Intelligent monitoring method and system for cable rubber wire production

Abstract

The application provides a cable rubber wire production intelligent monitoring method and system, and belongs to the technical field of intelligent monitoring of wire and cable manufacturing process. The method comprises the following steps: firstly, tension, wire diameter and die temperature sensors are deployed in the extrusion link to collect and calibrate sequence flow data in real time. The method uses a time window to extract tension abnormality index and material stress concentration characteristic value, combines temperature drift correction to calculate a correlation coefficient to identify a potential abnormal fluctuation interval; then, the abnormal data are injected into a historical fault knowledge graph to determine a risk level and a root cause type; a parameter coupling constraint matrix is established based on an extrusion process physical model to generate risk distribution characteristic data; when the risk reaches an intervention condition, a multi-parameter collaborative optimization is used to generate a preliminary intervention signal, and finally a closed-loop monitoring feedback is formed. The method can realize real-time discovery of abnormalities, accurate assessment of rupture risk and dynamic adjustment of production parameters, and significantly improves the stability of cable rubber wire production and product quality.

Description

Technical Field

[0001] This invention belongs to the field of intelligent monitoring technology for the manufacturing process of wires and cables, specifically relating to intelligent monitoring methods and systems for the production of cable rubber wires. Background Technology

[0002] In the extrusion molding process of cable rubber wire, wire diameter accuracy, tension stability, and die temperature control are the core elements determining product quality and production continuity. The extrusion process typically involves three coupled physical processes: screw plasticization, die forming, and traction stretching. Fluctuations in any of these parameters can cause quality problems such as wire diameter deviation, stress concentration, surface defects, and even wire breakage. Currently, the industry mainly uses two methods to monitor the extrusion process: one is to install sensors on the extruder, traction machine, and tension controller, setting upper and lower limits for parameters such as tension, wire diameter, and temperature, triggering an audible and visual alarm when a parameter exceeds the threshold; the other is to manually or semi-automatically measure the finished wire diameter at regular intervals and generate statistical control charts for offline analysis. However, these methods have revealed significant technical shortcomings in practical applications.

[0003] First, existing monitoring methods generally treat tension, wire diameter, and temperature as independent parameters for isolated analysis, ignoring the strong nonlinear coupling relationships between them. For example, an increase in die temperature can lead to a decrease in melt viscosity, which in turn causes a decrease in tension and fluctuations in wire diameter. Single-parameter threshold alarms are prone to false alarms or missed alarms in such multivariate coupled scenarios, making it difficult to identify complex anomalies caused by multiple factors. Second, when the system alarms, operators usually need to review multiple data curves and combine them with process experience to determine the risk level and root cause of the anomaly. The diagnostic process is time-consuming and highly subjective, failing to meet the real-time response requirements of high-speed continuous production. Third, after detecting wire diameter deviations or tension anomalies, the common practice is for operators to adjust the traction speed or screw speed individually based on experience. Due to the lack of a quantitative description of the interrelationships between various parameters, this trial-and-error adjustment of a single parameter is prone to triggering new fluctuations, resulting in long adjustment cycles and high scrap rates. In addition, most existing systems stop at the "detection-alarm" stage, and subsequent adjustment decisions and execution still rely on manual intervention, failing to form an automated feedback closed-loop control. This results in prolonged abnormal production states and difficulty in ensuring product quality consistency.

[0004] Therefore, there is an urgent need to provide an intelligent monitoring method and system that can perceive the multi-parameter coupling characteristics of the extrusion process in real time, automatically diagnose the root causes and risk levels of abnormalities, and perform collaborative optimization adjustments to improve the quality stability and automation level of cable rubber wire production. Summary of the Invention

[0005] This invention provides an intelligent monitoring method and system for the production of cable rubber wires, aiming to solve the technical problems mentioned in the background art.

[0006] In a first aspect, the present invention provides an intelligent monitoring method for the production of cable rubber wires, comprising: Step S1: Collect tension, wire diameter, and die temperature data during the extrusion process, and generate calibration sequence stream data after filtering and alignment; Step S2: Extract the tension anomaly index and material stress concentration feature value from the calibration sequence stream data in segments, calculate the correlation coefficient after temperature drift correction, and mark the potential abnormal fluctuation range when the correlation coefficient exceeds the preset correlation threshold and the die temperature is within the process allowable range. Step S3: Concatenate the tension anomaly index, material stress concentration characteristic value, die temperature and current working condition parameters of the potential abnormal fluctuation range into a query feature vector, inject it into the pre-constructed historical fault knowledge graph, and determine the risk level and root cause type of the potential abnormal fluctuation range; Step S4: Establish a parameter coupling constraint matrix based on the physical model of the extrusion process, and generate risk distribution characteristic data according to the parameter coupling constraint matrix, the risk level and the root cause type; Step S5: If the risk distribution characteristic data reaches the intervention trigger condition, multi-parameter collaborative optimization is performed with wire diameter stability as the objective function, tension safety range as the hard constraint, and the parameter coupling constraint matrix as the dynamic equality constraint to generate a collaborative adjustment vector and generate a preliminary intervention signal. Step S6: After performing dual verification of the device adjustment parameter range and device compatibility rules on the preliminary intervention signal, the device is adjusted, and the process returns to step S1 to form a closed-loop monitoring and adjustment mechanism.

[0007] Furthermore, step S1 specifically includes: High-frequency synchronous sampling is performed by high-precision tension sensors, laser diameter gauges and temperature sensors deployed in the extrusion process to obtain tension data, wire diameter data and die temperature data. The tension data, wire diameter data, and die temperature data are aligned by time series so that the data are arranged at equal intervals on the same time axis. The aligned data is denoised using Kalman filtering and outliers are removed to obtain the calibration sequence stream data.

[0008] Furthermore, step S2 specifically includes: An adaptive time window is used to segment the calibration sequence stream data, and the peak position, half-peak width, and right-tail integral of the probability density distribution of the tension change rate are extracted as the tension anomaly index. By combining the second-order difference of the wire diameter and the peak-valley difference, the characteristic value of material stress concentration is obtained; The correlation coefficients of the tension anomaly index and the material stress concentration characteristic value are calculated after temperature drift correction. If their absolute values ​​exceed the preset correlation threshold and the die temperature is within the process allowable range, they are marked as potential abnormal fluctuation ranges.

[0009] Furthermore, the extraction of the peak position, half-peak width, and right-tail integral of the probability density distribution of the tension change rate as a tension anomaly index specifically includes: The tension change rate sequence is obtained by calculating the first difference of the tension data within each time window. The probability density distribution of the tension change rate sequence is fitted by kernel density estimation. The change rate value corresponding to the maximum value of the probability density distribution is extracted as the peak position. The distribution width when the probability density distribution drops to half of the maximum value is extracted as the half-peak width. The cumulative probability of the change rate of the probability density distribution being greater than the preset abnormal change rate threshold is extracted as the right-tail integral. The peak position, half-peak width, and right-tail integral are used as the three components of the tension anomaly index.

[0010] Furthermore, the specific methods for obtaining material stress concentration characteristic values ​​by fusing the second-order difference of the line diameter and the peak-valley difference include: The strain rate mutation index is obtained by calculating the central second difference of the wire diameter data within each time window and taking the mean of the absolute values. The difference between the local maxima and local minima of the wire diameter within the time window is extracted to obtain the geometric defect magnitude index; The ratio of the geometric defect amplitude index to the strain rate mutation index is calculated to obtain the defect morphology discrimination index; The stress concentration characteristic value of the material is generated by weighted summation of the strain rate mutation index, geometric defect amplitude index, and defect morphology discrimination index.

[0011] Furthermore, step S3 specifically includes: The components of the tension anomaly index vector in the potential abnormal fluctuation range, the material stress concentration characteristic value, the average die temperature, and the screw speed setting, traction speed setting, and tension setting value under the current working condition are concatenated into a query feature vector and then normalized. The query feature vector is injected as a virtual query node into a pre-built historical fault knowledge graph. The historical fault knowledge graph includes fault case nodes, fault mode nodes, root cause type nodes, and risk level nodes. The edge types include instantiation relationships, attribution relationships, level mapping relationships, and parameter passing relationships. Multi-layer neighbor aggregation is performed using a graph attention network. The attention coefficient of the virtual query node to each fault case node is calculated, and a predetermined number of fault case nodes with the highest attention coefficients are selected as a set of similar cases. The risk level of each case in the set of similar cases is determined by weighted voting based on the normalized weight of the attention coefficient; the root cause type is determined by summing the attention weights of each root cause type node in groups and taking the maximum value.

[0012] Furthermore, step S4 specifically includes: Establish a screw plasticizing model, a die forming model, and a traction stretching model. The screw plasticizing model calculates the melt mass flow rate and melt temperature based on the screw speed. The die forming model calculates the die extrusion speed and extruded wire diameter based on the melt parameters. The traction stretching model calculates the finished wire diameter and actual tension based on the die parameters and traction speed. Screw speed, traction speed and tension setpoints are selected as state variables, finished wire diameter and actual tension are selected as observation variables. At the current steady-state operating point, the implicit equations of the three-layer model are linearized by Taylor expansion, and the Jacobian matrix is ​​extracted as the parameter coupling constraint matrix. Fourier transform is performed on the wire diameter data in the calibration sequence stream data to extract the dominant frequency component as the wire diameter change frequency data; statistical distribution fitting is performed on the tension anomaly index sequence in the calibration sequence stream data to obtain tension anomaly distribution data; an initial evaluation matrix is ​​constructed by aligning with the time axis; dynamic feature weights are assigned to the wire diameter dimension and tension dimension according to the ratio of the absolute value of the influence of each state variable on the observed variable in the parameter coupling constraint matrix, and weighted fusion is used to generate a comprehensive anomaly evaluation matrix; The comprehensive anomaly assessment matrix is ​​graded and labeled according to the risk level, the corresponding root cause type code is embedded, and the physical position coordinates of the cable in the extrusion process corresponding to each time window are generated according to the integral mapping of traction speed and time.

[0013] Furthermore, step S5 specifically includes: If the risk level in the risk distribution feature data reaches or exceeds the preset intervention level threshold, it is determined that the intervention trigger condition has been met. The objective function is to minimize the square mean of the difference between the measured wire diameter and the target wire diameter. The hard constraint is that the actual tension value is within the preset tension safety range. The soft constraint is the linear equation relationship between the increment of the observed variable and the increment of the state variable established by the parameter coupling constraint matrix. The boundary constraint is the equipment capability range of each adjustment amount. A multi-parameter collaborative optimization problem is constructed. The multi-parameter collaborative optimization problem is solved using a sequential quadratic programming algorithm to generate a collaborative adjustment vector that includes screw speed adjustment, traction speed adjustment, and tension setpoint adjustment. The coordinated adjustment vector, abnormal location coordinates, risk level, and root cause type are fused to generate a preliminary intervention signal.

[0014] Furthermore, step S6 specifically includes: The coordinated adjustment vector in the preliminary intervention signal is obtained, and the screw speed adjustment, traction speed adjustment and tension setpoint adjustment are compared with the adjustment parameter range of the corresponding equipment. The adjustment amount that exceeds the boundary is truncated to the boundary value to obtain the first verification result. The adjustment amounts in the first verification result are combined into a parameter adjustment set. It is then checked whether the set satisfies the speed response coupling constraint between the tension controller and the traction motor driver, the temperature coupling constraint between the screw speed adjustment and the melt temperature change, and the motor acceleration and deceleration capability constraint. If not satisfied, the minimum correction amount is solved by the pseudo-inverse of the parameter coupling constraint matrix to perform iterative correction and obtain the final adjustment parameters. The execution sequence is determined based on the risk level, and the final adjustment parameters are written into the corresponding control registers of the extruder frequency converter, traction motor driver and tension controller via industrial fieldbus to execute equipment adjustment; After the parameters are updated, return to step S1 to collect data again, calculate the deviation between the actual feedback value and the set value in the final adjustment parameters, and repeat steps S1 to S6 to form a closed-loop monitoring and adjustment mechanism. When the tension fluctuation amplitude and the wire diameter fluctuation amplitude are both lower than the corresponding preset stable threshold and no new abnormal fluctuation range is generated for several consecutive cycles, it is determined that a stable production state has been obtained.

[0015] Secondly, the present invention provides an intelligent monitoring system for the production of cable rubber wires, comprising: The data acquisition module collects tension, wire diameter, and die temperature data during the extrusion process, and generates calibration sequence stream data after filtering and alignment. The potential anomaly identification module is used to extract the tension anomaly index and material stress concentration characteristic value from the calibration sequence stream data in segments, calculate the correlation coefficient after temperature drift correction, and mark the potential anomaly fluctuation range when the correlation coefficient exceeds the preset correlation threshold and the die temperature is within the process allowable range. The knowledge graph diagnostic module is used to concatenate the tension anomaly index, material stress concentration characteristic value, die temperature and current working condition parameters of the potential abnormal fluctuation range into a query feature vector, inject it into the pre-constructed historical fault knowledge graph, and determine the risk level and root cause type of the potential abnormal fluctuation range. The risk distribution generation module is used to establish a parameter coupling constraint matrix based on the physical model of the extrusion process, and generate risk distribution feature data according to the parameter coupling constraint matrix, the risk level and the root cause type. The collaborative optimization decision module is used to perform multi-parameter collaborative optimization when the risk distribution characteristic data reaches the intervention trigger condition, with wire diameter stability as the objective function, tension safety range as the hard constraint, and the parameter coupling constraint matrix as the dynamic equation constraint, to generate a collaborative adjustment vector and generate a preliminary intervention signal. The verification and execution module is used to perform dual verification of the device adjustment parameter range and device compatibility rules on the preliminary intervention signal, then execute the device adjustment and return to trigger the data acquisition module to form a closed-loop monitoring and adjustment mechanism.

[0016] This invention offers the following advantages: First, by extracting the tension anomaly index based on probability density morphology and the material stress concentration feature value that integrates second-order difference and peak-valley difference, combined with the correlation coefficient after temperature drift correction, it can accurately identify potential abnormal fluctuation ranges, achieving a leap from traditional "post-event alarm" to "in-event prediction," significantly improving the timeliness and accuracy of anomaly detection. Second, by introducing a historical fault knowledge graph and graph attention network, it performs structured matching and weighted reasoning between the current anomaly feature vector and historical cases, automatically outputting risk level and root cause type. The diagnostic process requires no human experience intervention and has the advantages of strong interpretability and dynamic scalability. Third, based on the process physical model of screw plasticizing, die forming, and traction stretching, a parameter coupling constraint matrix is ​​established linearly, quantifying the sensitivity relationship between state variables and observed variables, and generating risk distribution feature data containing frequency domain, spatial domain, and dynamic weights, providing a physical constraint basis for collaborative optimization. Fourth, the parameter adjustment problem is modeled as a multi-parameter optimization problem with wire diameter stability as the objective function, tension safety range as the hard constraint, and coupling matrix as the dynamic equation constraint. A sequential quadratic programming approach is used to solve this problem, enabling a one-time output of a coordinated adjustment vector for screw speed, traction speed, and tension setpoints. This avoids the repeated fluctuations caused by traditional trial-and-error adjustments, significantly shortening adjustment time and reducing scrap rate. Fifth, through a dual mechanism of single-device boundary truncation and multi-constraint compatibility verification, combined with pseudo-inverse iterative correction of the coupling matrix, the safety and feasibility of adjustment commands are ensured. Finally, a closed-loop monitoring and adjustment mechanism is formed, automatically determining a stable production state when tension and wire diameter fluctuations are below the stability threshold and no new abnormal ranges are observed. This invention can capture minute anomalies in the extrusion process in real time, assess rupture risks in advance, and identify root causes. It achieves closed-loop intelligent management of anomaly detection and production control, significantly improving the stability and product quality of cable rubber wire production, while reducing material scrap and production downtime, improving production efficiency and safety. It possesses significant engineering application value and promotion potential. Attached Figure Description

[0017] Figure 1 A flowchart of the intelligent monitoring method for cable rubber wire production provided by the present invention; Figure 2 Comparison of tension fluctuation effects of the intelligent monitoring method for cable rubber wire production provided by the present invention; Figure 3 Comparison of wire diameter fluctuation effects of the intelligent monitoring method for cable rubber wire production provided by the present invention; Figure 4 The structural block diagram of the intelligent monitoring system for cable rubber wire production provided by the present invention. Detailed Implementation

[0018] To further understand the content of this invention, a detailed description of the invention is provided in conjunction with the accompanying drawings and embodiments. The specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention. It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0019] This invention is specifically applied to continuous extrusion production lines for rubber-insulated cables and rubber-sheathed wires, and is particularly suitable for online quality monitoring and automatic adjustment scenarios where strict requirements are placed on wire diameter accuracy, tension stability, and die temperature.

[0020] Firstly, such as Figure 1 This embodiment provides an intelligent monitoring method for the production of cable rubber wires, including: Step S1: Collect tension, wire diameter and die temperature data during the extrusion process, and generate calibration sequence stream data after filtering and alignment.

[0021] Specifically, step S1 includes: high-frequency synchronous sampling using a high-precision tension sensor, laser diameter gauge, and temperature sensor deployed in the extrusion process to acquire tension data, wire diameter data, and die temperature data; time-series alignment of the tension data, wire diameter data, and die temperature data to ensure that each data point is equidistant on the same time axis; and noise reduction processing of the aligned data using Kalman filtering to remove outliers, thereby obtaining calibration sequence stream data.

[0022] Specifically, step S1 first involves acquiring and preprocessing multi-source sensor data from the extrusion process. A high-precision tension sensor, a laser diameter gauge, and a temperature sensor are installed near the extruder die. The sampling frequency of all three is set to no less than 100 Hz to ensure real-time capture of tension fluctuations, wire diameter changes, and die temperature drift. The tension sensor uses a piezoelectric or strain gauge structure to directly measure the traction tension on the rubber wire; the laser diameter gauge acquires the wire diameter value through high-speed laser scanning; and the temperature sensor is embedded in the die's metal wall to measure the temperature at the melt outlet. These three sensors perform high-frequency synchronous sampling at the same time reference, outputting the original tension data sequence, the original wire diameter data sequence, and the original die temperature data sequence, respectively. Due to differences in the physical installation positions of the sensors, different signal transmission path lengths, and slight offsets in their respective sampling clocks, the three data sequences are not strictly aligned on the time axis. For example, the tension signal may lag behind the wire diameter signal by tens of milliseconds. Such timing misalignment, if directly used for subsequent correlation analysis, will introduce false causal relationships. To this end, the system performs time alignment on the acquired raw data sequence: using a unified time axis as a reference, the time point with the highest sampling rate is selected as the reference point, and other data sequences with lower sampling rates or fixed delays are resampled using linear interpolation or spline interpolation, so that tension value, wire diameter value and die temperature value exist simultaneously at each sampling moment, thereby generating an equidistant aligned data stream.

[0023] After alignment, the data still contains high-frequency components from mechanical vibration, electromagnetic interference, and sensor noise, and may also contain abnormal peaks caused by transient interference. To address this issue, the system employs Kalman filtering for denoising. Specifically, for each dimension of sensor data, a first- or second-order state-space model is established, using the actual physical quantity as the state variable and the measured value as the observation variable. The optimal state value is estimated in real-time through a two-step recursive process of prediction and update. Kalman filtering effectively suppresses Gaussian white noise while preserving the edge variation characteristics of tension, wire diameter, and temperature, avoiding the phase lag caused by simple low-pass filtering. Simultaneously with filtering, the system executes outlier removal logic: it calculates a standardized index for the filtering residual (the difference between the measured and predicted values) at each sampling point. If this index exceeds a preset threshold (e.g., three standard deviations), the point is considered an outlier and replaced with the filtered predicted value before output. The replaced data continues to participate in subsequent window sliding calculations to prevent abnormal peaks from distorting subsequent feature extraction results. After time alignment, Kalman filtering, and outlier removal, the system obtains a calibration sequence stream of data. In this data stream, tension, wire diameter, and die temperature are strictly synchronized in time, with significantly reduced noise levels and no isolated outliers. This calibration sequence stream provides a clean and aligned input basis for subsequent segmented extraction of tension anomaly indices and material stress concentration characteristics, ensuring the reliability of anomaly detection from the source. For example, when the die temperature slowly drifts due to heater malfunction, the calibrated temperature sequence accurately reflects this drift trend. Meanwhile, the aligned tension and wire diameter data synchronously capture the decrease in tension and increase in wire diameter caused by changes in melt viscosity due to temperature variations, providing the original time-matching basis for temperature drift correction and correlation coefficient calculation in step S2. Thus, by combining the filtering and alignment algorithm features in step S1 with the multi-sensor synchronous sampling technology in the cable rubber wire extrusion scenario, the system solves the problems of misjudgment and missed detection of anomalies caused by data asynchrony and noise interference in existing technologies, laying a reliable data foundation for fully closed-loop intelligent monitoring.

[0024] Step S2: Extract the tension anomaly index and material stress concentration characteristic value from the calibration sequence stream data in segments, calculate the correlation coefficient after temperature drift correction, and mark the potential abnormal fluctuation range when the correlation coefficient exceeds the preset correlation threshold and the die temperature is within the process allowable range.

[0025] Specifically, step S2 includes: segmenting the calibration sequence stream data using an adaptive time window, and extracting the peak position, half-peak width, and right-tail integral of the probability density distribution of the tension change rate as the tension anomaly index. This includes: calculating the first-order difference of the tension data within each time window to obtain the tension change rate sequence; fitting the probability density distribution of the tension change rate sequence using kernel density estimation; extracting the change rate value corresponding to the maximum value of the probability density distribution as the peak position; extracting the distribution width when the probability density distribution drops to half of the maximum value as the half-peak width; extracting the cumulative probability of the probability density distribution change rate being greater than a preset abnormal change rate threshold as the right-tail integral; and using the peak position, half-peak width, and right-tail integral as the three components of the tension anomaly index. Furthermore, the material stress concentration characteristic value is obtained by integrating the second-order difference of the wire diameter and the peak-valley difference, including: calculating the center second-order difference of the wire diameter data within each time window and taking the mean of the absolute values ​​to obtain the strain rate mutation index; extracting the difference between the local maxima and local minima of the wire diameter within the time window to obtain the geometric defect amplitude index; calculating the ratio of the geometric defect amplitude index to the strain rate mutation index to obtain the defect morphology discrimination index; and performing a weighted summation of the strain rate mutation index, the geometric defect amplitude index, and the defect morphology discrimination index to generate the material stress concentration characteristic value. Then, the correlation coefficient is calculated for the tension anomaly index and the material stress concentration characteristic value after temperature drift correction. If its absolute value exceeds a preset correlation threshold and the die temperature is within the process allowable range, it is marked as a potential abnormal fluctuation range.

[0026] More specifically, step S2 performs segmented feature extraction and potential anomaly detection on the calibration sequence stream data. The system uses an adaptive time window to segment the calibration sequence stream data. The window length is dynamically adjusted according to the current traction speed, so that each window corresponds to a fixed length of extruded cable, for example, each meter of cable is considered as an analysis unit. This avoids inconsistencies in the manifestation of the same physical defect in different time windows due to changes in production speed. For the tension data sequence within each time window, the system calculates its first difference to obtain the tension change rate sequence. This sequence eliminates the influence of the absolute value shift of tension and directly reflects the dynamic characteristics of tension fluctuations. The probability density distribution of the tension change rate sequence is fitted by kernel density estimation. This estimation method does not make Gaussian or symmetric assumptions about the distribution shape and can realistically characterize the complex distribution shapes such as bimodal and skewed tension change rates that may occur in actual production. Three components are extracted from this probability density distribution: the peak position, which is the tension change rate corresponding to the maximum probability density value, indicating the most common tension change trend within the window. If this value is far from zero, it indicates a systematic increase or decrease in tension. The half-peak width, which is the width between the two points when the probability density drops to half of its maximum value, quantifies the dispersion of tension fluctuations. An increased width indicates that tension control tends to be unstable. The right-tailed integral, which is the integral over the probability density region greater than the preset abnormal change rate threshold, reflects the probability of a large positive tension spike. The larger the integral value, the higher the risk of sudden load application. These three components together constitute the tension anomaly index, characterizing tension behavior from the three dimensions of trend, fluctuation, and spike, respectively.

[0027] For the wire diameter data sequence within the same time window, the system calculates the central second-order difference and takes the average of the absolute values ​​to obtain the strain rate mutation index. The central second-order difference is calculated by subtracting twice the average of the previous and next neighboring points from the current point value and then adding the neighboring point values. Its absolute average reflects the degree of local curvature change of the wire diameter curve and is sensitive to diameter mutations caused by internal stress concentration in the material. Simultaneously, the difference between the local maxima and minima of the wire diameter within the window is extracted to obtain the geometric defect amplitude index. This index quantifies the peak and valley depth of the wire diameter fluctuation within the window, directly corresponding to the bulge or depression size on the surface of the extruded cable. The ratio of the geometric defect amplitude index to the strain rate mutation index is used as a defect morphology discrimination index: a ratio less than one indicates that the wire diameter change is mainly gradual fluctuation, corresponding to uneven melt flow; a ratio greater than one indicates the presence of sharp mutations, corresponding to sudden disturbances such as mechanical vibration or traction slippage. The stress concentration characteristic value of the material is generated by weighted summation of the strain rate mutation index, geometric defect amplitude index and defect morphology discrimination index. The weighting coefficient is calibrated by historical process data, so that different types of stress concentration defects are separable on this characteristic value.

[0028] Both the tension anomaly index and the material stress concentration characteristic value are affected by die temperature drift. Taking the extrusion of natural rubber insulated wire as an example, when the die temperature rises from 90 degrees Celsius to 100 degrees Celsius, the melt viscosity decreases by about 20%, causing the peak position of the tension anomaly index to shift towards a negative value. The material stress concentration characteristic value also decreases due to the improved melt flowability. If not corrected, it will be misjudged as a process anomaly. The system performs temperature drift correction on the above two characteristics. Specifically, it subtracts the temperature compensation term obtained through univariate regression analysis from the difference between the current die temperature and the reference temperature to obtain the corrected characteristic value. Then, the Pearson correlation coefficient between the corrected tension anomaly index and the material stress concentration characteristic value is calculated. If the absolute value of the correlation coefficient exceeds the preset correlation threshold (e.g., 0.7) and passes the significance test (e.g., p-value less than 0.05), it indicates that there is a significant correlation between the statistical distribution characteristics of tension and the geometric defect characteristics of wire diameter. This correlation is a typical manifestation of a complex cause such as screw speed and traction speed mismatch and melt temperature anomaly. The system simultaneously determines whether the average die temperature of the current window is within the allowable range of the process. Only when the correlation coefficient exceeds the threshold and the die temperature is normal is the time window marked as a potential abnormal fluctuation range. For example, when a traction motor encoder failure causes periodic fluctuations in traction speed, the probability density distribution of the tension change rate shows a double peak, the half-peak width increases significantly, and the wire diameter data shows peak-valley fluctuations of the same frequency, the geometric defect amplitude index increases, and the corrected correlation coefficient is close to positive one. Based on this, the system accurately marks the potential abnormal fluctuation range when the temperature is normal. It can be seen that the algorithm features such as adaptive window segmentation, probability density statistics, second-order difference fusion, and temperature correction in step S2, together with the physical coupling relationship between tension, wire diameter, and temperature in the cable extrusion process, functionally support each other and jointly solve the problem that existing technologies are difficult to identify early composite anomalies and are susceptible to temperature interference, providing clear boundaries for abnormal time periods for subsequent knowledge graph diagnosis.

[0029] It should be noted that the system calculates the correlation coefficient and judges the die temperature status sequentially for each adaptive time window. If the absolute value of the correlation coefficient between the corrected tension anomaly index and the material stress concentration characteristic value within the current window does not exceed the preset correlation threshold, it is determined that there is no significant statistical correlation between tension fluctuations and wire diameter geometric defects within that window, and it is not marked as a potential abnormal fluctuation interval. The system then slides to the next time window to continue analysis. If the die temperature exceeds the process allowable range, regardless of whether the correlation coefficient exceeds the threshold, the system does not mark the window as a potential abnormal fluctuation interval and independently triggers a temperature anomaly alarm signal. This is because die temperature drift itself independently affects melt viscosity, thereby changing the response characteristics of tension and wire diameter. The calculated correlation coefficient may be distorted due to temperature interference and is not suitable as a criterion for composite anomalies. Only when both conditions are met—the absolute value of the correlation coefficient exceeding the preset correlation threshold and the die temperature being within the process allowable range—is the system marked the time window as a potential abnormal fluctuation interval. For cases where the correlation coefficient is within the threshold and the temperature is normal, the system determines that the current production status is stable and does not generate any markers. For cases where the temperature exceeds the allowable range, regardless of the correlation coefficient result, the system reports the temperature anomaly as an independent event, and the upper-level control system adjusts the heating or cooling according to the temperature control strategy, without proceeding to the subsequent knowledge graph diagnostic process. This ensures the specificity of the markers for potential abnormal fluctuation ranges, avoids false markers caused by temperature anomalies or random fluctuations, and ensures that the ranges entering step S3 are all periods in which tension-wire diameter coupling anomalies truly exist under suitable process conditions, thereby improving the utilization efficiency of diagnostic resources.

[0030] Step S3: Concatenate the tension anomaly index, material stress concentration characteristic value, die temperature and current working condition parameters of the potential abnormal fluctuation range into a query feature vector, inject it into the pre-constructed historical fault knowledge graph, and determine the risk level and root cause type of the potential abnormal fluctuation range.

[0031] Specifically, step S3 includes: concatenating the components of the tension anomaly index vector of the potential abnormal fluctuation range, the material stress concentration characteristic value, the average die temperature, and the screw speed setting, traction speed setting, and tension setting value under the current working condition into a query feature vector, and performing normalization processing; injecting the query feature vector as a virtual query node into a pre-constructed historical fault knowledge graph, which includes fault case nodes, fault mode nodes, root cause type nodes, and risk level nodes, with edge types including instantiation relations, attribution relations, level mapping relations, and parameter transfer relations; performing multi-layer neighbor aggregation through a graph attention network, calculating the attention coefficient of the virtual query node for each fault case node, and selecting a predetermined number of fault case nodes with the highest attention coefficient as a similar case set; performing weighted voting based on the attention coefficient normalization weight for each case in the similar case set to determine the risk level; and summing the attention weights of each root cause type node in groups and taking the maximum value to determine the root cause type.

[0032] More specifically, step S3 integrates multiple feature quantities corresponding to the potential abnormal fluctuation intervals marked in step S2 into a query feature vector, and uses a historical fault knowledge graph for intelligent diagnosis. For each marked time window, the system extracts three components of the tension anomaly index calculated within that window (peak position, half-peak width, and right-tail integral), material stress concentration feature value, average die temperature within the window, and the screw speed setpoint, traction speed setpoint, and tension setpoint read from the controller under the current operating conditions. These data are then concatenated into a multi-dimensional query feature vector. Since the dimensions and numerical ranges of different feature quantities vary greatly—for example, the peak position of the tension change rate is usually on the order of a few tenths of a Newton per second while the screw speed setpoint can reach tens of revolutions per minute—the system normalizes this vector, typically using maximum-minimum normalization or Z-score standardization, to unify the numerical range of each dimension to near zero or within the same range, thus preventing features with larger numerical values ​​from dominating the similarity matching results in subsequent calculations.

[0033] The normalized query feature vector is injected as a virtual query node into a pre-built historical fault knowledge graph. This knowledge graph is constructed based on fault cases accumulated during historical production processes, with each case corresponding to a labeled abnormal period. Node types in the graph include fault case nodes, fault mode nodes, root cause type nodes, and risk level nodes. Fault case nodes store the feature vector of the case; fault mode nodes describe the specific manifestation of the anomaly, such as "periodic tension fluctuations accompanied by synchronous fluctuations in wire diameter"; root cause type nodes record the root cause of the anomaly, such as "traction motor encoder wear" or "screw speed loop PID parameter mismatch"; risk level nodes label the risk level corresponding to the case, such as low risk, medium risk, and high risk. Edge types include instantiation relationships (fault case nodes belong to a fault mode node), attribution relationships (fault mode nodes are caused by a root cause type node), level mapping relationships (fault case nodes correspond to a risk level node), and parameter transfer relationships (similar associations between different fault cases in terms of operating parameters).

[0034] The system performs multi-layer neighbor aggregation on the knowledge graph after injecting virtual query nodes using a graph attention network. In the first layer aggregation, the virtual query node sends a query signal to its directly connected fault case nodes, and calculates the attention coefficient between the virtual node and each fault case node. The attention coefficient is calculated by concatenating the feature vectors of the virtual query node and the fault case nodes after linear transformation, mapping them to a scalar value through a single-layer feedforward neural network, and finally normalizing them using a softmax function so that the sum of the attention coefficients of all fault case nodes is 1. This attention coefficient reflects the similarity between the current query and each historical fault case. In the second layer aggregation, the system propagates the attention coefficients of fault case nodes to root cause type nodes and risk level nodes along the attribution relationship edges and the level mapping relationship edges. That is, the attention weight of a certain root cause type is equal to the sum of the attention coefficients of all fault case nodes belonging to it.

[0035] It should be noted that the linear transformation matrix parameters and single-layer feedforward neural network weight parameters in the graph attention network are obtained through the following training process: Before system deployment, labeled cases in the historical fault case database are used as training samples. The input of each sample is the feature vector of that case, and the labels include root cause type classification labels and risk level classification labels. Training adopts a multi-task joint loss function, where the root cause type classification task uses the cross-entropy loss function, and the risk level classification task uses the cross-entropy loss function. The two losses are weighted and summed with preset weights (e.g., each accounting for 0.5) to obtain the total loss function. The model parameters are optimized by mini-batch gradient descent until the total loss function converges on the validation set. After training, the model parameters are fixed and deployed to the production site. The system periodically (e.g., monthly) adds newly added labeled fault cases to the knowledge graph and updates the model parameters using incremental training to maintain the consistency between the diagnostic model and actual production.

[0036] After completing multi-level aggregation, the system selects a predetermined number of fault case nodes with the highest attention coefficients as a set of similar cases. This predetermined number can be set to five or ten. To determine the risk level, the system performs a weighted voting process on the risk level of each case in the set: the number of votes corresponding to each case's risk level is the case's attention coefficient. All cases are grouped and summed according to risk level to obtain the total attention weight for each risk level. The risk level with the highest weight is taken as the risk level of the current potential abnormal fluctuation range. To determine the root cause type, the system directly compares the attention weights of nodes for each root cause type, and takes the root cause type corresponding to the maximum weight as the diagnostic result. For example, in a certain anomaly, the three fault cases with the highest attention coefficients are "traction motor encoder failure (attention coefficient 0.5, high risk)," "traction drive parameter drift (0.3, medium risk)," and "tension sensor zero-point drift (0.2, low risk)." After weighted voting, the total weight for high risk is 0.5, medium risk is 0.3, and low risk is 0.2, thus determining the current risk level as high risk. Simultaneously, the attention weight of "traction motor encoder failure" as the root cause type node is 0.5, higher than the 0.3 for "traction drive parameter drift" and the 0.2 for "tension sensor zero-point drift." Therefore, the root cause type is determined to be traction motor encoder failure. The algorithmic features combining the attention network and knowledge graph in step S3, along with the historical experience accumulated from anomaly cases in the cable extrusion production scenario, functionally support each other, jointly solving the technical problem of existing technologies relying on manual experience for diagnosis and difficulty in quantifying risk levels. This provides clear risk level and root cause type inputs for subsequent collaborative optimization.

[0037] Step S4: Establish a parameter coupling constraint matrix based on the physical model of the extrusion process, and generate risk distribution characteristic data according to the parameter coupling constraint matrix, the risk level and the root cause type.

[0038] Step S4 includes: establishing a screw plasticizing model, a die forming model, and a traction stretching model. The screw plasticizing model calculates the melt mass flow rate and melt temperature based on the screw speed. The die forming model calculates the die extrusion speed and extruded wire diameter based on melt parameters. The traction stretching model calculates the finished wire diameter and actual tension based on die parameters and traction speed. The screw speed, traction speed, and tension are selected as state variables, and the finished wire diameter and actual tension are selected as observed variables. At the current steady-state operating point, the implicit equations solved simultaneously by the three models are linearized using Taylor expansion, and the Jacobian matrix is ​​extracted as the parameter coupling constraint matrix. Fourier transform is performed on the wire diameter data in the calibration sequence stream data. Leaf transformation is used to extract the dominant frequency component as the wire diameter change frequency data; statistical distribution fitting is performed on the tension anomaly index sequence in the calibration sequence stream data to obtain tension anomaly distribution data; an initial evaluation matrix is ​​constructed by aligning it with the time axis; dynamic feature weights are assigned to the wire diameter dimension and tension dimension according to the ratio of the absolute values ​​of the influence of each state variable on the observed variable in the parameter coupling constraint matrix, and weighted fusion is used to generate a comprehensive anomaly evaluation matrix; the comprehensive anomaly evaluation matrix is ​​graded and labeled according to the risk level, the corresponding root cause type code is embedded, and risk distribution feature data is generated according to the integral mapping of traction speed and time to the physical position coordinates of the cable in the extrusion process corresponding to each time window.

[0039] More specifically, step S4 establishes a parameter coupling constraint matrix based on the physical model of the extrusion process and integrates multi-source data to generate risk distribution characteristic data. The system first establishes a screw plasticizing model, a die forming model, and a traction stretching model. The screw plasticizing model calculates the melt mass flow rate and melt temperature based on the screw speed and material rheological parameters. The melt temperature is affected by both screw shear heat and barrel heating. Specifically, the screw plasticizing model is constructed using a melt transport theory based on power-law fluids. This model takes the screw speed (in revolutions per minute) and the rheological parameters of the rubber material (including consistency coefficient, flow index, and shear rate dependence) as inputs, and calculates the melt mass flow rate (in kilograms per hour) and melt temperature (in degrees Celsius) by solving the equilibrium equations of pressure flow and drag flow within the screw channel. Specifically, the melt mass flow rate is directly proportional to the screw speed and inversely influenced by the melt viscosity. The melt temperature is determined by the superposition of the screw shear heat (proportional to the square of the speed) and the heat flux of the barrel heater, taking into account convective heat transfer losses between the melt and the barrel wall. The die forming model uses the melt mass flow rate, melt temperature, and die geometry as inputs to calculate the die extrusion speed and extruded wire diameter. Specifically, the die forming model is established based on the law of conservation of mass and constitutive equations. This model uses the melt mass flow rate, melt temperature, and die geometry (including die diameter and shaping section length) output from the screw plasticizing model as inputs to calculate the die extrusion speed (unit: meters per minute) and extruded wire diameter (unit: millimeters). In this model, the die extrusion speed is equal to the melt mass flow rate divided by the melt density and the die cross-sectional area. The extruded wire diameter at the die exit is approximately equal to the die diameter, but it needs to be corrected for the melt exit expansion effect (i.e., the Burroughs effect). This correction factor has a nonlinear relationship with the melt's elastic parameters and shear rate. The traction-stretch model calculates the finished wire diameter and actual tension based on the die extrusion speed, traction speed, and the tensile properties of the rubber material. Specifically, the traction-stretch model is established based on a series of Hooke's law and viscoelastic constitutive models. This model takes the die extrusion speed, extruded wire diameter, and traction speed (unit: meters per minute) output by the die forming model as inputs, and introduces the tensile property parameters of the rubber material (including tensile modulus and relaxation time). By solving the tensile stress equilibrium differential equation, the finished wire diameter and actual tension are calculated. The finished wire diameter equals the extruded wire diameter divided by the draw ratio (the draw ratio is defined as the ratio of the traction speed to the die extrusion speed, scaled by the square root). The actual tension equals the integral of the internal stress along the cross-section generated by elastic deformation and viscous dissipation of the rubber wire during the stretching process. This tension has an approximately linear relationship with the excess of the draw ratio (actual draw ratio minus equilibrium draw ratio), and the proportionality coefficient is determined by the dynamic modulus of the material. The above three models together constitute a set of nonlinear equations describing the steady-state behavior of the extrusion process, providing a physical basis for subsequent linearization and extraction of the Jacobian matrix.

[0040] The system selects screw speed, traction speed, and tension setpoints as state variables, and finished wire diameter and actual tension as observed variables. At the current steady-state operating point, a Taylor expansion of the implicit equations is performed, retaining first-order terms to linearize the nonlinear model. The Jacobian matrix between the increments of state variables and observed variables is extracted. Each element of this matrix represents the change in an observed variable caused by a unit change in a state variable, i.e., the parameter coupling sensitivity. This Jacobian matrix serves as the parameter coupling constraint matrix, used to quantitatively describe the interrelationships between various process parameters during extrusion. For example, the partial derivative of screw speed with respect to finished wire diameter is positive, indicating that increasing the speed will thicken the wire; the partial derivative of traction speed with respect to finished wire diameter is negative, indicating that increasing traction will thin the wire; the partial derivative of tension setpoint with respect to actual tension is close to 1, but its effect on wire diameter is weak. This coupling constraint matrix provides the physical constraint basis for subsequent collaborative optimization.

[0041] During the generation of risk distribution characteristic data, the system performs a Fast Fourier Transform (FFT) on the wire diameter data in the calibration sequence stream data. Specifically, it extracts a predetermined length (e.g., corresponding to a 50-meter cable) of wire diameter data sequence from the current time window, inputs it into the FFT algorithm, calculates the spectral distribution, and extracts the frequency component with the largest amplitude in the spectrum as the wire diameter change frequency data. This frequency value can be used to determine whether there is a periodic disturbance source, such as traction wheel eccentricity or screw speed pulsation. Simultaneously, the system constructs a sequence of tension anomaly indices calculated for each time window in chronological order, performs statistical distribution fitting on this sequence, and uses kernel density estimation to obtain its probability density curve. Then, it extracts distribution morphology parameters (such as skewness, kurtosis, and bimodal coefficient) as tension anomaly distribution data to characterize the overall stability of tension fluctuations. After aligning along the time axis, the system merges the wire diameter change frequency data, tension anomaly distribution parameters, and the original anomaly scores for each window to construct an initial evaluation matrix. Each row of the matrix corresponds to a time window, and each column corresponds to a characteristic index. The system reads the parameter coupling constraint matrix (i.e., the Jacobian matrix), where each element represents the sensitivity of a state variable (e.g., screw speed) to an observed variable (e.g., finished wire diameter). For the wire diameter dimension, the system extracts the absolute values ​​of the sensitivity of all state variables to the finished wire diameter from the coupling matrix, sums them, and uses this sum as the baseline weight for the wire diameter dimension. Similarly, for the tension dimension, the system extracts the sum of the absolute values ​​of the sensitivity of all state variables to the actual tension and uses this sum as the baseline weight for the tension dimension. Then, the ratio of the two baseline weights is calculated. For example, if the total sensitivity of the wire diameter dimension is twice that of the tension dimension, the dynamic feature weight for the wire diameter dimension is set to two-thirds, and for the tension dimension, it is set to one-third. Based on these dynamic weights, the system performs a weighted linear combination of the data in the wire diameter-related and tension-related columns of the initial evaluation matrix, fusing them to generate a comprehensive anomaly evaluation matrix. Each element of this matrix quantifies the comprehensive anomaly degree of the corresponding time window under multi-parameter coupling.

[0042] Based on the risk levels determined in step S3, the system hierarchically labels the comprehensive anomaly assessment matrix, assigning different color codes or numerical markers to high-risk, medium-risk, and low-risk individuals. Simultaneously, it embeds the root cause type code determined in step S3, ensuring that each anomaly window contains both a quantitative anomaly degree and a semantic diagnostic conclusion. Based on the integral of traction speed over time, the system calculates the cable length coordinates corresponding to each time window, mapping them to its physical location in the extrusion process, such as the meter coordinates measured from the die exit. Combining the above information, the system generates risk distribution characteristic data, which structurally describes the distribution of comprehensive anomaly degree, risk level zoning, root cause type distribution, and the dominant frequency characteristics of wire diameter variation along the length of the extruded cable. For example, in a certain production batch, the risk distribution characteristic data shows a high-risk anomaly in the range of 50 to 52 meters from the die exit, with the root cause type code being a traction motor encoder failure and the dominant frequency of wire diameter variation being 2 Hz, consistent with the fluctuation frequency of traction speed within this range. This provides precise location information and physical basis for the collaborative optimization in step S5. In step S4, the algorithm features such as physical model linearization to extract coupling matrix, frequency domain feature extraction, and dynamic weight fusion support each other functionally with the physical characteristics of multivariable coupling in the cable extrusion process. Together, they solve the technical problems of existing technologies lacking quantitative constraint relationships between parameters and being unable to generate structured risk distribution data.

[0043] Step S5: If the risk distribution characteristic data reaches the intervention trigger condition, multi-parameter collaborative optimization is performed with wire diameter stability as the objective function, tension safety range as the hard constraint, and the parameter coupling constraint matrix as the dynamic equality constraint to generate a collaborative adjustment vector and generate a preliminary intervention signal.

[0044] Specifically, step S5 includes: if the risk level in the risk distribution characteristic data reaches or exceeds the preset intervention level threshold, then the intervention trigger condition is determined to be met; a multi-parameter collaborative optimization problem is constructed with the objective function being the minimization of the square mean of the difference between the wire diameter measurement value and the target wire diameter, the hard constraint being that the actual tension value is within the preset tension safety range, the soft constraint being the linear equation relationship between the increment of the observed variable and the increment of the state variable established by the parameter coupling constraint matrix, and the boundary constraint being the equipment capability range of each adjustment amount; the multi-parameter collaborative optimization problem is solved using a sequential quadratic programming algorithm to generate a collaborative adjustment vector containing the screw speed adjustment amount, the traction speed adjustment amount, and the tension setpoint adjustment amount; the collaborative adjustment vector, the abnormal position coordinates, the risk level, and the root cause type are fused to generate a preliminary intervention signal.

[0045] More specifically, step S5 determines whether to trigger intervention based on risk distribution characteristic data and generates adjustment instructions using a multi-parameter collaborative optimization approach. The system reads the risk distribution characteristic data generated in step S4 and extracts the risk level information contained therein. If the risk level corresponding to the current time window reaches or exceeds the preset intervention level threshold (e.g., high-risk level triggers intervention while medium- and low-risk levels do not), then the intervention triggering condition is deemed met, and the system enters the collaborative optimization decision-making process.

[0046] The optimization problem is constructed with wire diameter stability as the core objective. The objective function is set to minimize the mean square of the difference between the measured wire diameter and the target wire diameter, i.e., minimizing the mean square error of the wire diameter deviation within a prediction period. This objective directly corresponds to the key indicator of extruded product quality control—wire diameter accuracy. The hard constraint condition is that the actual tension value must be within the preset tension safety range. This range is determined by the tensile strength of the cable material and the production process specifications. For example, the lower limit of the tension safety range for natural rubber insulated wire is the minimum tension to prevent slack, and the upper limit is the maximum tension to prevent breakage. Violating this constraint will lead to wire breakage or winding failure, so it is an inviolable hard constraint. The soft constraint condition is the linear equation relationship determined by the parameter coupling constraint matrix established in step S4, i.e., the increment of the observed variable (the increment of the finished wire diameter and the increment of the actual tension) is equal to the Jacobian matrix multiplied by the increment of the state variable (the increment of the screw speed, the increment of the traction speed, and the increment of the tension setpoint). This equation relationship originates from the linearization approximation of the physical model of the extrusion process near the working point and reflects the coupling sensitivity between parameters. Since the model has linearization errors, it is introduced as a penalty term in the objective function as an equation soft constraint. Boundary constraints are the equipment capacity range for each adjustment. For example, the screw speed adjustment must not exceed the maximum single change step size and absolute upper and lower limits allowed by the frequency converter; the traction speed adjustment is limited by the rated speed and acceleration / deceleration capability of the motor; and the tension setpoint adjustment is limited by the output range of the tension controller.

[0047] The aforementioned multi-parameter collaborative optimization problem is solved using a sequential quadratic programming algorithm. In each iteration, this algorithm approximates the original nonlinear constraint optimization problem as a quadratic programming subproblem. The search direction is obtained by solving the subproblem, and the optimization variables are updated after determining the step size through a one-dimensional search. This iteration is repeated until convergence to the optimal solution. The solution is a three-dimensional vector containing the screw speed adjustment, traction speed adjustment, and tension setpoint adjustment, denoted as the collaborative adjustment vector. For example, in the extrusion production of a batch of cable rubber wire, the target wire diameter is 10 mm, the current measured wire diameter is 10.3 mm, and the actual tension value is close to the lower limit of the safe range, indicating a high risk level. The system uses a parameter coupling constraint matrix as a soft constraint, where the Jacobian matrix shows that a 1% increase in screw speed approximately increases the wire diameter by 0.2 mm, and a 1% increase in traction speed approximately decreases the wire diameter by 0.15 mm. The collaborative adjustment vector obtained after solving the sequential quadratic programming is: screw speed reduced by 2%, traction speed increased by 3%, and tension setpoint increased by 5%. This solution restores the wire diameter to the target value of 10 mm while increasing the tension to the middle of the safe range, and all adjustments do not exceed the equipment's capacity limits.

[0048] After generating the coordinated adjustment vector, the system integrates the coordinated adjustment vector, the abnormal location coordinates obtained in step S4 (e.g., the range of 50 to 52 meters from the die opening), and the risk level and root cause type determined in step S3 to generate a preliminary intervention signal. This signal not only contains specific parameter adjustment values ​​but also carries the spatial location of the abnormality and diagnostic conclusions, facilitating subsequent steps to determine execution priorities based on risk levels and using the abnormal location information for quality traceability. The sequential quadratic programming optimization algorithm based on physical model constraints in step S5 functionally supports the coupling relationship between wire diameter, tension, and process parameters during cable extrusion and the physical limitations of the equipment. Together, they solve the problems of repeated fluctuations and long adjustment times caused by single-parameter trial-and-error adjustments in existing technologies, achieving one-time coordinated optimization of multiple parameters.

[0049] Step S6: After performing dual verification of the device adjustment parameter range and device compatibility rules on the preliminary intervention signal, the device is adjusted, and the process returns to step S1 to form a closed-loop monitoring and adjustment mechanism.

[0050] Specifically, step S6 includes: obtaining the coordinated adjustment vector in the preliminary intervention signal; comparing the screw speed adjustment, traction speed adjustment, and tension setpoint adjustment with the adjustment parameter range of the corresponding equipment; truncating the adjustment amounts exceeding the boundary to the boundary value to obtain the first verification result; combining the adjustment amounts in the first verification result into a parameter adjustment set; verifying whether it satisfies the speed response coupling constraint between the tension controller and the traction motor driver, the temperature coupling constraint between the screw speed adjustment and the melt temperature change, and the motor acceleration / deceleration capability constraint; if not satisfied, solving for the minimum value based on the pseudo-inverse of the parameter coupling constraint matrix. The correction amount is iteratively corrected to obtain the final adjustment parameters; the execution sequence is determined according to the risk level, and the final adjustment parameters are written into the corresponding control registers of the extruder frequency converter, traction motor driver and tension controller through the industrial fieldbus to execute equipment adjustment; after the parameters are updated, the data is collected again in step S1, the deviation between the actual feedback value and the set value in the final adjustment parameters is calculated, and steps S1 to S6 are repeated to form a closed-loop monitoring and adjustment mechanism. When the tension fluctuation amplitude and the wire diameter fluctuation amplitude are both lower than the corresponding preset stable threshold and no new abnormal fluctuation range is generated for several consecutive cycles, it is determined that a stable production state has been obtained.

[0051] More specifically, step S6 performs double verification on the preliminary intervention signal generated in step S5, then implements equipment adjustment and forms a closed-loop monitoring mechanism. The system obtains the coordinated adjustment vector in the preliminary intervention signal, which includes the screw speed adjustment, traction speed adjustment, and tension setpoint adjustment. The first verification targets the adjustment parameter range of a single piece of equipment: the screw speed adjustment is compared with the maximum single change step size and absolute upper and lower limits of the speed allowed by the extruder frequency converter; if it exceeds these limits, the adjustment is truncated to the boundary value. Similarly, the traction speed adjustment and tension setpoint adjustment are compared with the output range of the traction motor driver and tension controller, respectively, and truncated to obtain the first verification result. The truncation operation ensures that each adjustment can be safely executed within the physical limits of its respective actuator.

[0052] The second verification checks the compatibility rules of the combined adjustments. The system combines the adjustment values ​​from the first verification result into a parameter adjustment set and checks whether this set satisfies three types of constraints. The first type is the speed response coupling constraint between the tension controller and the traction motor driver, that is, the rate of change of traction speed cannot exceed the response bandwidth of the tension controller, otherwise the tension will overshoot or oscillate. The second type is the temperature coupling constraint between screw speed adjustment and melt temperature change, that is, the shear heat caused by a sudden increase in screw speed should not cause the melt temperature to exceed the process upper limit, so the increase in speed must match the rate of temperature rise. The third type is the motor acceleration and deceleration capability constraint, that is, the acceleration of the screw motor and the traction motor cannot exceed the rated acceleration and deceleration capability of their respective drivers. If the parameter adjustment set satisfies all the above constraints, it is directly used as the final adjustment parameter; if not, the system solves for the minimum correction amount based on the pseudo-inverse of the parameter coupling constraint matrix established in step S4. Specifically, the unmet constraints are represented as a system of linear inequalities. With the goal of minimizing the L2 norm of the difference between the original adjustment and the correction, the least squares solution is obtained using the pseudo-inverse of the constraint matrix to obtain the corrected adjustment. The process is iteratively verified until all compatibility constraints are satisfied, thereby obtaining the final adjustment parameters.

[0053] After completing the dual verification, the system determines the execution sequence based on the risk level determined in step S3: adjustment instructions corresponding to high-risk levels are executed immediately with the highest priority; medium-risk levels can be executed sequentially within the current control cycle; and low-risk levels are delayed until the next control cycle or executed after operator confirmation. The system writes the final adjustment parameters into the speed setting register of the extruder inverter, the speed setting register of the traction motor driver, and the tension setting register of the tension controller via an industrial fieldbus (such as Profinet or EtherCAT), and executes equipment adjustment. After the parameters are updated, the system automatically returns to step S1, re-collects tension, wire diameter, and die temperature data, and calculates the deviation between the actual feedback value and the set value in the final adjustment parameters. Steps S1 to S6 are repeated to form a closed-loop monitoring and adjustment mechanism. Taking cable rubber wire extrusion as an example, after one coordinated adjustment, the wire diameter gradually approaches the target value of 10 mm from 10.3 mm, and the tension rises from the lower limit of the safe range to the middle. At this time, the system continues to monitor. If, in subsequent consecutive cycles (e.g., ten cycles), the tension fluctuation amplitude and wire diameter fluctuation amplitude are both lower than the corresponding preset stability threshold (e.g., fluctuation amplitude less than ±2%), and no new potential abnormal fluctuation range is generated, the system determines that a stable production state has been achieved, and the current parameters can be maintained and continuously monitored. The algorithm features of dual verification, pseudo-inverse iterative correction, and closed-loop feedback adjustment in step S6 functionally support the physical limitations of the cable extrusion equipment and the controller response characteristics, jointly solving the technical problems in the prior art where adjustment commands may exceed the equipment safety boundary and the lack of feedback verification after adjustment leads to repeated abnormalities, thus realizing safe and reliable fully automatic closed-loop control.

[0054] In a comparative test of continuous extrusion production, the traditional cable monitoring method and the intelligent monitoring method for cable rubber wire production of this invention showed significant differences in tension fluctuation, wire diameter fluctuation, and the number of breakage risks. The traditional method, using single-parameter threshold alarms and manual adjustments, resulted in a peak tension fluctuation of 22 Newtons after a traction speed disturbance, exceeding the safety limit by approximately 83%, and a peak wire diameter deviation of 0.18 mm. In contrast, the method of this invention, through potential abnormal zone prediction, knowledge graph diagnosis, and multi-parameter collaborative optimization, consistently controlled the tension fluctuation amplitude within the safe range of 12 Newtons under the same disturbance conditions, reducing the peak wire diameter deviation to below 0.02 mm. These comparisons demonstrate that this invention effectively suppresses drastic fluctuations in tension and wire diameter, fundamentally eliminating the risk of cable breakage caused by tension impacts, and significantly improving the safety and stability of the extrusion process.

[0055] Specifically, refer to the appendix Figure 2 Appendix Figure 3 The tension fluctuation comparison curve and the wire diameter fluctuation comparison curve shown in this embodiment further quantify the performance differences between the present invention and the prior art in the production process. Figure 2 In the diagram, the horizontal axis represents time (minutes), and the vertical axis represents the tension fluctuation amplitude (Newtons), with a safety upper limit of 12 Newtons. The traditional method (solid line) shows a sharp increase in tension fluctuation to a peak of 22 Newtons within approximately 0.5 minutes after the introduction of interference, exceeding the safety upper limit by about 83%, and oscillating at this high level for over 1.5 minutes. The method of this invention (dashed line) shows only a slight increase after the introduction of interference, followed by a rapid drop back to around 8.5 Newtons. Throughout the adjustment process, the tension fluctuation remains far below the 12 Newton safety upper limit. Figure 3 In the figure, the horizontal axis represents time (minutes), and the vertical axis represents wire diameter deviation (millimeters). The dashed line represents the control target band of 0.05 millimeters. The traditional method (solid line) produces a positive wire diameter deviation of 0.18 millimeters approximately 5 minutes after the disturbance occurs, and it gradually converges after about 3.5 minutes of trial and error adjustments. The method of this invention (dashed line) has a maximum wire diameter deviation of only -0.02 millimeters after the disturbance occurs, remaining consistently within ±0.05 millimeters, far superior to the dynamic accuracy of the traditional method. Both figures demonstrate that this invention, through potential anomaly prediction and collaborative optimization mechanisms, significantly reduces the peak values ​​of tension fluctuations and wire diameter fluctuations, and greatly compresses the adjustment time, fully verifying the significant technical advantages of this invention in suppressing process fluctuations and improving product quality consistency.

[0056] Secondly, such as Figure 4 As shown, this embodiment provides an intelligent monitoring system for cable rubber wire production, including: The data acquisition module collects tension, wire diameter, and die temperature data during the extrusion process, and generates a calibration sequence stream data after filtering and alignment. The potential anomaly identification module extracts the tension anomaly index and material stress concentration feature values ​​from the calibration sequence stream data in segments, calculates the correlation coefficient after temperature drift correction, and marks a potential anomaly fluctuation range when the correlation coefficient exceeds a preset correlation threshold and the die temperature is within the allowable process range. The knowledge graph diagnosis module concatenates the tension anomaly index, material stress concentration feature values, die temperature, and current operating parameters of the potential anomaly fluctuation range into a query feature vector, injects it into a pre-constructed historical fault knowledge graph, and determines the risk level and root cause type of the potential anomaly fluctuation range. The distribution generation module is used to establish a parameter coupling constraint matrix based on the physical model of the extrusion process, and generate risk distribution characteristic data according to the parameter coupling constraint matrix, the risk level, and the root cause type. The collaborative optimization decision module is used to perform multi-parameter collaborative optimization when the risk distribution characteristic data reaches the intervention trigger condition, with wire diameter stability as the objective function, tension safety range as the hard constraint, and the parameter coupling constraint matrix as the dynamic equality constraint, to generate a collaborative adjustment vector and generate an initial intervention signal. The verification and execution module is used to perform dual verification of the initial intervention signal based on the equipment adjustment parameter range and equipment compatibility rules, and then execute equipment adjustment and return to trigger the data acquisition module to form a closed-loop monitoring and adjustment mechanism.

[0057] Specifically, each module is housed in an industrial control computer and programmable logic controller (PLC) deployed alongside the extrusion production line. The data acquisition module performs high-frequency synchronous sampling using a piezoelectric tension sensor, a high-speed laser diameter gauge, and a thermocouple temperature sensor installed at the extruder die. The sampling frequency is no less than 100 Hz. The three signals are sent to the analog input channel of the PLC via signal conditioning circuitry, and then read by the industrial control computer via industrial Ethernet. The industrial control computer runs a Kalman filter algorithm to perform time alignment, noise filtering, and outlier removal on the three data streams, generating a calibration sequence stream data which is then stored in a real-time database.

[0058] The potential anomaly identification module runs as software within the central processing unit of the industrial control computer. This module reads calibration sequence stream data from a real-time database and dynamically adjusts the time window length based on the current traction speed, ensuring each window corresponds to a fixed length of extruded cable. Within each window, the module calculates the first-order difference of tension and extracts the peak position, half-peak width, and right-tail integral of the probability density distribution using kernel density estimation, forming three components of the tension anomaly index. Simultaneously, it calculates the mean of the absolute values ​​of the second-order difference at the center of the cable diameter to obtain the strain rate mutation index, extracts the difference between the local maxima and minima of the cable diameter to obtain the geometric defect amplitude index, and calculates the ratio and weighted sum of these two indices to generate material stress concentration characteristic values. The module calculates the correlation coefficient after drift correction of the above two characteristics based on the die temperature. When the absolute value of the correlation coefficient exceeds a preset correlation threshold and the die temperature is within the allowable process range, the time window is marked as a potential anomaly fluctuation range, and the feature data within the window is pushed to the knowledge graph diagnostic module.

[0059] The knowledge graph diagnostic module is deployed on the solid-state drive and memory of the industrial control computer. It pre-stores historical fault knowledge graphs, organized as a graph database, containing fault case nodes, fault mode nodes, root cause type nodes, and risk level nodes. Edge types include instantiation relationships, attribution relationships, level mapping relationships, and parameter transfer relationships. The module concatenates the three components of the tension anomaly index of the received potential abnormal fluctuation range, material stress concentration characteristic values, average die temperature, and screw speed setpoints, traction speed setpoints, and tension setpoints read from the programmable logic controller into a query feature vector, which is then normalized. This vector is injected into the knowledge graph as a virtual query node. The module runs a graph attention network algorithm to calculate the attention coefficient between the virtual node and each fault case node. After multi-level neighbor aggregation, the five fault case nodes with the highest attention coefficients are selected as a similar case set. The risk level of each case in the set is determined by weighted voting. The root cause type is determined by summing the attention weights of each root cause type node in groups and taking the maximum value.

[0060] The risk distribution generation module also runs on the industrial control computer. This module first establishes a screw plasticizing model, a die forming model, and a traction and tension model based on the physics of the extrusion process. Screw speed, traction speed, and tension setpoints are selected as state variables, while finished wire diameter and actual tension are selected as observed variables. At the current steady-state operating point, the module performs a Taylor series linearization of the simultaneous equations of the three models, extracts the Jacobian matrix as the parameter coupling constraint matrix, and stores it in memory. This module performs a Fast Fourier Transform on the wire diameter data in the calibration sequence stream data, extracting the dominant frequency component as the wire diameter change frequency data. It then performs statistical distribution fitting on the tension anomaly index sequence for each time window to obtain tension anomaly distribution data. After constructing an initial evaluation matrix aligned with the time axis, dynamic feature weights are assigned to the wire diameter and tension dimensions based on the ratio of the absolute values ​​of the influence of each state variable on the observed variables in the parameter coupling constraint matrix. These weighted weights are then fused to generate a comprehensive anomaly evaluation matrix. The risk level output by the knowledge graph diagnostic module is then graded and labeled, and root cause type codes are embedded. Simultaneously, the physical coordinates of the cable corresponding to each time window are mapped based on the integral of the traction speed over time, generating risk distribution feature data.

[0061] The collaborative optimization decision-making module runs on the industrial control computer. This module determines whether the risk level in the risk distribution characteristic data reaches or exceeds a preset intervention level threshold; if so, it triggers the optimization process. The objective function of the optimization problem is set as minimizing the squared mean of the difference between the measured wire diameter and the target wire diameter. The hard constraint is that the actual tension value is within a preset tension safety range. The dynamic equation soft constraint is the linear equation relationship between the increments of observed variables and state variables established by the parameter coupling constraint matrix. The boundary constraint is the equipment capability range of each adjustment. The module calls a sequential quadratic programming algorithm library to solve the optimization problem, generating a collaborative adjustment vector containing screw speed adjustment, traction speed adjustment, and tension setpoint adjustment. This vector is then fused with the abnormal location coordinates, risk level, and root cause type to form a preliminary intervention signal, which is passed to the verification and execution module via memory sharing.

[0062] The verification and execution module is deployed in the industrial control computer and programmable logic controller. The industrial control computer first compares the screw speed adjustment, traction speed adjustment, and tension setpoint adjustment in the initial intervention signal with the corresponding equipment's adjustment parameter ranges. Adjustments exceeding the boundaries are truncated to the boundary values, yielding the first verification result. Subsequently, the adjustment values ​​in this result are combined into a parameter adjustment set to verify whether they satisfy the speed response coupling constraints between the tension controller and the traction motor driver, the temperature coupling constraints between screw speed adjustment and melt temperature change, and the motor acceleration / deceleration capability constraints. If not satisfied, iterative correction is performed based on the pseudo-inverse of the parameter coupling constraint matrix to solve for the minimum correction amount, obtaining the final adjustment parameters. According to the execution sequence determined by the risk level, the industrial control computer writes the final adjustment parameters into the speed setpoint register of the extruder inverter, the speed setpoint register of the traction motor driver, and the tension setpoint register of the tension controller via the Profinet industrial fieldbus, completing the equipment adjustment. After the parameters are updated, the system automatically triggers the data acquisition module to restart data acquisition, forming a closed-loop monitoring and adjustment mechanism. The algorithm features in the above modules and the hardware such as sensors, actuators and controllers in the extrusion production site support each other in terms of function, and together solve the technical problems of the separation of monitoring and execution and the lack of coordination in parameter adjustment in the existing technology.

[0063] The above description is merely a specific embodiment of this specification. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the scope of protection of this specification is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this specification, and these modifications or substitutions should all be covered within the scope of protection of this specification.

Claims

1. A method for intelligent monitoring of cable rubber wire production, characterized in that, include: Step S1: Collect tension, wire diameter, and die temperature data during the extrusion process, and generate calibration sequence stream data after filtering and alignment; Step S2: Extract the tension anomaly index and material stress concentration feature value from the calibration sequence stream data in segments, calculate the correlation coefficient after temperature drift correction, and mark the potential abnormal fluctuation range when the correlation coefficient exceeds the preset correlation threshold and the die temperature is within the process allowable range. Step S3: Concatenate the tension anomaly index, material stress concentration characteristic value, die temperature and current working condition parameters of the potential abnormal fluctuation range into a query feature vector, inject it into the pre-constructed historical fault knowledge graph, and determine the risk level and root cause type of the potential abnormal fluctuation range; Step S4: Establish a parameter coupling constraint matrix based on the physical model of the extrusion process, and generate risk distribution characteristic data according to the parameter coupling constraint matrix, the risk level and the root cause type; Step S5: If the risk distribution characteristic data reaches the intervention trigger condition, multi-parameter collaborative optimization is performed with wire diameter stability as the objective function, tension safety range as the hard constraint, and the parameter coupling constraint matrix as the dynamic equality constraint to generate a collaborative adjustment vector and generate a preliminary intervention signal. Step S6: After performing dual verification of the device adjustment parameter range and device compatibility rules on the preliminary intervention signal, the device is adjusted, and the process returns to step S1 to form a closed-loop monitoring and adjustment mechanism.

2. The intelligent monitoring method for cable rubber wire production according to claim 1, characterized in that, Step S1 specifically includes: High-frequency synchronous sampling is performed by high-precision tension sensors, laser diameter gauges and temperature sensors deployed in the extrusion process to obtain tension data, wire diameter data and die temperature data. The tension data, wire diameter data, and die temperature data are aligned by time series so that the data are arranged at equal intervals on the same time axis. The aligned data is denoised using Kalman filtering and outliers are removed to obtain the calibration sequence stream data.

3. The intelligent monitoring method for cable rubber wire production according to claim 1, characterized in that, Step S2 specifically includes: An adaptive time window is used to segment the calibration sequence stream data, and the peak position, half-peak width, and right-tail integral of the probability density distribution of the tension change rate are extracted as the tension anomaly index. By combining the second-order difference of the wire diameter and the peak-valley difference, the characteristic value of material stress concentration is obtained; The correlation coefficients of the tension anomaly index and the material stress concentration characteristic value are calculated after temperature drift correction. If their absolute values ​​exceed the preset correlation threshold and the die temperature is within the process allowable range, they are marked as potential abnormal fluctuation ranges.

4. The intelligent monitoring method for cable rubber wire production according to claim 3, characterized in that, The peak position, half-peak width, and right-tail integral of the probability density distribution of the extracted tension change rate are specifically used as the tension anomaly index, including: The tension change rate sequence is obtained by calculating the first difference of the tension data within each time window. The probability density distribution of the tension change rate sequence is fitted by kernel density estimation. The change rate value corresponding to the maximum value of the probability density distribution is extracted as the peak position. The distribution width when the probability density distribution drops to half of the maximum value is extracted as the half-peak width. The cumulative probability of the change rate of the probability density distribution being greater than the preset abnormal change rate threshold is extracted as the right-tail integral. The peak position, half-peak width, and right-tail integral are used as the three components of the tension anomaly index.

5. The intelligent monitoring method for cable rubber wire production according to claim 3 or 4, characterized in that, The material stress concentration characteristic values ​​obtained by fusing the second-order difference of the line diameter and the peak-valley difference specifically include: The strain rate mutation index is obtained by calculating the central second difference of the wire diameter data within each time window and taking the mean of the absolute values. The difference between the local maxima and local minima of the wire diameter within the time window is extracted to obtain the geometric defect magnitude index; The ratio of the geometric defect amplitude index to the strain rate mutation index is calculated to obtain the defect morphology discrimination index; The stress concentration characteristic value of the material is generated by weighted summation of the strain rate mutation index, geometric defect amplitude index, and defect morphology discrimination index.

6. The intelligent monitoring method for cable rubber wire production according to claim 1, characterized in that, Step S3 specifically includes: The components of the tension anomaly index vector in the potential abnormal fluctuation range, the material stress concentration characteristic value, the average die temperature, and the screw speed setting, traction speed setting, and tension setting value under the current working condition are concatenated into a query feature vector and then normalized. The query feature vector is injected as a virtual query node into a pre-built historical fault knowledge graph. The historical fault knowledge graph includes fault case nodes, fault mode nodes, root cause type nodes, and risk level nodes. The edge types include instantiation relationships, attribution relationships, level mapping relationships, and parameter passing relationships. Multi-layer neighbor aggregation is performed using a graph attention network. The attention coefficient of the virtual query node to each fault case node is calculated, and a predetermined number of fault case nodes with the highest attention coefficients are selected as a set of similar cases. The risk level of each case in the set of similar cases is determined by weighted voting based on the normalized weight of the attention coefficient; the root cause type is determined by summing the attention weights of each root cause type node in groups and taking the maximum value.

7. The intelligent monitoring method for cable rubber wire production according to claim 1, characterized in that, Step S4 specifically includes: Establish a screw plasticizing model, a die forming model, and a traction stretching model. The screw plasticizing model calculates the melt mass flow rate and melt temperature based on the screw speed. The die forming model calculates the die extrusion speed and extruded wire diameter based on the melt parameters. The traction stretching model calculates the finished wire diameter and actual tension based on the die parameters and traction speed. Screw speed, traction speed and tension setpoints are selected as state variables, finished wire diameter and actual tension are selected as observation variables. At the current steady-state operating point, the implicit equations of the three-layer model are linearized by Taylor expansion, and the Jacobian matrix is ​​extracted as the parameter coupling constraint matrix. Fourier transform is performed on the wire diameter data in the calibration sequence stream data to extract the dominant frequency component as the wire diameter change frequency data; statistical distribution fitting is performed on the tension anomaly index sequence in the calibration sequence stream data to obtain tension anomaly distribution data; an initial evaluation matrix is ​​constructed by aligning with the time axis; dynamic feature weights are assigned to the wire diameter dimension and tension dimension according to the ratio of the absolute value of the influence of each state variable on the observed variable in the parameter coupling constraint matrix, and weighted fusion is used to generate a comprehensive anomaly evaluation matrix; The comprehensive anomaly assessment matrix is ​​graded and labeled according to the risk level, the corresponding root cause type code is embedded, and the physical position coordinates of the cable in the extrusion process corresponding to each time window are generated according to the integral mapping of traction speed and time.

8. The intelligent monitoring method for cable rubber wire production according to claim 1, characterized in that, Step S5 specifically includes: If the risk level in the risk distribution feature data reaches or exceeds the preset intervention level threshold, it is determined that the intervention trigger condition has been met. The objective function is to minimize the square mean of the difference between the measured wire diameter and the target wire diameter. The hard constraint is that the actual tension value is within the preset tension safety range. The soft constraint is the linear equation relationship between the increment of the observed variable and the increment of the state variable established by the parameter coupling constraint matrix. The boundary constraint is the equipment capability range of each adjustment amount. A multi-parameter collaborative optimization problem is constructed. The multi-parameter collaborative optimization problem is solved using a sequential quadratic programming algorithm to generate a collaborative adjustment vector that includes screw speed adjustment, traction speed adjustment, and tension setpoint adjustment. The coordinated adjustment vector, abnormal location coordinates, risk level, and root cause type are fused to generate a preliminary intervention signal.

9. The intelligent monitoring method for cable rubber wire production according to claim 1, characterized in that, Step S6 specifically includes: The coordinated adjustment vector in the preliminary intervention signal is obtained, and the screw speed adjustment, traction speed adjustment and tension setpoint adjustment are compared with the adjustment parameter range of the corresponding equipment. The adjustment amount that exceeds the boundary is truncated to the boundary value to obtain the first verification result. The adjustment amounts in the first verification result are combined into a parameter adjustment set. It is then checked whether the set satisfies the speed response coupling constraint between the tension controller and the traction motor driver, the temperature coupling constraint between the screw speed adjustment and the melt temperature change, and the motor acceleration and deceleration capability constraint. If not satisfied, the minimum correction amount is solved by the pseudo-inverse of the parameter coupling constraint matrix to perform iterative correction and obtain the final adjustment parameters. The execution sequence is determined based on the risk level, and the final adjustment parameters are written into the corresponding control registers of the extruder frequency converter, traction motor driver and tension controller via industrial fieldbus to execute equipment adjustment; After the parameters are updated, return to step S1 to collect data again, calculate the deviation between the actual feedback value and the set value in the final adjustment parameters, and repeat steps S1 to S6 to form a closed-loop monitoring and adjustment mechanism. When the tension fluctuation amplitude and the wire diameter fluctuation amplitude are both lower than the corresponding preset stable threshold and no new abnormal fluctuation range is generated for several consecutive cycles, it is determined that a stable production state has been obtained.

10. An intelligent monitoring system for the production of cable rubber wires, characterized in that, include: The data acquisition module collects tension, wire diameter, and die temperature data during the extrusion process, and generates calibration sequence stream data after filtering and alignment. The potential anomaly identification module is used to extract the tension anomaly index and material stress concentration characteristic value from the calibration sequence stream data in segments, calculate the correlation coefficient after temperature drift correction, and mark the potential anomaly fluctuation range when the correlation coefficient exceeds the preset correlation threshold and the die temperature is within the process allowable range. The knowledge graph diagnostic module is used to concatenate the tension anomaly index, material stress concentration characteristic value, die temperature and current working condition parameters of the potential abnormal fluctuation range into a query feature vector, inject it into the pre-constructed historical fault knowledge graph, and determine the risk level and root cause type of the potential abnormal fluctuation range. The risk distribution generation module is used to establish a parameter coupling constraint matrix based on the physical model of the extrusion process, and generate risk distribution feature data according to the parameter coupling constraint matrix, the risk level and the root cause type. The collaborative optimization decision module is used to perform multi-parameter collaborative optimization when the risk distribution characteristic data reaches the intervention trigger condition, with wire diameter stability as the objective function, tension safety range as the hard constraint, and the parameter coupling constraint matrix as the dynamic equation constraint, to generate a collaborative adjustment vector and generate a preliminary intervention signal. The verification and execution module is used to perform dual verification of the device adjustment parameter range and device compatibility rules on the preliminary intervention signal, then execute the device adjustment and return to trigger the data acquisition module to form a closed-loop monitoring and adjustment mechanism.

Images