: The fast prediction model evaluates this approach. , , .

[0122] Actual Execution and Results: The adjustment instruction was issued. Monitoring 20 minutes later showed: Actual The actual defect probability drops to 0.046. When it drops to 0.06, energy consumption per unit length increases. Actual average speed It closely matches the prediction.

[0123] Comparative analysis with traditional single-target regulation: Under the same perturbation, if the traditional approach is adopted with quality as the sole objective (minimization)... The optimal solution calculated by the system is: , , The prediction result is: , , .

[0124] The 72-hour long-term simulation data of the two strategies are compared and summarized in Table 2 below: Table 2:

[0125] The wire and cable industry has long faced the challenge of balancing quality, cost, and delivery time. Traditional automated systems often prioritize optimizing a single quality indicator, leading to high energy consumption and limited production capacity. This invention, through the construction of a multi-objective adaptive optimization module, deeply integrates materials science models, equipment physical models, and intelligent optimization algorithms, achieving: From singular focus on quality to comprehensive optimization: Proactively seeking the optimal solution between energy consumption and output while ensuring controllable quality risks. From static rules to dynamic adaptation: The optimization target weights can be dynamically adjusted according to market demand and energy prices, resulting in agile responses. From manual experience to model-based decision-making: Reliance on process expert experience is reduced, improving the scientific rigor and consistency of decision-making. This marks a crucial leap from automation to intelligence in wire and cable production quality control, providing a practical and feasible technical path for the industry to implement green and intelligent manufacturing.

[0126] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0127] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. It should be noted that any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A multi-dimensional detection method for quality defects in wire and cable manufacturing, characterized in that, The method is applied to a continuous production line for wires and cables, and includes the following steps: S1: Real-time acquisition of raw data streams reflecting the dynamics of the production process through high-precision multimodal sensor arrays deployed at key process nodes of the production line; S2: Input the raw data stream into the online material microstate deduction module, and deduce the material microstate parameters distributed along the cable length direction online based on the preset physicochemical kinetic model; S3: Through the multi-dimensional macroscopic performance testing unit, the cable semi-finished products on the production line are continuously tested online to obtain multi-dimensional macroscopic performance parameters, including appearance morphology, electrical characteristics, thermal behavior and mechanical properties. S4: Based on a unified time reference, the microscopic state parameters of the material and the multidimensional macroscopic performance parameters are spatiotemporally aligned and fused to construct a high-dimensional defect feature vector; S5: Input the high-dimensional defect feature vector into a pre-trained defect recognition model, which is a graph neural network embedded with material physics constraints, and output the defect type and probability of the cable. S6: Through the closed-loop feedback control interface, the defect type and probability, as well as the related abnormal process parameters, are fed back to the process control system of the production line in real time to trigger dynamic adjustment of key process parameters.

2. The multi-dimensional detection method for quality defects in wire and cable manufacturing according to claim 1, characterized in that, In step S1, the high-precision multimodal sensing array includes at least: a distributed fiber optic temperature sensor deployed at the extruder or cross-linking pipe to measure the material temperature field; a laser displacement sensor group deployed at the extrusion die outlet to measure the insulation layer size and concentricity; a tension-velocity composite sensor deployed in the traction section to measure instantaneous tension and linear velocity; and a composite probe deployed inside the cross-linking pipe to measure ambient humidity and oxygen concentration.

3. The multi-dimensional detection method for quality defects in wire and cable manufacturing according to claim 2, characterized in that, In step S2, the online material microstate deduction module calls the corresponding physicochemical kinetic model to perform deduction based on the type of cable material currently being produced on the production line. When the material is cross-linked polyethylene, the cross-linking kinetic model based on the Arrhenius equation is called, and the cross-linking degree distribution of the cable insulation layer along the length direction is deduced based on the temperature field measured by the distributed optical fiber temperature sensor and the historical linear velocity data measured by the tension-velocity composite sensor. When the material is polyvinyl chloride, the crystallization kinetic model based on the Avrami equation is called, and the crystallinity distribution of the cable sheath layer along the length direction is deduced based on the cooling rate measured by the thermal behavior detection subsystem.

4. The multi-dimensional detection method for quality defects in wire and cable manufacturing according to claim 1, characterized in that, In step S3, the multi-dimensional macroscopic performance testing unit includes: an appearance defect detection subsystem, which identifies and quantifies scratches, bubbles, impurities, and color difference defects on the cable surface based on a linear CCD camera and machine vision algorithms; an electrical performance testing subsystem, which measures the conductor conductivity and dielectric loss factor of the cable based on a non-contact eddy current probe and a dielectric strength testing module; a thermal behavior testing subsystem, which measures the temperature field on the cable surface and the thermal diffusivity of the insulation layer based on an infrared thermal imager and a flash thermal diffusivity measuring device; and a mechanical performance testing subsystem, which measures the mechanical loss factor or dynamic modulus of the cable based on an electromagnetic excitation and laser vibration measuring device.

5. The multi-dimensional detection method for quality defects in wire and cable manufacturing according to claim 1, characterized in that, Step S4 specifically includes: S41: Synchronize the data streams output by the high-precision multimodal sensing array, the online material microstate inference module, and the multidimensional macroscopic performance detection unit to a unified time reference; S42: Divide the synchronized time series data into preset time windows and sliding steps; S43: Within each time window, extract core parameters from the material's microscopic state parameters and the multidimensional macroscopic performance parameters, and calculate dynamic indicators reflecting process disturbances; S44: Extract and combine time-domain statistical features, frequency-domain features, and spatiotemporal correlation features from the core parameters and the dynamic indicators to generate the high-dimensional defect feature vector.

6. The multi-dimensional detection method for quality defects in wire and cable manufacturing according to claim 1, characterized in that, In step S5, the defect identification model is a graph neural network, and its architecture is configured as follows: the cable is discretized into a series of spatial nodes, and the initial feature vector of each node contains its corresponding local material micro-state parameters and multi-dimensional macro-performance parameters; a graph structure is constructed according to the spatial adjacency relationship of the nodes, and the feature information of the nodes and their neighborhoods is aggregated using graph convolution operations; in the training loss function of the model, a material physical constraint term based on the physical laws of materials is introduced to penalize the model output prediction results that violate the physical laws.

7. The multi-dimensional detection method for quality defects in wire and cable manufacturing according to claim 6, characterized in that, The physical constraints on the material include: when the degree of crosslinking is higher than a preset threshold, the corresponding dielectric loss factor should be lower than a preset upper limit.

8. The multi-dimensional detection method for quality defects in wire and cable manufacturing according to claim 1, characterized in that, In step S6, the closed-loop feedback control interface executes the following logic: when the probability of a specific defect output by the defect identification model exceeds a preset threshold, and the root cause analysis confirms that the defect has a stable and strong correlation with the abnormality of one or more process parameters, a control command is generated; the control command is sent to the corresponding process control system through an industrial standard communication protocol, and the control command includes adjusting the set temperature of the extruder heating zone, adjusting the traction line speed, or triggering control of the auxiliary actuator.

9. A multi-dimensional detection method for quality defects in wire and cable manufacturing according to claim 8, characterized in that, The method also includes an anti-oscillation mechanism: during a preset latching time after a control command is executed, new control commands for the same controlled process parameter are suspended.

10. A multi-dimensional detection method for quality defects in wire and cable manufacturing according to claim 1, characterized in that, The method further includes calibration and update steps: periodically acquiring standard samples of the cable and conducting offline laboratory tests to obtain benchmark values ​​for the material's microscopic state parameters and multi-dimensional macroscopic performance parameters; calculating the deviation between the online simulation and test values ​​and the benchmark values; using the deviation to fine-tune and calibrate the model parameters of the online simulation module for the material's microscopic state; and incrementally learning and updating the defect identification model.

Images