Cable insulation layer eccentricity compensation control method based on laser detection

Abstract

This invention relates to a laser-based detection-based method for controlling the eccentricity of cable insulation layers, aiming to solve the problem of difficult eccentricity control caused by the unstable flow of molten polymer. The core scheme involves embedding a micro-stress sensor array in the extruder head to simultaneously collect multiple process parameters such as stress, temperature, pressure, and tension, integrating them into a multi-source raw dataset. A high spatiotemporal resolution dynamic stress field is reconstructed through spatiotemporal fusion, interpolation, and rheological correction. Furthermore, a physically meaningful three-dimensional stress fingerprint vector is extracted, and a response spectrum between stress and eccentricity evolution is constructed by combining historical calibration data, enabling real-time prediction of eccentricity trends. Based on the prediction results, servo-driven compensation adjustment is implemented in the extruder head to achieve dynamic feedforward control of concentricity, and model fine-tuning is performed based on laser detection feedback. This method improves the uniformity and process adaptability of insulation layer forming, and reduces the risk of eccentricity defects caused by fluctuations in operating conditions.

Description

Technical Field

[0001] This invention relates to the field of intelligent control and stress field physical modeling technology for cable extrusion processes, and in particular to a method for compensating for cable insulation eccentricity based on laser detection. Background Technology

[0002] In the field of intelligent control for cable extrusion processes, the mainstream methods for monitoring and compensating for insulation layer eccentricity generally employ feedback closed-loop control strategies based on direct measurement signals. Existing technologies typically use an external laser detection module to scan the insulation layer thickness in real time, calculating the eccentricity direction and amplitude. Then, using techniques such as PID control, expanded state variable methods, or pre-aiming distance and curvature calculation, the measured eccentricity data is fed back to the extruder head to drive the servo mechanism for adjustment. Some advanced technologies attempt to introduce black-box modeling methods such as neural network fitting, state observers, and RBF (Rapid Reduction Function) to predict and compensate for eccentricity evolution. Other multi-source data fusion technologies use indirect parameters such as temperature, pressure, and tension as feature inputs to optimize the extruder head position adjustment effect. The overall industry development trend is characterized by: improved detection accuracy, faster response speed, intelligent compensation algorithms, multi-source data fusion, and continuously enhanced model adaptive capabilities.

[0003] Previous representative technologies mostly relied on the integration of wall thickness laser detection and servo control to automatically adjust the eccentricity during cable forming. These solutions have good steady-state response under standard process conditions, but when there is a lack of direct eccentricity signals or when encountering complex process fluctuations, the mapping relationship between indirect parameters and eccentricity is unclear, making it difficult to form high-precision feedforward prediction and compensation. Methods based on mainstream PID control structures, fitting algorithms, and state observers can fit certain causal relationship curves through historical data, but they are essentially statistical correlation paradigms that ignore the physical intrinsic state correlation characteristics of the material forming process. They are easily affected by process disturbances, and the model's generalization ability and physical interpretability are limited.

[0004] Existing technologies suffer from the following prominent problems: First, the mapping relationship between indirect process parameters (such as die head temperature, melt pressure, and traction tension) and cable insulation eccentricity is complex and constrained by material composition, flow channel structure, and melt rheological properties, making it difficult to accurately describe using traditional mathematical formulas or statistical models. Furthermore, the industry generally lacks mechanisms to transform multi-source process data into traceable physical state representations, resulting in compensation control relying solely on reverse feedback signals and lacking foresight. Second, black-box prediction methods (such as neural networks, RBF, and LSTM) struggle to explain the physical causes of process state changes and eccentricity evolution, hindering on-site process optimization and anomaly analysis. Third, existing compensation modules mostly only adjust after eccentricity detection anomalies, failing to incorporate real-time evolution of the melt's internal mechanical state for proactive prediction and coordinated compensation, thus limiting response speed and robustness.

[0005] Therefore, there is an urgent need in this field for a novel technical solution that can move beyond the framework of indirect parameter statistical correlation and establish a physically traceable correlation between process parameters and melt stress fields. This solution should be able to accurately map multi-source process parameters to the dynamic stress field distribution of the melt without direct measurement signals. Through physical field fingerprint feature extraction, it should generate interpretable eccentricity trend prediction results, thereby driving die head linkage compensation. This technology can not only greatly improve the accuracy and response speed of eccentricity feedforward compensation, but also provide engineering operability and physical reliability for the digital and intelligent control of cable extrusion processes, offering a higher level of theoretical and applied foundation for industry process optimization and intelligent manufacturing upgrades. Summary of the Invention

[0006] This application provides a laser-based method for compensating and controlling the eccentricity of cable insulation, which aims to solve one of the problems or issues of the prior art mentioned in the background section.

[0007] The cable insulation eccentricity compensation and control method based on laser detection provided in this application specifically includes: S1: Acquire circumferential micro-zone stress response signals and axial micro-zone stress response signals at several key cross-sections of the cable extruder head, and simultaneously collect real-time melt temperature gradient data, melt pressure distribution data and traction tension data to form a multi-source process parameter raw dataset.

[0008] S2: Based on the original dataset of the multi-source process parameters, construct a two-dimensional dynamic stress field snapshot sequence with high spatiotemporal resolution, and map the discrete sensing signals into the dynamic stress field distribution characteristics of the melt in the continuous spatial domain.

[0009] S3: Perform principal component-topological feature fusion encoding on the dynamic stress field distribution characteristics of the melt to extract the polarity angle distribution of the circumferential stress gradient, the coordinate position of the abrupt change point of the radial stress attenuation slope, and the spatial offset of the stress standing wave nodal line, which reflect the asymmetry of melt flow, and generate a three-dimensional stress fingerprint vector with physical meaning.

[0010] S4: By utilizing the mapping relationship between different three-dimensional stress fingerprint vectors calibrated in historical qualified batch data and the actual eccentricity direction and eccentricity amplitude obtained by the laser detection module, a table-based stress-eccentricity response spectrum is constructed, and a physical traceable correlation model between indirect parameters and core state variables is established.

[0011] S5: Input the currently generated three-dimensional stress fingerprint vector as an index into the stress-eccentricity response map, retrieve the nearest response entry, and output the eccentricity evolution prediction result containing the predicted direction and degree of eccentricity trend.

[0012] S6: Based on the current extrusion line speed of the cable extruder head and the die structure parameters, the predicted direction of the eccentricity trend in the eccentricity evolution prediction result is converted into a compensation displacement vector in the extruder head plane coordinate system, and a servo drive target position correction command is generated.

[0013] S7: Drive the extruder head to perform linkage adjustment action according to the target position correction command of the servo driver, change the spatial distribution of melt extrusion, so as to counteract the predicted eccentricity trend of the insulation layer and realize feedforward compensation control.

[0014] S8: Determine whether the deviation between the actual eccentricity data fed back by the laser detection module and the eccentricity evolution prediction result continues to exceed the preset dynamic tolerance threshold. If the condition is met, trigger the fingerprint model online fine-tuning process to update the local response region of the stress-eccentricity response spectrum.

[0015] The cable insulation eccentricity compensation and control method based on laser detection provided in this application has the following beneficial effects: (1) The feedforward compensation paradigm proposed in this application effectively overcomes the problems of poor generalization ability and weak adaptability to working conditions caused by the reliance on statistical correlation modeling in the traditional cable insulation layer eccentricity control. By starting from the physical essence of material forming, an interpretable causal link of "process parameters → melt stress field → eccentricity evolution" is constructed, which significantly improves the physical credibility and process transparency of the control strategy. Compared with the existing methods that only fit the black box relationship between indirect parameters and eccentricity based on historical data, this scheme is based on the acquisition of high spatiotemporal resolution dynamic stress field snapshots by embedded stress sensing array. Combined with the fusion of multi-source information of temperature, pressure and tension, it realizes the refined perception of melt flow state. This allows the originally unmeasurable eccentricity trend to be characterized by a three-dimensional stress fingerprint vector with clear mechanical meaning. Thus, without introducing any neural network or complex observer, it realizes the real-time prediction of the eccentricity direction and amplitude evolution trend, which greatly improves the response sensitivity and prediction accuracy of feedforward compensation.

[0016] (2) By establishing an offline calibrated stress-eccentric response spectrum and combining it with an online nearest neighbor retrieval mechanism, this scheme realizes a lightweight operation mode that can complete local knowledge updates without global model retraining, which significantly reduces the computational overhead of the algorithm and the difficulty of engineering deployment. It is especially suitable for industrial scenarios with multiple varieties and small batches of continuous production. Furthermore, the adaptive spatial mapping module dynamically calculates the compensation displacement vector according to the current linear velocity and the mold geometry, and directly drives the servo system to adjust the position of the extruder head, avoiding the lag and oscillation risks in traditional PID feedback control, making the compensation action more forward-looking and time-matched. At the same time, the closed-loop verification mechanism only triggers the local fine-tuning of the fingerprint model when the detection deviation continues to exceed the limit, which not only ensures the stability of the system in the long-term operation, but also prevents false updates caused by noise interference, realizing the coordination and unity of model evolution capability and robustness. Without adopting complex structures such as RBF approximation, LSTM prediction, and aiming control, the technical goal of high-performance feedforward control is achieved.

[0017] (3) This technical approach transforms the eccentricity control problem into a traceable stress field morphological analysis task, fundamentally changing the previous passive control paradigm that relied on indirect variable correlation or pure data-driven approaches. It constructs an observable, representable, and adjustable full-process control closed loop, which not only enhances the interpretability and fault diagnosis capabilities of the system but also provides a reliable physical basis for subsequent process optimization. Since the entire process is based on a micro-sensor array and a lightweight lookup table inference architecture, it has good real-time performance and low latency characteristics, meeting the millisecond-level response requirements in high-speed extrusion scenarios. In addition, the framework has moderate requirements for equipment modification and is easy to integrate into existing production line control systems, possessing strong scalability and engineering implementation value. In summary, this solution, without relying on the technical means listed in the circumvention content, guides feature construction through physical mechanisms and uses stress fingerprints as a bridge to achieve pre-identification and accurate compensation of eccentricity trends. It effectively solves the contradiction between dynamic adaptability, control accuracy, and implementation cost in traditional methods, providing a new solution for high-quality cable manufacturing that combines scientific depth and engineering practicality. Attached Figure Description

[0018] Figure 1 This is the main flowchart of a cable insulation eccentricity compensation control method based on laser detection.

[0019] Figure 2 This is a sub-flowchart of a cable insulation eccentricity compensation control method based on laser detection.

[0020] Figure 3 This is another sub-flowchart of the cable insulation eccentricity compensation control method based on laser detection. Detailed Implementation

[0021] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0022] The following disclosure provides many different embodiments or examples for implementing different structures of the invention. To simplify the disclosure, specific examples of components and arrangements are described below. Of course, these are merely examples and are not intended to limit the invention. Furthermore, reference numerals and / or letters may be repeated in different examples; such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0023] like Figure 1 As shown, this application provides a cable insulation eccentricity compensation and control method based on laser detection, specifically including: S1: Acquire circumferential micro-zone stress response signals and axial micro-zone stress response signals at several key cross-sections of the cable extruder head, and simultaneously collect real-time melt temperature gradient data, melt pressure distribution data and traction tension data to form a multi-source process parameter raw dataset.

[0024] S2: Based on the original dataset of the multi-source process parameters, construct a two-dimensional dynamic stress field snapshot sequence with high spatiotemporal resolution, and map the discrete sensing signals into the dynamic stress field distribution characteristics of the melt in the continuous spatial domain.

[0025] S3: Perform principal component-topological feature fusion encoding on the dynamic stress field distribution characteristics of the melt to extract the polarity angle distribution of the circumferential stress gradient, the coordinate position of the abrupt change point of the radial stress attenuation slope, and the spatial offset of the stress standing wave nodal line, which reflect the asymmetry of melt flow, and generate a three-dimensional stress fingerprint vector with physical meaning.

[0026] S4: By utilizing the mapping relationship between different three-dimensional stress fingerprint vectors calibrated in historical qualified batch data and the actual eccentricity direction and eccentricity amplitude obtained by the laser detection module, a table-based stress-eccentricity response spectrum is constructed, and a physical traceable correlation model between indirect parameters and core state variables is established.

[0027] S5: Input the currently generated three-dimensional stress fingerprint vector as an index into the stress-eccentricity response map, retrieve the nearest response entry, and output the eccentricity evolution prediction result containing the predicted direction and degree of eccentricity trend.

[0028] S6: Based on the current extrusion line speed of the cable extruder head and the die structure parameters, the predicted direction of the eccentricity trend in the eccentricity evolution prediction result is converted into a compensation displacement vector in the extruder head plane coordinate system, and a servo drive target position correction command is generated.

[0029] S7: Drive the extruder head to perform linkage adjustment action according to the target position correction command of the servo driver, change the spatial distribution of melt extrusion, so as to counteract the predicted eccentricity trend of the insulation layer and realize feedforward compensation control.

[0030] S8: Determine whether the deviation between the actual eccentricity data fed back by the laser detection module and the eccentricity evolution prediction result continues to exceed the preset dynamic tolerance threshold. If the condition is met, trigger the fingerprint model online fine-tuning process to update the local response region of the stress-eccentricity response spectrum.

[0031] Step S1 involves acquiring circumferential and axial micro-area stress response signals at several key cross-sections of the cable extruder head, and simultaneously collecting real-time melt temperature gradient data, melt pressure distribution data, and traction tension data to form a multi-source process parameter raw dataset. The key cross-sections refer to specific locations selected along the axial direction of the cable extruder head that have a decisive influence on the uniformity of the insulation layer extrusion. These cross-sections cover the critical process stages from the melt entering the die head to extrusion molding, including but not limited to: the melt inlet cross-section (CS1), the end cross-section of the flow divider cone (CS2), and the front section of the molding zone (CS3). By deploying a sensor array within the annular region of these key cross-sections, the synchronous capture of multi-dimensional stress signals affecting the insulation layer's eccentricity is achieved. Step S1 specifically includes: S1.1: Embedded micro-stress sensor arrays are deployed in the annular distribution area of ​​key cross-sections (e.g., CS1, CS2, CS3) of the cable extruder head. Multiple piezoelectric micro-area stress-sensitive units are uniformly arranged circumferentially on the inner wall of the extruder head, generating circumferential and axial micro-area stress response signal acquisition channels with spatial resolution. The sensor arrays for each key cross-section are deployed in the same way, but their acquired signals collectively constitute a multidimensional dataset characterizing the stress evolution of the melt throughout the entire process from inlet to forming.

[0032] When implementing embedded micro-stress sensor array deployment in the annular distribution area of ​​the key cross-section of the cable extruder head, the circumferential geometry of the inner wall of the extruder head is used as a spatial arrangement reference. Input conditions include the circumferential coordinate data of the inner wall of the extruder head, the material properties of the extruder head, and the number and specifications of available piezoelectric micro-area stress-sensitive units. An equidistant segmentation calculation is performed on the circumferential coordinate data, dividing the circumferential inner wall of the extruder head into several equal-length regions. These regions are then used as a reference to determine the installation center position of each piezoelectric micro-area stress-sensitive unit. The base material of the piezoelectric unit is matched using material properties to ensure no significant damping mismatch in the stress transmission path. Embedding grooves are prefabricated at designated positions through machining to fix the piezoelectric sensitive units. Positioning fixtures are applied to each embedding groove, and insertion and contact surface pre-tightening treatment of the piezoelectric micro-area stress-sensitive unit is performed to eliminate installation gaps and prevent signal drift caused by micro-vibrations during operation. The output of the piezoelectric unit is introduced into two types of acquisition channels: a circumferential acquisition channel to capture stress change signals along the circumferential direction of the inner wall of the extruder head, and an axial acquisition channel to capture stress change signals along the longitudinal direction of the extruder head. Shielding and grounding measures are introduced during the wiring process to suppress electromagnetic interference. Through the above layout and channel construction processing, the spatial arrangement result is transformed into micro-area stress response signal acquisition channels with spatial resolution capabilities, corresponding to circumferential and axial directions respectively, realizing the hardware prerequisites for multi-directional stress measurement.

[0033] For example, in a diameter of The circumferential length of the inner wall of the extruder head, measured in millimeters, is determined using a CAD model to be [missing information]. millimeters, at intervals A piezoelectric micro-region stress-sensitive element is arranged in millimeters, resulting in an element count of... Each unit's base is made of alloy steel similar to the head material to reduce stress wave transmission loss at the interface. Before installation, the surface roughness of the base-head contact surface is machined to a certain degree. Micrometers. The embedding groove depth is set to the thickness of the sensitive cell. To ensure sufficient pre-tightening during installation, the circumferential output of the sensitive unit is connected to the circumferential acquisition channel, and the axial output is connected to the axial acquisition channel. The wiring uses twisted-pair shielded cable and is grounded at both ends of the receiver. After deployment, vibration tests were conducted to verify the sensitivity and signal noise levels of each channel. The results showed that the stress signal amplitude of the circumferential channel remained stable within the design range, and the response delay of the axial channel was significantly reduced, meeting the technical requirements for high spatiotemporal resolution measurement.

[0034] S1.2: Based on the circumferential micro-area stress response signal acquisition channel and the axial micro-area stress response signal acquisition channel, high-frequency synchronous sampling and timing alignment processing are performed to eliminate the phase delay difference between different sensing nodes and generate a discretized circumferential micro-area stress response signal sequence and a discretized axial micro-area stress response signal sequence with a unified timestamp reference.

[0035] With spatially resolvable circumferential and axial micro-area stress response signal acquisition channels already established, the analog voltage signals output by each sensing node are converted into digital signals using a high-performance data acquisition card. A unified hardware trigger source is used at the acquisition card as the sampling start condition to ensure the synchronization of sampling actions across channels. A unified sampling frequency is set for the multi-channel digital signals obtained through synchronous sampling. The sampling frequency is set to be at least twice the highest characteristic frequency of the transient stress fluctuation in the melt to meet the bandwidth requirements of the Nyquist sampling theorem and avoid aliasing effects. The digital signals are sent to the central processing unit using a multi-channel parallel DMA transmission mode. A high-precision timestamp output from a clock module is added to each sampling packet, with a timestamp accuracy better than microseconds, serving as a reference for subsequent timing alignment. By comparing the timestamps acquired by each sensing node, the phase delay difference is calculated and a delay vector is formed. The delay difference is calculated using the following formula: in For a certain node's timestamp, The reference node timestamp is used. High-precision interpolation resampling is performed based on the delay vector to remap the signals of each node to a unified time reference, eliminating phase misalignment caused by sensor response lag, and a phase-adjustable FIR filter is used to compensate for spectral distortion in the interpolation process. Through the above processing method, the sensor array deployment results of the previous step are transformed into discretized circumferential micro-area stress response signal sequences and discretized axial micro-area stress response signal sequences with a unified timestamp reference, realizing the temporal consistency and high-frequency response matching of stress data from multiple sensor nodes.

[0036] For example, on a cable extrusion production line, 16 piezoelectric stress sensing units arranged circumferentially and 8 piezoelectric stress sensing units arranged axially are all connected to a PCIe data acquisition card supporting 32-channel parallel acquisition, with a sampling frequency set to 250kHz. The hardware trigger signal is provided by a photoelectric trigger installed in the extruder head, and the trigger signal is synchronized to all channels. The timestamp is provided by a high-precision clock module calibrated with GPS, with an accuracy of 0.5 microseconds. During the acquisition process, the reference node is selected as the first circumferential sensor, and the delay difference between the fifth circumferential sensor and the reference node is calculated as... Seconds, after interpolation and resampling, the phase error is less than The spectrum was then corrected using a phase-compensated FIR filter, with the filter order set to 128 and a passband range of 0–80 kHz. In this scenario, the processed discretized circumferential and axial micro-region stress signals have the same time reference and sampling points, ensuring temporal consistency when subsequently fused with multi-source data such as temperature, pressure, and traction tension, significantly improving the accuracy and stability of stress field reconstruction.

[0037] S1.3: Using a thermocouple array and pressure transmitter arranged along the extrusion channel, thermodynamic state data during the melt flow process are captured in real time to extract radially distributed temperature change rate data and axially distributed pressure fluctuation data, generating a real-time melt temperature gradient data set and a melt pressure distribution data set.

[0038] Initialize and verify the sensor coordinates and sampling configuration of the thermocouple array and pressure transmitter already arranged along the extrusion channel to ensure that each measuring point is in a critical process position and has stable sampling capability.

[0039] The output electromotive force signal of the thermocouple array is synchronously sampled at the millisecond level, the original voltage value is converted into the corresponding temperature value, and the temperature difference between each radially adjacent measuring point is calculated based on the radial position index.

[0040] A radial temperature change rate parameter is constructed using temperature difference and radial spacing data, and the temperature gradient is calculated using the following formula: in, Let i be the temperature value at radial position i. The temperature value at radial position i+1 is... This represents the radial spacing.

[0041] Phase alignment processing is performed on the sampling data of the pressure transmitter arranged along the axial direction, the instantaneous pressure values ​​of different measuring points are arranged in a consistent manner according to the timestamp, and the pressure difference between adjacent measuring points along the axial direction is calculated.

[0042] Pressure fluctuation parameters are derived using pressure difference and axial spacing data, and the pressure gradient is calculated using the following formula: in, This represents the pressure value at axial position j. This represents the pressure value at the axial position j+1. This represents the axial spacing.

[0043] The radial temperature change rate sequence was constructed into a real-time melt temperature gradient data set, and the axial pressure gradient sequence was constructed into a melt pressure distribution data set. Timestamps and spatial index labels were added to each data set.

[0044] By combining thermocouple arrays and pressure transmitters for acquisition and gradient derivation processing, the synchronous stress data from the previous step is converted in parallel into a set of real-time melt temperature gradient data and a set of melt pressure distribution data that characterize the thermodynamic state of melt flow, thereby achieving non-invasive high spatiotemporal resolution process status monitoring.

[0045] For example, in a high-precision cable extrusion production line, four thermocouple measuring points are arranged in the radial direction, with a spacing of [missing information]. The temperature values ​​collected in each sampling cycle are as follows: , , , Degrees Celsius. In the calculation of the radial temperature change rate, the first temperature gradient is... = ℃ / mm, the second segment is = ℃ / mm. Three pressure transmitters are arranged axially, with a spacing of [missing information]. The pressures during the sampling period are as follows: , , MPa, the first pressure gradient is = MPa / mm, the second segment is = MPa / mm. Under these conditions, the temperature gradient set and pressure distribution set generated are updated within millisecond time resolution and transmitted to the stress field reconstruction module, significantly improving the thermodynamic accuracy and dynamic response speed of the fluid dynamics state model.

[0046] S1.4: Based on the torque sensor output at the traction machine drive end, calculate the longitudinal force characteristics under the current production line operation state to isolate mechanical transmission noise and extract the effective load component, and generate a scalar of traction tension data characterizing the cable forming resistance.

[0047] The traction machine drive end is the core equipment providing linear traction force on the cable insulation extrusion production line. Based on the torque sensor output signal of the traction machine drive end, bandpass filtering and fast Fourier transform operations are performed on the real-time sampled data of the sensor to isolate the effective spectral components located outside the natural frequency of the traction system structure. These effective spectral components are input into a mechanical transmission noise stripping model. This model constructs a noise feature library based on the traction machine gear ratio, bearing vibration mode, and motor electromagnetic force fluctuation characteristics. Frequency domain matching operations are used to identify and eliminate noise components matching the aforementioned feature library. Steady-state window mean filtering is performed on the noise-stripped torque signal to suppress high-frequency random fluctuations and retain the low-frequency mechanical response reflecting the load change trend of the traction machine. Based on the torque signal output from the steady-state window mean filter, longitudinal force calculation is performed according to the mechanical efficiency coefficient of the transmission chain and the traction roller radius parameters. The traction tension is calculated using the following formula: in, For traction tension data scalar, This is the steady-state torque value after noise stripping. The radius of the traction roller is... This is the mechanical efficiency coefficient of the transmission chain. The calculated traction tension data scalar is stored in the corresponding field of the original dataset of multi-source process parameters. Through the above processing method, the torque sensor output of the previous step is converted into a longitudinal force index with stripping noise and physical accuracy, thereby realizing the effective quantification of cable forming resistance.

[0048] For example, a strain gauge torque sensor with a range of 500 N·m and an accuracy of 0.1%FS is installed at the drive end of the traction machine, and the sampling frequency is set to 2 kHz. The filtering bandwidth is selected from 5 to 150 Hz to avoid the gear meshing frequency of 35 Hz and the fundamental frequency of the motor's electromagnetic force of 50 Hz. In the noise stripping model, the gear ratio is set to 3.5, the mechanical efficiency coefficient is set to 0.92, and the radius of the traction roller is accurately measured to be 0.25 m. After the sampled data is frequency-domain matched to strip the gear meshing peak and electromagnetic force harmonic peak, the steady-state window length is set to 2 s for mean filtering to obtain the steady-state torque value. N·m. Substituting this torque value into the formula: The calculation result is N is written as a scalar value of traction tension into the original dataset of multi-source process parameters. In actual production verification, this tension value can stably reflect the change in insulation layer forming resistance, and the quality of the input signal of the corresponding eccentricity prediction model is significantly improved, and the response speed of the feedforward compensation control is greatly improved.

[0049] S1.5: Perform multi-source data spatiotemporal fusion encapsulation processing on the discretized circumferential micro-region stress response signal sequence, discretized axial micro-region stress response signal sequence, real-time melt temperature gradient data set, melt pressure distribution data set, and traction tension data scalar to construct a structured data package containing complete process state dimensions and generate a multi-source process parameter raw dataset.

[0050] Step S2: Based on the original dataset of multi-source process parameters, a high spatiotemporal resolution two-dimensional dynamic stress field snapshot sequence is constructed to map discrete sensing signals into the dynamic stress field distribution characteristics of the melt in a continuous spatial domain. Specifically, this includes: S2.1: Perform timestamp alignment processing on the circumferential micro-region stress response signal, axial micro-region stress response signal, real-time melt temperature gradient data, melt pressure distribution data and traction tension data in the original dataset of multi-source process parameters to eliminate phase deviation caused by the difference in sampling clocks of multiple sensors and generate synchronized multi-source sensor data frames with a unified time reference.

[0051] For the circumferential micro-region stress response signal, axial micro-region stress response signal, real-time melt temperature gradient data, melt pressure distribution data and traction tension data in the original dataset of multi-source process parameters, a time index record of sensor sampling events is established to identify the sampling timestamp differences of various signals.

[0052] Based on time index records, the offset of sampling clocks of different sensors is calculated. A reference clock source is used as a global reference. The sampling time of each signal is adjusted by interpolation and resampling algorithms to match the global time reference.

[0053] For the adjusted signal, phase difference calculation is performed. The residual phase deviation is confirmed by comparing the peak positions of the periodic response components, and the deviation value is input into the phase compensator for correction.

[0054] During the phase compensation process, a linear phase filter is used to achieve time domain shifting, ensuring that each signal corresponds to the same physical event at a unified moment, thus avoiding spatial mapping distortion during multi-source data fusion.

[0055] A time-series normalization operation is performed on the signal sequence processed with a unified time base to keep the time index interval of each synchronized data frame constant, which facilitates the subsequent continuous mapping calculation of the two-dimensional dynamic stress field.

[0056] The above processing method transforms the original dataset of multi-source process parameters from the previous step into a synchronized multi-source sensor data frame with a unified time reference, achieving temporal consistency of multi-source heterogeneous data and providing stable input for subsequent spatial interpolation and physical characteristic compensation.

[0057] For example, on a high-precision cable extrusion production line, the sampling frequency of the circumferential micro-stress sensor is configured at 500Hz, the axial micro-stress sensor at 480Hz, the thermocouple array at 200Hz, the pressure transmitter at 250Hz, and the traction tension sensor at 100Hz. By establishing a time index record, the average clock offset between the circumferential and axial sensors was measured to be... The average clock offset between the thermocouple array and the pressure transmitter is in milliseconds. The average clock offset of the traction tension data in milliseconds is Milliseconds. Using a reference clock source of 500Hz, each signal is resampled to a uniform sampling frequency through linear interpolation, and the phase difference is calculated using the formula. ,in This is the residual time deviation. Given the signal period, the phase difference of the traction tension signal is calculated as follows: The degree is zeroed after shift compensation using a linear phase filter. The time index interval of the processed synchronized multi-source sensor data frame is stabilized at... The results showed that in the subsequent construction of the two-dimensional stress field, the temporal matching error was significantly reduced and the spatial mapping accuracy was significantly improved, ensuring the input quality of the eccentricity trend prediction model.

[0058] S2.2: Based on the sensor physical coordinate layout information in the synchronized multi-source sensing data frame, perform Kriging spatial interpolation algorithm processing to map the stress measurement values ​​of discrete points into a continuous gridded stress distribution matrix covering the key cross section of the extruder head, and generate preliminary stress field grid data containing spatial location information.

[0059] Based on the sensor physical coordinate layout information in the synchronized multi-source sensing data frames, the set of scalar coordinates of the measurement points corresponding to the circumferential and axial stress response signals and their distribution mapping relationship on the key cross-section of the extruder head are determined. This set of positional coordinate scalars and the corresponding stress measurements form a paired dataset, which serves as the initial input matrix for the spatial interpolation algorithm. A two-dimensional coordinate domain model of the extruder head cross-section is constructed, and the semivariogram type, scale parameter, and smoothing factor for Kriging spatial interpolation are set. The semivariogram values ​​are calculated based on the actual distance between the sensors. The Kriging estimation formula is then used. in, For interpolation estimates, Let i be the spatial coordinates of the i-th sensor node. The Kriging weighting coefficients are determined by solving the Kriging equations, satisfying the unbiasedness and minimum variance conditions. The interpolation results are mapped to a regular grid region covering the key cross-section of the extruder head, forming a gridded matrix data containing the spatial coordinates of each grid center and the corresponding interpolated stress value. The gridded matrix undergoes data integrity checks, eliminating outlier interpolation points and compensating for missing boundary regions, generating preliminary stress field grid data containing spatial location information. Through Kriging spatial interpolation, the synchronized multi-source sensor data frames from the previous step are transformed into a regular gridded stress distribution matrix in a continuous spatial domain, achieving spatial continuity modeling of discrete measurements.

[0060] For example, on the cross-section of the cable extruder head, 16 piezoelectric stress-sensitive elements are arranged circumferentially, and 8 sensing nodes are arranged axially, forming a two-dimensional coordinate set of a total of 128 measurement points, with the coordinate unit being millimeters. The synchronized multi-source sensing data frame contains the stress value of each measurement point, in MPa, with a measured range of 3.5 to 5.2 MPa. A two-dimensional coordinate domain model with a cross-sectional radius of 50 mm is constructed, using a spherical semivariogram, with a scale parameter set to 30 mm and a smoothing factor set to 0.1. The distance between any two nodes is calculated, and the semivariogram value is derived accordingly. A Kriging equation system is established, and the weight coefficients are solved. The interpolated estimates are calculated within a radius of 0 to 50 mm, with a grid resolution of 1 mm, generating a regular coordinate grid matrix with a diameter of 100 mm. Each grid center corresponds to the interpolated stress value. Four anomalous interpolation points are detected in the boundary region, with stress values ​​exceeding the maximum measured value by 0.5 MPa. After performing boundary compensation calculations, the anomalous values ​​are replaced. The final output is a preliminary stress field mesh data containing 5025 mesh nodes, with spatial continuity meeting the accuracy requirements for subsequent viscosity-temperature coupling correction steps.

[0061] S2.3: Using the real-time melt temperature gradient data in the preliminary stress field grid data, construct the melt viscosity-temperature coupling correction coefficient field, and perform non-uniform medium rheological property compensation calculation on the stress values ​​in the preliminary stress field grid data to eliminate the influence of material modulus changes caused by temperature fluctuations on stress readings, and generate rheologically corrected dynamic stress field distribution data.

[0062] Based on the real-time melt temperature gradient data in the preliminary stress field grid data, a viscosity-temperature coupling correction coefficient field directly corresponding to the melt temperature change is established to quantify the dynamic adjustment factor of the material modulus at different spatial locations as it deviates from the reference value with temperature.

[0063] The temperature gradient data is mapped to the stress field grid coordinate system according to the correspondence between grid points. Node-level interpolation matching is performed to generate the instantaneous temperature deviation scalar value of each grid node, which serves as the basic input for subsequent rheological compensation.

[0064] Based on the material's viscosity-temperature characteristic curve, a coefficient is calculated for each temperature deviation scalar value using an empirical formula: in, For correction factor, Viscosity-temperature sensitivity coefficient Let be the temperature deviation of the i-th grid node.

[0065] The correction coefficient field is multiplied by the initial stress field grid data to obtain the stress numerical matrix after temperature fluctuation compensation.

[0066] A non-uniform medium rheological property correction factor field is introduced for the compensated stress numerical matrix. A second-order correction model is constructed based on the nonlinear relationship between melt viscosity and temperature, and a dual-coefficient iterative adjustment is performed to ensure that the corrected stress data simultaneously meets the dual constraints of material modulus and viscosity variation.

[0067] The output is rheologically corrected dynamic stress field distribution data, providing an accurate input basis for subsequent physical consistency smoothing filtering.

[0068] By constructing a viscosity-temperature coupling correction coefficient field and performing dot product compensation processing, the preliminary stress field mesh data generated in the previous step is transformed into rheologically corrected dynamic stress field distribution data that takes into account both the elimination of temperature fluctuation effects and the compensation for non-uniform rheological properties of materials, thereby improving the physical accuracy of stress measurement.

[0069] For example, on a critical section of the extruder head in a high-precision cable production line, the initial stress field mesh data has a size of 50×50 nodes, with a sampling frequency of 500Hz per node. Temperature gradient data is collected by eight thermocouples distributed radially, with a reference temperature set at 220℃. The material viscosity-temperature sensitivity coefficient α is experimentally measured to be 0.005 MPa / ℃. At a certain moment, the temperature deviation ΔTi of the i-th node is 5℃, and the correction factor is calculated as follows. The result was 1.025. Multiplying this coefficient by the initial stress value of 40 MPa yielded a compensated stress value of 41.0 MPa. In the full-field calculation, the stress value at the node with the largest temperature deviation was significantly increased after compensation compared to the initial value, while the stress value at the node with the lowest temperature deviation was slightly decreased, forming a spatially resolved correction coefficient field. This result, after adjustment using a second-order rheological correction model, showed a significant reduction in the noise peak of the rheologically corrected dynamic stress field distribution. The structural characteristics better matched the actual melt flow, and the performance of subsequent smoothing and filtering was greatly improved.

[0070] S2.4: Based on the rheologically corrected dynamic stress field distribution data, the melt continuity equation constraint condition is introduced to perform physical consistency smoothing filtering to suppress high-frequency measurement noise and ensure the divergence conservation characteristics of the stress field vector field, thereby generating a smoothed dynamic stress field sequence that conforms to the basic laws of fluid mechanics.

[0071] S2.5: The smoothed dynamic stress field sequence is sliced ​​and encapsulated according to a preset high-frequency sampling period to discretize the stress evolution process in the continuous time domain into a series of two-dimensional image data units with independent spatiotemporal labels, thereby generating the final high spatiotemporal resolution two-dimensional dynamic stress field snapshot sequence.

[0072] like Figure 2 As shown, step S3 involves performing principal component-topological feature fusion encoding on the dynamic stress field distribution characteristics of the melt to extract the polarity angle distribution of the circumferential stress gradient, the coordinates of the abrupt change points of the radial stress attenuation slope, and the spatial offset of the stress standing wave nodal lines, reflecting the asymmetry of melt flow, thereby generating a physically meaningful three-dimensional stress fingerprint vector. Specifically, this includes: S3.1: Perform spatial domain discretization meshing processing on the high spatiotemporal resolution two-dimensional dynamic stress field snapshot sequence to transform the continuously distributed dynamic stress field distribution characteristics of the melt into a set of regularly arranged micro-region stress tensor matrices, so as to obtain a spatially discretized stress field dataset containing circumferential micro-region stress response signals and axial micro-region stress response signals, which serves as the standardized input object for subsequent dimensionality reduction processing.

[0073] For the high spatiotemporal resolution two-dimensional dynamic stress field snapshot sequence input, regularized grid division parameters are set in the spatial domain, including grid cell side length, number of circumferential divisions and number of axial divisions, to ensure that the physical size of each grid cell matches the sampling accuracy of the sensor array.

[0074] Based on the set meshing parameters, the continuously distributed dynamic stress field snapshot sequence of the melt is discretized into segments according to both circumferential and axial directions, ensuring that the stress vector corresponding to each segment covers the corresponding physical space range.

[0075] The discretized data is processed by coordinate mapping, which maps the spatial position index of each pixel in the snapshot sequence to the physical coordinates of the corresponding grid cell, forming a set of regularly arranged micro-region stress tensor matrices, where each matrix element contains circumferential stress components and axial stress components.

[0076] Unit normalization is performed on the stress components of each element in the set of micro-region stress tensors to unify the circumferential and axial micro-region stress response signals from different sources into a numerical domain with consistent dimensions, thus avoiding weight imbalance caused by scale differences in subsequent feature analysis.

[0077] Structured label encoding is applied to the normalized set of micro-region stress tensor matrices, embedding grid location information, sampling timestamp information, and sensor source identification information into the matrix data structure to form a spatially discretized stress field dataset, which serves as a standardized input object for principal component analysis and topological feature extraction.

[0078] By using spatial domain discretization meshing, the continuous spatial domain dynamic stress field snapshot from the previous step is transformed into field data with regular topological structure and unified physical dimensions, thus achieving the transformation of stress field characteristics from continuous distribution to structured matrix.

[0079] For example, in the cable extruder head section radius of... millimeters, number of circumferential sensing nodes is The number of axial nodes is In each production scenario, a high spatiotemporal resolution two-dimensional dynamic stress field snapshot sequence is arranged circumferentially. degrees, axial direction per The mesh is divided into millimeters. The resulting micro-region stress tensor matrix has a size of [value missing]. Each matrix element contains a circumferential stress value. With axial stress value The numerical unit is uniformly set in megapascals (MPa). For MPa to Stress values ​​in the MPa range are normalized and mapped to... The normalization formula for the interval is: ,in This represents the current stress value of the mesh element. and These represent the minimum and maximum values ​​of the stress components of all elements in the matrix, respectively. After the above processing, a spatially discretized stress field dataset containing location coordinates, timestamps, and sensor source identifiers is formed. This dataset can significantly improve the stability of feature extraction in subsequent principal component analysis and maintain the traceability of physical quantities in the topological feature fusion encoding stage.

[0080] S3.2: Based on the spatially discretized stress field dataset, perform covariance matrix construction and eigenvalue decomposition operations, and use principal component analysis algorithm to extract the preceding principal component load vectors whose cumulative contribution rate exceeds a preset threshold. Compress the high-dimensional redundant micro-region stress tensor matrix set into a low-dimensional principal component score coefficient sequence to eliminate collinear interference between multi-source process parameters and retain the main energy distribution characteristics of melt flow.

[0081] Based on the stress tensor matrices of each micro-region in the spatially discretized stress field dataset, a covariance matrix is ​​constructed among the variables to comprehensively characterize the statistical correlation between circumferential and axial stress components. Perform eigenvalue decomposition on the covariance matrix to obtain each eigenvalue and its corresponding eigenvector, thereby quantifying the variance contribution of stress modes in different directions. Based on a preset threshold set by the cumulative contribution rate, the preceding principal component load vectors whose cumulative contribution rate exceeds the threshold are selected from the feature value sequence to ensure that the main energy distribution characteristics of the melt flow are preserved and noise components are eliminated. The selected principal component load vectors are used to form a dimension reduction transformation matrix, and a linear transformation is applied to the original high-dimensional micro-region stress tensor matrix set to map it to the low-dimensional principal component score space. Principal component score coefficient sequences are generated in a low-dimensional space to replace the original multi-source redundant stress data, thereby effectively eliminating collinearity interference among multi-source process parameters. By employing the aforementioned methods of covariance matrix construction, eigenvalue decomposition, and principal component selection, the spatially discretized stress field dataset from the previous step is transformed into a low-dimensional principal component score coefficient sequence, achieving the expected technical effect of dimensionality reduction while preserving the main melt flow pattern. For example, in a high-precision cable extrusion production line, the spatially discretized stress field dataset contains a tensor matrix set consisting of 64 circumferential stress sampling points and 16 axial stress sampling points. When constructing the covariance matrix, the stress values ​​of all sampling points are centralized, and the average value is removed from each sampling sequence. Eigenvalue decomposition is performed on the resulting covariance matrix, yielding 80 eigenvalues. The cumulative contribution rate of the first 5 eigenvalues ​​reaches 0.92, and when the preset threshold is 0.9, these 5 eigenvectors are selected as the principal component load vectors. The principal component scores are calculated using the following formula: in, Principal component score coefficient matrix, The normalized matrix of the spatially discretized stress field. The principal component loading matrix is ​​obtained through transformation. The matrix dimension was reduced from 80×100 samples to 5×100 samples, significantly improving computational efficiency. Under different production line speed conditions, this low-dimensional scoring sequence can still completely retain the main energy patterns of melt flow, and the stability of subsequent topological feature extraction is significantly improved, and the accuracy of eccentricity trend prediction is greatly enhanced.

[0082] S3.3: Perform persistent homology topological feature extraction processing on the low-dimensional principal component score coefficient sequence, construct a one-dimensional complex filtering sequence using the sliding window method and calculate the Betti number change trajectory, identify the generation and annihilation events of connected components in stress concentration regions from the topological structure evolution, and output a set of topological invariant features characterizing the evolution path of micro-defects inside the melt, thereby realizing cross-domain mapping from statistical features to geometric topological features.

[0083] S3.4: Based on the topological invariant feature set, perform polar coordinate transformation and gradient operator convolution operation to calculate the polarity angle distribution of the circumferential stress gradient, the coordinate position of the abrupt change point of the radial stress attenuation slope, and the spatial offset of the stress standing wave nodal line, respectively. This transforms the abstract topological invariant feature set into three scalar physical parameters with clear mechanical interpretation: the polarity angle distribution of the circumferential stress gradient, the coordinate position of the abrupt change point of the radial stress attenuation slope, and the spatial offset of the stress standing wave nodal line.

[0084] Based on the connected component generation and annihilation event records in the topological invariant feature set, the polar coordinate transformation operator is called to transform the topological coordinates of each node in the two-dimensional dynamic stress field space grid, mapping the original topological feature positions represented in Cartesian coordinates to a polar coordinate dataset consisting of a combination of circumferential angles and radial distances.

[0085] Gradient operator convolution kernels are loaded onto the polar coordinate dataset, and first-order gradient operations are performed in the circumferential direction to capture the rate of change of the stress field in the circumferential direction, thereby obtaining the original polar angle distribution matrix.

[0086] The polarity angle distribution matrix is ​​searched for the position of the maximum amplitude. The angle value corresponding to the stress gradient direction reversal point is extracted and used as the scalar output of the polarity angle distribution of the circumferential stress gradient.

[0087] Perform a second-order gradient operation in the radial direction to calculate the rate of change of the slope of the radial stress profile curve, and perform abrupt change detection on the curve rate of change sequence to locate the radial coordinate corresponding to the sudden increase in the gradient rate of change. Define the radial coordinate as the location parameter of the abrupt change point of the radial stress attenuation slope.

[0088] For the topological invariant identifier of the stress field standing wave mode, the spatial offset of the standing wave nodal line is calculated by performing difference calculation between the nodal displacement data in polar coordinates and the reference standing wave nodal position, and the offset is output in scalar form.

[0089] By using polar coordinate transformation and gradient operator convolution, the topological invariant feature set is transformed into three scalar physical parameters with clear mechanical interpretation: the polar angle distribution of the circumferential stress gradient, the coordinate position of the abrupt change point of the radial stress attenuation slope, and the spatial offset of the stress standing wave nodal line. This enables a quantitative characterization of the asymmetric properties of melt flow.

[0090] For example, during the operation of a high-precision cable extrusion production line, the initial coordinates of the circumferential node distribution in the collected topological invariant feature set are (12.0mm, 5.0mm), the initial coordinate range of the radial node distribution is 0-10mm, and the reference position of the standing wave nodal line is 3.5mm radially. After polar coordinate transformation, the maximum gradient reversal point appears in the circumferential node angle distribution matrix at... The position is defined as the polarity angular distribution of the circumferential stress gradient. After second-order gradient calculation and abrupt change detection, the radial stress profile curve shows a significant slope change at 7.2 mm radially; this value is recorded as the coordinate position of the abrupt change point in the radial stress attenuation slope. The offset of the standing wave nodal nodes is calculated using displacement difference. mm. The above three parameters are combined into a physical parameter vector and input into the subsequent processing module. It is verified that the output stress fingerprint vector can significantly improve the accuracy of predicting the eccentricity trend in stress-eccentricity response spectrum retrieval, and improve the response speed and stability of compensation control in actual operation.

[0091] S3.5: Perform vector splicing and normalization processing on the polarity angle distribution of the circumferential stress gradient, the coordinate position of the abrupt change point of the radial stress attenuation slope, and the spatial offset of the stress standing wave nodal line. Combine them according to the predefined physical semantic order to generate a three-dimensional stress fingerprint vector with unique identification, and complete the final conversion from multi-source heterogeneous sensing data to a single standardized state index, so as to directly use it as the retrieval key value of the stress-eccentric response spectrum.

[0092] The scalar physical parameter inputs—the polarity angle distribution of the circumferential stress gradient, the coordinates of the abrupt change points in the radial stress attenuation slope, and the spatial offset of the stress standing wave nodal line—are initialized using a uniform format, mapping them to preset index positions according to their physical quantity types. Based on these index positions, a physical feature column vector containing three scalar components is created. Physical semantic order constraints are applied during data construction to ensure that each component has unique contextual relevance in subsequent retrievals. Dimension elimination is performed on the column vector using a min-max normalization method. in For the original physical quantity, This represents the minimum value of the physical quantity in the historical sample. To achieve the maximum value, this normalization process eliminates the numerical range differences between different physical quantities. A vector concatenation operation is then performed on the normalized column vector, combining the three components into a three-dimensional vector of length three according to a predefined physical semantic order. Euclidean norm normalization is then applied to this vector. in This is the concatenated vector. Given its Euclidean length, this process maps the final vector to a unit sphere to improve scale consistency in spatial retrieval. The normalized vector is then encapsulated into a three-dimensional stress fingerprint vector with a unique identifier, achieving the final conversion from multi-source heterogeneous sensor data format to a single normalized state index. Through the aforementioned chained processing method, the physical quantity results from the previous step are transformed into normalized data that can be directly used for stress-eccentric response spectrum retrieval, achieving a precise mapping from stress flow characteristics to eccentricity trend retrieval keys.

[0093] For example, during the extrusion process of a high-precision cable insulation layer, the measured value of the polarity angle distribution of the circumferential stress gradient is... The coordinates of the point where the radial stress attenuation slope abruptly changes are: Millimeters, the spatial offset of the stress standing wave nodal line is Millimeters. The minimum value of the polarity angle distribution of the circumferential stress gradient in the historical sample database is... Degree, maximum value is Degree, location of radial abrupt change point, minimum value of X component millimeters, maximum value millimeters, minimum value of Y component millimeters, maximum value Millimeters, minimum stress nodal line offset is millimeters, maximum value is Millimeters. Performing min-max normalization, the normalized value of the polarity angle distribution of the circumferential stress gradient is calculated as: ≈ The normalized value of the X component at the radial mutation point location is ≈ The normalized value of the Y component is ≈ The normalized value of the nodal line offset is ≈ The three normalized values ​​are arranged into a vector according to their physical semantic order. Calculate the Euclidean norm ≈ Then, after normalization, the final three-dimensional stress fingerprint vector is approximately The vector was used to locate the corresponding eccentricity direction and degree entries in the stress-eccentricity response map retrieval, verifying that the response time is significantly improved and the prediction accuracy is greatly enhanced in high-speed production lines.

[0094] like Figure 3 As shown, step S4 involves: utilizing the mapping relationship between different three-dimensional stress fingerprint vectors calibrated in historical qualified batch data and the actual eccentricity direction and amplitude obtained from laser detection to construct a lookupable stress-eccentricity response spectrum, establishing a physically traceable correlation model between indirect parameters and core state variables. Specifically, this includes: S4.1: Perform spatiotemporal alignment processing on the original dataset of multi-source process parameters from the production process of historical qualified batches, and generate a calibration sample sequence containing timestamps by synchronizing the actual eccentricity direction and eccentricity amplitude data fed back by the laser detection module.

[0095] S4.2: Based on the original data of multi-source process parameters in the calibration sample sequence, perform principal component-topological feature fusion encoding to extract the polarity angle distribution of the circumferential stress gradient, the coordinate position of the abrupt change point of the radial stress attenuation slope, and the spatial offset of the stress standing wave nodal line, which reflect the asymmetry of melt flow, and generate a three-dimensional stress fingerprint vector corresponding to the calibration sample sequence.

[0096] Based on the original data of multi-source process parameters in the calibration sample sequence containing timestamps, the existing datasets of circumferential micro-region stress response signals and axial micro-region stress response signals are called to construct a spatially discretized stress field matrix as the initial input object for encoding processing.

[0097] A covariance matrix construction operation is performed on the spatially discretized stress field matrix to establish a multidimensional covariance matrix with the stress components of the entire matrix as variables, ensuring a global description of the statistical correlation of different sensing components.

[0098] The covariance matrix is ​​subjected to eigenvalue decomposition, the resulting eigenvectors are sorted according to the size of the eigenvalues, and the preceding principal component load vectors are selected based on the cumulative contribution rate threshold to form a dimension reduction transformation matrix, thereby realizing the mapping of the high-dimensional stress field matrix to the low-dimensional principal component score coefficient sequence.

[0099] Persistent homology topological feature extraction is performed on the low-dimensional principal component score coefficient sequence. A one-dimensional complex filtering sequence is constructed through a sliding time window, and the Betti number change trajectory is calculated to capture the topological invariant feature set of the generation and annihilation of connected components in the stress concentration region inside the melt.

[0100] By performing polar coordinate transformation and gradient operator convolution on the topological invariant feature set, the polarity angle distribution of the circumferential stress gradient, the coordinate position of the abrupt change point of the radial stress attenuation slope, and the spatial offset of the stress standing wave nodal line are calculated, transforming the topological abstract quantities into scalar physical parameters with mechanical meaning.

[0101] The polarity angle distribution of the circumferential stress gradient, the coordinates of the abrupt change point of the radial stress attenuation slope, and the spatial offset of the stress standing wave nodal line are processed by vector splicing and normalization, and combined into a three-dimensional stress fingerprint vector according to the physical semantic order, which serves as the index data unit for the subsequent establishment of stress-eccentric response spectrum mapping.

[0102] By using principal component analysis and topological feature fusion processing, the spatiotemporal aligned process parameter data from the previous step is transformed into a physically interpretable three-dimensional stress fingerprint vector, achieving the conversion effect of multi-source heterogeneous inputs to standardized state key values.

[0103] For example, on a high-precision cable production line, the calibration sample sequence includes 500 sets of timestamped circumferential and axial stress response signals. The signal sampling frequency is 2kHz, the temperature gradient data resolution is 0.1℃, and the melt pressure distribution data resolution is 0.05MPa. When constructing the spatially discretized stress field matrix, the number of circumferential sampling points is set to 24, the number of axial sampling points to 12, and the matrix dimension to 288×500. After constructing the covariance matrix, a contribution rate ranking list is calculated through eigenvalue decomposition. The top 7 principal component load vectors with a cumulative contribution rate exceeding 0.85 are selected to form a dimension-reduced matrix. In the persistent cohomology analysis, the sliding window length is set to 50 frames, the step size to 10 frames, and the Betti number variation identifies an average of 4 connected component generation events and 3 annihilation events per batch. The formula for calculating the polarity angle distribution of the circumferential stress gradient in the polar coordinate transformation is: ,in The value of the i-th circumferential gradient component; the coordinates of the abrupt change point in the radial stress attenuation slope are detected by gradient operator convolution, with the average position index being the 5th radial sampling point; the standardized spatial offset of the stress standing wave nodal line ranges from... The range is between 0.4 and 0.6. After normalization, the components of the three-dimensional stress fingerprint vector are all within the range of... Input 1 to 1 as the index key to the subsequent stress-eccentricity response map. The results show that this processing flow can significantly improve the retrievalability of fingerprint and eccentricity mapping, reduce index collisions between different batch process states, and achieve stable and high-precision construction of a physically traceable association model.

[0104] S4.3: Using the three-dimensional stress fingerprint vector corresponding to the calibration sample sequence as the index key, associate the corresponding actual eccentricity direction and eccentricity amplitude data, perform multi-dimensional spatial clustering analysis to divide the eccentricity evolution characteristic regions under different flow patterns and generate eccentricity evolution characteristic clusters.

[0105] The generated 3D stress fingerprint vectors are indexed and correlated with the actual eccentricity direction and amplitude data obtained by the laser detection module. A clear mapping input set is established by constructing joint data pairs. This mapping input set is then normalized according to the physical dimensions of each index to eliminate numerical scale bias caused by different measurement units, forming a dimensionless vector set suitable for spatial analysis. Based on the dimensionless vector set, a distance metric method required for cluster analysis is selected, using high-dimensional Euclidean distance as the basic metric. By calculating the spatial distance matrix between any two sets of vectors, a quantitative basis is provided for subsequent flow pattern classification. Density-based clustering algorithm parameters, including neighborhood radius and minimum sample size, are applied to the distance matrix to adapt to the complex distribution characteristics of melt flow patterns and allow the identification of non-spherical cluster structures. An iterative solution process is performed on the clustering algorithm. Combining the position of the 3D stress fingerprint vector at the cluster center with the corresponding eccentricity parameter change trend, samples with similar spatial distribution and similar eccentricity evolution paths are grouped into the same feature cluster, forming a set of eccentricity evolution feature clusters. Through the above multidimensional spatial clustering analysis, the three-dimensional stress fingerprint vector index result of the previous step is transformed into an eccentric evolution feature region divided according to the flow morphology, so as to achieve the eccentric trend grouping effect based on physical morphological characteristics.

[0106] For example, in a high-precision cable production line, 5000 sets of three-dimensional stress fingerprint vectors and corresponding actual eccentricity direction and amplitude data of historical qualified batches are collected. The polarity angle distribution range of the circumferential stress gradient is within... arrive Between degrees, the coordinates of the point where the radial stress attenuation slope abruptly changes are located in arrive Between millimeters, the spatial offset of the stress standing wave nodal line is within arrive Between millimeters. The above three-dimensional vector and eccentricity parameter are normalized to the [0,1] interval, and the high-dimensional Euclidean distance calculation formula is used: ,in , , Represents the first group of normalized stress fingerprint vector components. , , This represents the quantity of the second component. The neighborhood radius of the clustering algorithm is set to... With a minimum sample size of 10, DBSCAN clustering iterations were performed, resulting in 8 eccentric evolution feature clusters. Each cluster corresponds to a specific melt flow morphology category and eccentricity variation pattern. Validation results show that the eccentricity prediction error within the same feature cluster is significantly lower than that across clusters, ensuring the physical consistency and prediction stability of the mapping relationship.

[0107] S4.4: Based on the data distribution density within the eccentric evolution feature cluster, perform local weighted regression fitting to establish a nonlinear mapping function from the three-dimensional stress fingerprint vector space coordinates to the eccentricity trend prediction direction and the degree of eccentricity trend prediction, and generate stress-eccentricity response mapping entries.

[0108] Based on the spatial distribution density data of eccentric evolutionary feature clusters, the coordinate positions of the center points of each cluster in the three-dimensional stress fingerprint vector space are determined, serving as the core reference nodes of the local fitting region.

[0109] For each sample point in the fitting region, the Euclidean distance between it and the cluster center is calculated, and a weight coefficient matrix is ​​generated according to a preset kernel function to give higher contribution to neighboring samples during the regression fitting process.

[0110] A weighted matrix multiplication operation is performed on the weight coefficient matrix and the eccentric direction label and eccentric amplitude label of the sample points to form a weighted response set of direction and amplitude, which serves as the training input for the nonlinear regression model.

[0111] The local polynomial regression method is used to map the spatial coordinates of the three-dimensional stress fingerprint vector to the direction and amplitude response sets respectively, and construct a bi-branch nonlinear mapping function with clustering characteristics, in which each branch independently fits the direction prediction and amplitude prediction.

[0112] In the fitting operation, least squares error optimization is performed using an objective function of the following form: in, Output predicted values ​​for the model. For eccentric label values, Given the number of samples, the optimal fitting coefficients are obtained by optimizing the direction and amplitude branches respectively.

[0113] The optimized direction and amplitude bi-branch mapping function is uniformly encoded within each cluster fitting region to form a stress-eccentric response mapping table entry.

[0114] By combining local weighted regression fitting with bi-branch nonlinear mapping, the eccentric evolution feature clusters from the previous step are transformed into quantitative response data that supports the retrieval index, thus achieving an accurate mapping from the three-dimensional stress fingerprint vector to the eccentricity trend prediction direction and the degree of eccentricity trend prediction.

[0115] For example, in a 300mm² insulation layer extrusion production scenario, the collected three-dimensional stress fingerprint vector spatial coordinates are concentrated in five feature clusters, with cluster centers located at (0.32, 0.45, 0.18), (0.56, 0.12, 0.41), (0.78, 0.54, 0.37), (0.15, 0.64, 0.53), and (0.49, 0.38, 0.62), respectively. In the local weighted regression process, a Gaussian kernel is selected as the weight kernel function, and the bandwidth parameter is set to 0.2. The weight value of each sample is calculated; for example, the weight value of a sample 0.05 from the cluster center reaches 0.98, while the weight value of a sample 0.25 from the cluster center decreases to 0.37. The weight values ​​are multiplied by the historical eccentricity direction label and amplitude label to obtain the weighted response set. A third-order polynomial regression model is constructed with the three-dimensional coordinates as independent variables, fitting the direction and amplitude branches respectively. The objective function is then used... The two branches were optimized separately. The direction prediction branch outputs an eccentricity angle of 42.3° within the cluster (0.56, 0.12, 0.41), and the amplitude branch outputs an eccentricity compensation of 0.37 mm. During the verification stage, the difference between the prediction results and the actual laser detection values ​​was significantly reduced, ensuring the foresight and accuracy of the compensation control.

[0116] S4.5: The stress-eccentric response mapping entries are structurally encapsulated to construct a stress-eccentric response spectrum that supports fast nearest neighbor retrieval, forming a physically traceable correlation model between indirect parameters and core state variables.

[0117] The stress-eccentricity response mapping entries generated by local weighted regression fitting are indexed and structured. The 3D stress fingerprint vector, predicted eccentricity trend direction, and predicted eccentricity trend degree of each entry are stored in a unified format in the spatial retrieval index unit. Based on these index units, a data domain partitioning operation is performed, establishing multi-level hash mapping buckets according to the numerical ranges of each component of the 3D stress fingerprint vector. This achieves segmented management of spatial coordinates and reduces the load of full-domain traversal during retrieval. Keyword index construction is then performed using the segmented hash mapping bucket data, combining the 3D stress fingerprint vector with its corresponding eccentricity prediction data to form unique key-value pairs, and generating a high-dimensional vector index matrix that supports Euclidean distance calculation. Compact storage optimization is performed on the aforementioned high-dimensional vector index matrix. This reduces storage footprint by compressing redundant eccentric prediction data and sparsifying infrequently used keys, while retaining necessary spatial proximity metadata to support fast nearest neighbor retrieval. The optimized high-dimensional vector index matrix is ​​then associated with a hash mapping bucket structure to form a stress-eccentric response map structure object supporting dual retrieval paths, balancing precise positioning with fast approximate matching. This structured encapsulation transforms the mapping entries from the previous step into a rapidly indexable stress-eccentric response map, enabling the construction of a physically traceable correlation model between indirect parameters and core state variables.

[0118] For example, when processing a batch of historical qualified batch data, the component values ​​of the three-dimensional stress fingerprint vector are: the polarity angle distribution of the circumferential stress gradient (0.785), the coordinate position of the abrupt change point of the radial stress attenuation slope (2.35), and the spatial offset of the stress standing wave nodal line (0.12). The corresponding predicted eccentricity trend direction is 45 degrees, and the predicted degree of eccentricity trend is 0.018 mm. When storing this entry in the index unit, it is first mapped to hash buckets according to the numerical range of each component. The polarity angle distribution of the circumferential stress gradient falls into the 0.7~0.8 interval bucket, the coordinate position of the abrupt change point of the radial stress attenuation slope falls into the 2.3~2.4 interval bucket, and the spatial offset of the stress standing wave nodal line falls into the 0.1~0.15 interval bucket. Then... As key values, the combined eccentricity prediction direction and degree are stored in a high-dimensional vector index matrix. In the process of implementing compact storage, the Euclidean distance from the current data cluster center is used. Redundant data is filtered out, making the matrix storage more sparse. The final output stress-eccentricity response map can directly locate the corresponding eccentricity prediction data based on the above-mentioned structural objects during the retrieval stage, significantly improving the retrieval speed and maintaining the integrity of the physical mapping relationship.

[0119] Step S5: The currently generated three-dimensional stress fingerprint vector is used as an index input into the stress-eccentricity response map. The nearest neighbor response entry is retrieved, and the eccentricity evolution prediction result, including the predicted direction and degree of eccentricity trend, is output. Specifically, this includes: S5.1: Perform normalization preprocessing on the currently generated 3D stress fingerprint vector to eliminate dimensional differences and generate a standardized 3D stress fingerprint vector, ensuring data consistency in subsequent retrieval operations.

[0120] S5.2: Based on the standardized three-dimensional stress fingerprint vector, the spatial distance value between it and all historical calibration entries in the stress-eccentricity response spectrum is calculated using a high-dimensional spatial Euclidean distance metric algorithm to generate a multi-dimensional distance feature set.

[0121] Based on the normalized three-dimensional stress fingerprint vector, the three-dimensional stress fingerprint vector index key of all historical calibration entries in the loaded stress-eccentric response spectrum.

[0122] For each historical calibration entry's three-dimensional stress fingerprint vector index key, a dimension-correspondence difference operation is performed to obtain the deviation components from the standardized three-dimensional stress fingerprint vector in each dimension.

[0123] The deviation component sequence is squared to eliminate the influence of positive and negative signs and is used for subsequent distance accumulation calculation.

[0124] The squared deviation components of each dimension are summed in index order to form a scalar value of the sum of squares reflecting the differences in global spatial location.

[0125] A square root operation is performed on the sum of squares and scalar values ​​to generate the Euclidean distance between the historical calibration entry and the standardized three-dimensional stress fingerprint vector, forming a high-dimensional spatial distance metric result.

[0126] The Euclidean distance values ​​calculated for each historical calibration entry are stored sequentially according to the entry index to form a multidimensional distance feature set.

[0127] Through the above processing method, the fingerprint vector normalized in the previous step is transformed into a multi-dimensional spatial distance feature that can be used for nearest neighbor retrieval, thereby realizing the spatial distance metric required for predicting the eccentric evolution trend.

[0128] For example, under the operating conditions of a high-precision cable production line, the standardized three-dimensional stress fingerprint vector is set to (0.52, 0.33, 0.41), and the stress-eccentricity response spectrum contains 1000 historical calibration entries, among which the three-dimensional stress fingerprint vector of a certain entry is (0.55, 0.30, 0.45). According to the above formula, the first-dimensional deviation is 0.52. 0.55= 0.03, squared value 0.0009; second dimension bias 0.33. 0.30 = 0.03, the squared value is 0.0009; the third dimension bias is 0.41. 0.45= The sum of squares is 0.004, and the square value is 0.0016. The sum of squares is 0.0009 + 0.0009 + 0.0016 = 0.0034, and the square root is 0.0583. This value is stored at the corresponding index position in the multidimensional distance feature set. Under this production line condition, after the distance feature set is constructed, it can be directly used to locate the nearest response item, retrieve the predicted eccentricity trend direction and degree, and significantly improve the response speed and prediction accuracy of feedforward compensation control.

[0129] S5.3: Perform minimum value filtering on the multidimensional distance feature set to locate the nearest response entry with the minimum spatial distance value, and extract the historical eccentricity direction label and historical eccentricity amplitude label associated with the nearest response entry.

[0130] S5.4: Based on the historical eccentricity direction label and the historical eccentricity amplitude label, and combined with the current extrusion line speed dynamic weight coefficient, linear interpolation correction processing is performed to generate an eccentricity evolution prediction result that includes the predicted eccentricity trend direction and the predicted eccentricity trend degree.

[0131] Based on the historical eccentricity direction labels and historical eccentricity amplitude labels extracted from the nearest neighbor response entries filtered by S5.3, the current real-time speed of the extrusion line is set as the input condition for the calculation of dynamic weighting coefficients, and a linear interpolation operation model for direction and amplitude correction is constructed.

[0132] The directional component correction process is performed by using the dynamic weight coefficient corresponding to the historical eccentricity direction label and the current extrusion line speed. The angle value of the historical direction label is multiplied with the current dynamic weight coefficient, and then weighted and combined with the preset interpolation reference direction to generate the correction angle value of the eccentricity trend prediction direction.

[0133] The amplitude component correction process is performed by using the historical eccentricity amplitude label and the dynamic weight coefficient corresponding to the current extrusion line speed. The value of the historical amplitude label is multiplied by the current dynamic weight coefficient, and then combined with the preset interpolation benchmark amplitude for weighted combination to generate the correction amplitude value of the eccentricity trend prediction degree.

[0134] For the direction correction and amplitude correction processes, a unified linear interpolation formula is used for calculation: in, This is the corrected predicted value. For dynamic weighting coefficients, Historical tag values, This is the interpolation reference value.

[0135] The direction correction angle value and the magnitude correction magnitude value are encapsulated into eccentricity trend prediction direction and eccentricity trend prediction degree data structures, respectively, to form a complete eccentricity evolution prediction result.

[0136] Through the above weighted interpolation correction process, the historical tag information obtained in S5.3 is transformed into prediction data that takes into account the dynamic characteristics of the extrusion line, so as to realize the real-time adaptability and physical traceability of the eccentric evolution prediction results in terms of direction and amplitude.

[0137] For example, in a high-speed cable extrusion production line, the current real-time extrusion line speed is measured to be 20.0 m / s, the dynamic weighting coefficient is 0.65 after process calibration, the historical eccentricity direction label is 30.0°, the historical eccentricity amplitude label is 0.40 mm, the interpolation reference direction is 28.0°, and the interpolation reference amplitude is 0.35 mm. Based on the above data, the direction correction process is substituted into the formula: The calculated correction direction value is 29.3°. The amplitude correction process is then substituted into the formula: The calculated correction amplitude is 0.3825 mm. The output eccentricity evolution prediction results include a direction of 29.3° and an amplitude of 0.3825 mm. It has been verified that the deviation between the prediction results and the actual eccentricity measurement values ​​under this speed condition is significantly reduced, supporting the accurate generation of the compensation displacement vector and the optimization of the servo drive control commands in subsequent S6.

[0138] Step S6: Based on the current extrusion line speed and die structure parameters, the predicted eccentricity trend direction in the eccentricity evolution prediction result is converted into a compensation displacement vector in the extruder head plane coordinate system, generating a servo drive target position correction command. Specifically, this includes: S6.1: Based on the currently collected extrusion line speed data and the pre-stored die structure geometric parameters, the predicted direction of the eccentricity trend in the eccentricity evolution prediction result is processed by spatial coordinate system transformation to generate the initial offset angle vector in the extruder head plane coordinate system.

[0139] S6.2: Using the initial bias angle vector and the eccentricity trend prediction data, combined with the melt rheological property coefficient, perform dynamic gain calculation to calculate the theoretical compensation displacement scalar value reflecting the melt hysteresis effect.

[0140] S6.3: Perform vector synthesis operation based on the theoretical compensation displacement scalar value and the initial offset angle vector to construct the original compensation displacement vector containing directional and amplitude components, so as to determine the target movement trajectory of the extruder head on the two-dimensional plane.

[0141] A two-dimensional planar vector synthesis model is established based on the theoretical compensation displacement scalar value and the initial offset angle vector to construct an original compensation displacement vector that simultaneously contains direction and amplitude information. For the input theoretical compensation displacement scalar value, amplitude normalization is performed to convert it into a scalar unit consistent with the extruder head plane coordinate system to eliminate unit differences. A polar coordinate direction basis vector is constructed using the direction component contained in the initial offset angle vector, maintaining a rotating reference system completely consistent with the predicted eccentricity trend direction. The normalized theoretical compensation displacement scalar value and the direction basis vector are multiplied by a scalar to combine amplitude and direction, outputting an intermediate vector for direction and amplitude synthesis. A Cartesian coordinate projection transformation is performed on the intermediate vector to convert the polar coordinate synthesis result into component representations on the X and Y axes, forming dual-axis displacement data that can directly drive servo motion. The above calculations are completed using the vector synthesis formula: in, The original compensation displacement vector, This is the normalized theoretical compensation displacement scalar value. This represents the direction component of the initial offset angle vector. Through the vector synthesis process described above, the result of the previous step is transformed into dual-axis displacement target data that can be used for motion control, thus constructing the two-dimensional planar target movement trajectory of the extruder head.

[0142] For example, in a high-precision cable production line, the real-time extrusion line speed is 1.25 m / s, and the initial offset direction angle given by the die structure parameters is... The degree of eccentricity prediction is 0.32 mm, and the melt rheological coefficient is 1.08. After dynamic gain calculation in step S6.2, the theoretical compensation displacement scalar value is obtained. mm. This scalar value, after amplitude normalization, is maintained in a planar coordinate system unit system, and the direction basis vector is... Degree generation. Calculated using the synthesis formula: the X-direction component is... = mm, Y-axis component is = mm. After performing the Cartesian coordinate projection transformation, the original compensated displacement vector is obtained as ( , (mm). This vector exhibits smooth trajectory and precise direction in the actuator response verification, which can significantly improve the feedforward control effect of eccentricity trend correction.

[0143] S6.4: Apply servo system kinematic constraints to the original compensated displacement vector and perform amplitude limiting and smoothing filtering to eliminate high-frequency jitter and generate an optimized compensated displacement vector that conforms to the driver response characteristics.

[0144] For the original compensated displacement vector containing directional and amplitude components generated in the previous sub-step S6.3, the kinematic constraint parameters of the servo system are loaded as input conditions. Combined with the maximum stroke limit, acceleration limit, and transient vibration tolerance of the drive mechanism, the amplitude component in the original compensated displacement vector is subjected to amplitude limiting calculation to prevent the mechanical actuator from exceeding the design displacement range. A time-series response model is constructed based on the amplitude-limited compensated displacement vector. Dynamic smoothing filtering preprocessing based on second-order difference is performed using the boundary conditions of acceleration and velocity to eliminate the risk of response overshoot caused by high-frequency transient changes. The smoothed displacement vector is input into a discretized filter structure matching the sampling period of the servo drive. Cross-period buffer compensation is achieved through convolution operations on the continuous displacement trajectory, reducing vibration noise coupling in the mechanical domain. A curvature adjustment algorithm based on third-order polynomial fitting is applied to the convolution-processed displacement vector to adjust the radius of curvature of the vector path to optimize kinematic smoothness and avoid inertial shocks caused by excessively fast steering or rapid changes in direction. The output of the above processing actions, after amplitude limiting, smoothing, and curvature optimization, ensures that its dynamic characteristics match the response bandwidth of the servo driver and meet the mechanical domain execution stability requirements. By applying kinematic constraints to the servo system and sequentially performing processes such as amplitude limiting, dynamic smoothing, convolution buffering, and curvature optimization, the original compensation displacement vector from the previous step is transformed into an optimized compensation displacement vector that conforms to the driver's response characteristics, achieving high-precision displacement control without high-frequency jitter.

[0145] For example, on a high-precision cable extrusion production line operating at 0.8 m / s, the original compensated displacement vector amplitude is 0.65 mm, the direction angle is 48°, the maximum safe travel amplitude allowed by the servo drive is 0.50 mm, the maximum safe acceleration is 15 mm / s², and the inertial shock tolerance is 5 mm / s. After inputting the amplitude component into the limiting module, a limiting value of 0.50 mm is obtained. The limiting vector is then processed by a second-order differential smoothing filter with a sampling period of 5 ms. The filter coefficients are set to 0.25, 0.5, and 0.25, resulting in a smoothed output amplitude of 0.48 mm. The smoothed vector is then input into a cross-period convolution buffer module with a convolution kernel length of 3 sampling periods, resulting in a buffered amplitude of 0.47 mm. Third-order polynomial curvature adjustment is performed, adjusting the fitted curvature radius from the original 2.5 mm to 3.2 mm, and optimizing the direction angle to 47.5°. The final output optimized compensation displacement vector amplitude is 0.47 mm and the direction angle is 47.5°. When executing the displacement trajectory within the response bandwidth of the servo driver, the high-frequency jitter amplitude is significantly reduced, and the displacement trajectory exhibits good smoothness and execution accuracy in mechanical domain motion.

[0146] S6.5: Based on the optimized compensation displacement vector, perform a pulse equivalent mapping transformation operation to convert it into a digital target position correction command that can be recognized by the servo driver, so as to drive the extruder head to perform precise linkage adjustment actions.

[0147] Based on optimized compensation displacement vector data conforming to the driver's response characteristics, the scaling factor between the servo driver's pulse equivalent parameters and the displacement in the planar coordinate system is obtained as the input index. A linear mapping calculation from displacement components to pulse components is performed, mapping the direction and amplitude components in the two-dimensional plane to the pulse steps of the servo driver's X and Y axes, respectively. The pulse equivalent mapping formula is then used... in The target number of pulses, The displacement is in mm. The pulse equivalent coefficient (pulse / mm) is used to calculate the target pulse number for each axis, ensuring synchronization between different axes. The mapped pulse number is integerized and rounded to compensate for errors, avoiding positioning deviations caused by floating-point pulse values. Combining the servo drive communication protocol format, the integerized pulse value is split into high-byte and low-byte segments and a CRC checksum is generated, forming a digital target position correction command data packet that the servo drive can parse. By sending this digital command data packet, the optimized compensation displacement vector is converted into precise execution pulse quantities for the servo drive, achieving synchronous linkage control of multi-axis spatial displacement of the extruder head. Through pulse equivalent mapping transformation, the planar compensation vector from the previous step is converted into target pulse commands executable by the servo control, achieving precise transmission from the digital domain to the mechanical domain.

[0148] For example, in the scenario of cable insulation eccentricity compensation control, the X-axis component of the optimized compensation displacement vector is 1.25mm, and the Y-axis component is... The servo driver pulse equivalent coefficient K is set to 2000 pulses / mm, with a diameter of 0.85mm. The target pulse count for the X-axis is calculated using the mapping formula. The result equals 2500 pulses, and the target pulse number on the Y-axis is... The result equals 1700pulse. (The rest of the text appears to be a mix of numbers and symbols, possibly related to a specific program or event.) The 1700 bytes are integerized and rounded to zero to compensate for rounding errors. They are then segmented into high and low bytes according to the servo driver communication protocol, with a CRC checksum added. The instruction packet is sent to the servo system via the fieldbus, driving a positive X-axis displacement of 1.25 mm and a negative Y-axis displacement of 0.85 mm, achieving synchronous adjustment of the extruder head in a two-dimensional plane. Tests show that, under the above parameter configuration, the servo driver response time remains in the millisecond range, the displacement execution error is significantly reduced, the concentricity of the insulation layer is significantly improved after compensation, and the real-time eccentricity control effect is stable.

[0149] Step S7: Based on the target position correction command of the servo driver, the extruder head is driven to perform a linkage adjustment action to change the melt extrusion space distribution, thereby counteracting the predicted insulation layer eccentricity trend and achieving feedforward compensation control. Specifically, this includes: S7.1: Parse and process the target position correction command of the servo drive, extract the angular component containing the predicted direction of eccentricity and the amplitude component containing the predicted degree of eccentricity, and generate the compensation displacement vector data in the extruder head plane coordinate system to be executed.

[0150] S7.2: Based on the current extrusion line speed and die structure parameters, perform dynamic gain calibration calculation on the compensation displacement vector data to eliminate mechanical transmission lag effect and nonlinear friction interference, and generate the real-time displacement control quantity of the extruder head after speed compensation.

[0151] Based on the compensated displacement vector data in the extruder head plane coordinate system obtained through analysis, the currently acquired extrusion linear velocity value and die structure parameter set are used as the dynamic calibration input source. The extrusion linear velocity data is input to the mechanical transmission delay analysis module to establish a quantitative relationship between speed changes and the servo system response time, obtaining the speed-related hysteresis coefficient matrix. The die structure parameters are input to the friction torque modeling module, and the nonlinear friction interference value is calculated by combining the die lip gap width, melt contact area, and viscosity coefficient. The speed-related hysteresis coefficient matrix and the nonlinear friction interference value are input into the gain correction formula to perform real-time correction calculations. in, This is the dynamic gain correction value. This is the theoretical gain coefficient. This is the speed lag scaling factor. This refers to the extrusion line speed. As the friction reference value, This is the melt viscosity coefficient. The contact area of ​​the die lip is defined. The dynamic gain correction value and the compensated displacement vector data are multiplied to generate a set of displacement control vectors after speed and friction compensation. Inertial filtering and numerical smoothing are performed on the displacement control vector set to remove control signal spikes introduced by instantaneous speed fluctuations, outputting the real-time displacement control quantity of the extruder head after speed compensation. Through the above dynamic gain calibration process, the compensated displacement vector data from the previous step is transformed into a displacement control quantity that conforms to the actual mechanical response characteristics, thereby eliminating the mechanical transmission hysteresis effect and nonlinear friction interference.

[0152] For example, in a high-speed insulation layer extrusion production scenario, the extrusion line speed is collected in real time. m / s, die lip clearance width is mm, melt viscosity coefficient Pa·s, contact area of ​​mold lip cm². Velocity lag proportional factor. Friction reference value N. The theoretical gain coefficient is set to Substituting the above parameters into the dynamic gain correction formula, the velocity-related lag coefficient matrix is ​​calculated to be... The nonlinear friction interference value is After performing the correction calculation, the dynamic gain correction value is obtained. To compensate for the magnitude of the displacement vector mm is the input, which is multiplied by the correction gain to obtain the calibrated displacement control amplitude. The amplitude of the control signal is stabilized at mm, while maintaining the original angular components. After inertial filtering and smoothing, the amplitude is stabilized at... Within a mm range, there are no sharp fluctuations. When this control variable is applied to drive the extruder head, the output trajectory remains consistent in the mechanical domain, significantly improving the execution accuracy and response stability of the feedforward compensation control.

[0153] S7.3: The real-time displacement control quantity of the extruder head after speed compensation is used to drive the multi-axis linkage servo motor to perform differential stepping motion, converting the control quantity in the digital domain into a high-precision spatial displacement trajectory of the extruder head in the mechanical domain.

[0154] Based on the real-time displacement control value of the extruder head after speed compensation, this value is used as the core input parameter for the multi-axis linkage servo motor drive.

[0155] Digital signal formatting is performed on the control interface of the multi-axis servo motor to convert the displacement control quantity into a fixed-length data frame structure that can be accepted by the servo controller, and a synchronization trigger flag is embedded in the data frame to ensure the consistency of multi-axis start-up.

[0156] The servo controller calls the subdivision pulse generation module to construct a differential step motion command sequence with a high subdivision level according to the target number of pulses corresponding to the displacement control quantity, and embeds the velocity feedforward coefficient in the pulse interval to maintain motion smoothness.

[0157] Multi-axis synchronous scheduling is performed on the differential step motion command sequence, and the pulse sequences of each axis are rearranged in time slices according to spatial coordination relationship to ensure that the displacement trajectory of each coordinate axis meets the preset linkage geometric constraint conditions during execution.

[0158] The synchronized pulse sequence is sent to each servo motor drive module. The motor encoder collects the feedback position in real time and compares it with the target position. Closed-loop pulse compensation is performed to correct the offset error caused by inertia or friction, and finally a high-precision spatial displacement trajectory of the extruder head is generated.

[0159] Through the above-mentioned multi-axis control and closed-loop correction method, the real-time displacement control quantity after speed compensation in the previous step is transformed into a precise motion trajectory that can be executed in the physical domain, thereby achieving high resolution and high stability of extruder head position adjustment.

[0160] For example, in a high-precision cable extrusion production line, the real-time displacement control value of the extruder head is set to 0.025mm, the control system is configured with dual servo motors for X and Y axes, and the servo subdivision level is set to 20,000 pulses / mm. The 0.025mm value is input to the pulse generation module, and the target pulse count is calculated. = For 500 pulses, speed feedforward correction is performed, with the feedforward coefficient set to... The corrected target pulse number is = The pulse sequences of both axes are synchronized according to the geometric constraints of the mold lip, ensuring that the X-axis output pulses account for 60% of the total and the Y-axis for 40%. During execution, the current position is acquired through the encoder, compared with the target position error, and the number of compensation pulses is calculated based on the deviation value. For example, if the error is... Upon pulse activation, a compensation pulse is immediately inserted to eliminate the deviation. The extruder head spatial trajectory output after this process deviates less than the expected trajectory in measurement verification. μm, significantly improving the accuracy and stability of linkage adjustment.

[0161] S7.4: Adjust the spatial geometric distribution of the die lip gap of the extruder head according to the high-precision spatial displacement trajectory of the extruder head, change the velocity field distribution characteristics of the molten polymer in the annular flow channel, and generate a melt extrusion spatial distribution pattern with directional offset characteristics.

[0162] S7.5: Based on the spatial distribution pattern of melt extrusion with directional offset characteristics, reverse filling compensation is performed on the wall thickness of the insulation layer being formed to counteract the eccentricity trend of the insulation layer caused by process fluctuations, thereby achieving feedforward correction control of the concentricity of the cable insulation layer.

[0163] Based on the spatial distribution of melt extrusion with directional offset characteristics generated by the extruder head, online measurement data of the current forming insulation layer wall thickness is obtained as the initial condition for reverse compensation calculation.

[0164] The wall thickness spatial distribution data and the target concentric wall thickness model are differentially processed to obtain the wall thickness deviation distribution matrix, and the deviation direction vector field and deviation amplitude scalar field are generated accordingly.

[0165] A reverse rotation transformation along the circumferential direction is applied to the deviation direction vector field, so that the flow direction of the compensating melt is 180° opposite to the deviation direction, ensuring that the compensating flow can fill the insufficient thickness area.

[0166] The theoretical filling flow distribution matrix is ​​obtained by multiplying the deviation amplitude scalar field with the local rheological coefficient of the melt, which is used to guide the local flow adjustment of directional melt delivery.

[0167] Finite element method is used to apply circumferential and radial boundary condition constraints to the filling flow distribution matrix, calculate the velocity field redistribution of the compensating melt in the annular flow channel, and map the results back to the extruder head die lip geometry control unit to adjust the gap of each region to form the compensating flow channel.

[0168] The melt velocity field adjusted by the compensation channel is superimposed with the existing molding velocity field to generate a new distribution of the overall extrusion velocity field, thereby achieving reverse filling compensation of wall thickness.

[0169] By using the above-mentioned reverse filling compensation process, the spatial distribution of the directional offset melt in the previous step is transformed into a molding control index for wall thickness equalization, thereby achieving feedforward correction control of the concentricity of the cable insulation layer.

[0170] For example, in a high-precision cable production scenario, the measured circumferential wall thickness deviation of the insulation layer ranges from 0.15mm to 0.35mm, the target wall thickness is 2.00mm, and the existing melt rheological coefficient has an average radial value of 2.5×10⁻⁶. 4 Pa·s. Input conditions are that the deviation direction is concentrated in a 90°±15° loop, and the maximum value of the deviation amplitude matrix is ​​0.35 mm. Differential operations yield the deviation amplitude scalar field, which is multiplied by the rheological coefficient to form the theoretical filling flow distribution matrix, where the maximum value is... = Pa·s·mm. The finite element method uses circumferential boundary conditions to fix the outer radius of the mold lip and radial boundary conditions to constrain the outer wall cooling zone. The calculated compensation velocity field in the deviation region is increased by [value missing]. The wall thickness of the deviation area increased by 0.34 mm after being superimposed with the original velocity field at m / s, which is close to the target wall thickness. This resulted in a significant improvement in the concentricity of the insulation layer and a stable forming process.

[0171] Step S8: Determine whether the deviation between the actual eccentricity data fed back by the laser detection module and the eccentricity evolution prediction result continuously exceeds a preset dynamic tolerance threshold. If the condition is met, trigger the online fine-tuning process of the fingerprint model to update the local response region of the stress-eccentricity response spectrum. Specifically, this includes: S8.1: Obtain the actual eccentricity direction and actual eccentricity amplitude data output by the laser detection module, and simultaneously extract the eccentricity trend prediction direction and eccentricity trend prediction degree data generated by the stress-eccentricity response spectrum retrieval at the current moment. Use the above four sets of data as the basic input source for deviation calculation.

[0172] S8.2: Perform vector angle difference calculation based on the actual eccentricity direction and the predicted eccentricity trend direction, and simultaneously perform scalar numerical subtraction operation based on the actual eccentricity amplitude and the predicted eccentricity trend degree to generate a comprehensive deviation vector sequence containing the direction deviation angle and amplitude deviation.

[0173] S8.3: Perform sliding time window statistical test processing on the comprehensive deviation vector sequence to determine whether the direction deviation angle and amplitude deviation in multiple consecutive sampling periods continuously exceed the preset dynamic tolerance threshold. If the continuous over-limit condition is met, generate a model fine-tuning trigger command.

[0174] S8.4: In response to the model fine-tuning trigger command, locate the index position of the current three-dimensional stress fingerprint vector in the stress-eccentricity response map, and select the historical qualified batch calibration data within the neighborhood of the index position as the local training sample set to limit the search space for model update.

[0175] S8.5: Using the local training sample set, perform gradient descent iterative optimization on the mapping weight coefficients of the corresponding region in the stress-eccentricity response map, recalibrate the nonlinear correlation curve between the three-dimensional stress fingerprint vector and the actual eccentricity direction and actual eccentricity amplitude, and output the updated local response map region to complete the online fine-tuning of the fingerprint model.

[0176] For those skilled in the art, various other corresponding changes and modifications can be made based on the technical solutions and concepts described above, and all such changes and modifications should fall within the protection scope of the claims of this invention.

[0177] Unless otherwise defined, the technical or scientific terms used herein shall have the ordinary meaning as understood by one of ordinary skill in the art to which this application pertains. The terms “first,” “second,” “third,” and similar terms used in this patent application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms “an” or “a” and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising” or “including” and similar terms mean that the elements or objects preceding “comprising” or “including” encompass the elements or objects listed following “comprising” or “including” and their equivalents, and do not exclude other elements or objects. The “multiple” mentioned in the embodiments of this application refers to two or more. A and / or B indicate three possibilities: A; B; and A and B.

[0178] The above description is merely an exemplary embodiment of this application, but the scope of protection of this application 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 application, and such modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A laser-based method for compensating and controlling cable insulation eccentricity, specifically including: S1: Obtain the original dataset of multi-source process parameters at several key cross-sections of the cable extruder head; S2: Based on the original dataset of the multi-source process parameters, construct a two-dimensional dynamic stress field snapshot sequence with high spatiotemporal resolution, and map the discrete sensing signals into the dynamic stress field distribution characteristics of the melt in the continuous spatial domain. S3: Based on the dynamic stress field distribution characteristics of the melt, extract a three-dimensional stress fingerprint vector that reflects the asymmetry of melt flow; S4: Construct a stress-eccentricity response spectrum by using the mapping relationship between the different three-dimensional stress fingerprint vectors calibrated in the historical qualified batch data and the actual eccentricity direction and eccentricity amplitude obtained by the laser detection module; S5: Input the currently generated three-dimensional stress fingerprint vector as an index into the stress-eccentricity response map, retrieve the nearest response entry, and output the eccentricity evolution prediction result containing the predicted direction and degree of eccentricity trend. S6: Based on the current extrusion line speed of the cable extruder head and the die structure parameters, the predicted eccentricity trend direction is converted into a compensation displacement vector in the extruder head plane coordinate system, and a servo drive target position correction command is generated. S7: Based on the target position correction command of the servo driver, drive the extruder head to perform a linkage adjustment action to change the melt extrusion space distribution.

2. The cable insulation eccentricity compensation control method based on laser detection according to claim 1, characterized in that, Following step S7, step S8 is also included: If the deviation between the actual eccentricity direction and eccentricity amplitude fed back by the laser detection module and the eccentricity evolution prediction result continues to exceed the preset dynamic tolerance threshold, then the local response region of the stress-eccentricity response spectrum is updated.

3. The cable insulation eccentricity compensation control method based on laser detection according to claim 1, characterized in that, The original dataset of multi-source process parameters includes: circumferential micro-region stress response signal, axial micro-region stress response signal, and simultaneously collects real-time melt temperature gradient data, melt pressure distribution data and traction tension data.

4. The cable insulation eccentricity compensation control method based on laser detection according to claim 1, characterized in that, The three-dimensional stress fingerprint vector includes: the polarity angle distribution of the circumferential stress gradient, the coordinate position of the abrupt change point of the radial stress attenuation slope, and the spatial offset of the stress standing wave nodal line.

5. The cable insulation eccentricity compensation control method based on laser detection according to claim 3, characterized in that, Step S3 specifically includes: The high spatiotemporal resolution two-dimensional dynamic stress field snapshot sequence is subjected to mesh generation processing to transform the continuously distributed dynamic stress field distribution characteristics of the melt into a set of regularly arranged micro-region stress tensor matrices, so as to obtain a spatially discretized stress field dataset containing the circumferential micro-region stress response signal and the axial micro-region stress response signal. Based on the spatially discretized stress field dataset, covariance matrix construction and eigenvalue decomposition are performed. Principal component analysis algorithm is used to extract the preceding principal component load vectors whose cumulative contribution rate exceeds a preset threshold. The high-dimensional redundant micro-region stress tensor matrix set is compressed into a low-dimensional principal component score coefficient sequence. Persistent homology topological feature extraction processing is performed on the low-dimensional principal component score coefficient sequence. A one-dimensional complex filtering sequence is constructed using the sliding window method, and the Betti number change trajectory is calculated. The generation and annihilation events of connected components in stress concentration regions are identified from the topological evolution, thereby outputting a set of topological invariant features characterizing the evolution path of micro-defects inside the melt. Based on the topological invariant feature set, polar coordinate transformation and gradient operator convolution are performed to calculate the polar angle distribution of the circumferential stress gradient, the coordinate position of the abrupt change point of the radial stress attenuation slope, and the spatial offset of the stress standing wave nodal line, respectively. The polarity angle distribution of the circumferential stress gradient, the coordinates of the abrupt change point of the radial stress attenuation slope, and the spatial offset of the stress standing wave nodal line are subjected to vector splicing and normalization processing, and combined to generate a three-dimensional stress fingerprint vector with unique identification.

6. The cable insulation eccentricity compensation control method based on laser detection according to claim 1, characterized in that, Step S4 specifically includes: Spatiotemporal alignment processing is performed on the original dataset of multi-source process parameters from the historical qualified batch production process to generate a calibration sample sequence containing timestamps; Based on the original data of multi-source process parameters in the calibration sample sequence, the polarity angle distribution of the circumferential stress gradient, the coordinate position of the abrupt change point of the radial stress attenuation slope, and the spatial offset of the stress standing wave nodal line, which reflect the asymmetry of melt flow, are extracted to generate a three-dimensional stress fingerprint vector corresponding to the calibration sample sequence. Using the three-dimensional stress fingerprint vector corresponding to the calibrated sample sequence as an index key, the corresponding actual eccentricity direction and eccentricity amplitude data are associated to divide the eccentricity evolution characteristic regions under different flow patterns and generate eccentricity evolution characteristic clusters. Based on the data distribution density within the eccentric evolution feature cluster, a nonlinear mapping function is established from the three-dimensional stress fingerprint vector space coordinates to the eccentricity trend prediction direction and the degree of eccentricity trend prediction, and stress-eccentricity response mapping entries are generated. The stress-eccentricity response mapping entries are structured and encapsulated to construct a stress-eccentricity response spectrum that supports fast nearest neighbor retrieval.

7. The cable insulation eccentricity compensation control method based on laser detection according to claim 1, characterized in that, Step S7 includes: The servo drive target position correction command is parsed and processed to extract the angle component containing the predicted direction of eccentricity and the amplitude component containing the predicted degree of eccentricity, and to generate the compensation displacement vector data in the extruder head plane coordinate system to be executed. Based on the current extrusion line speed and die structure parameters, dynamic gain calibration calculation is performed on the compensated displacement vector data to generate the real-time displacement control quantity of the extruder head after speed compensation. By using the real-time displacement control of the extruder head after speed compensation to drive a multi-axis linkage servo motor to perform differential stepping motion, a high-precision spatial displacement trajectory of the extruder head is obtained. The spatial geometric distribution of the die lip gap of the extruder head is adjusted according to the high-precision spatial displacement trajectory of the extruder head, thereby changing the velocity field distribution characteristics of the molten polymer in the annular flow channel and generating a spatial distribution pattern of melt extrusion with directional offset characteristics. Based on the spatial distribution pattern of melt extrusion with directional offset characteristics, reverse filling compensation is performed on the wall thickness of the insulation layer being formed, thereby achieving feedforward correction control of the concentricity of the cable insulation layer.

8. The cable insulation eccentricity compensation control method based on laser detection according to claim 3, characterized in that, Multiple piezoelectric micro-zone stress-sensitive units are uniformly arranged in the annular distribution area of ​​the key cross section of the cable extruder head to generate a circumferential micro-zone stress response signal acquisition channel and an axial micro-zone stress response signal acquisition channel with spatial resolution capability. The micro-zone stress response signal acquisition channel is used to acquire the circumferential micro-zone stress response signal, and the axial micro-zone stress response signal acquisition channel is used to acquire the axial micro-zone stress response signal.

9. The cable insulation eccentricity compensation control method based on laser detection according to claim 3, characterized in that, A thermocouple array and a pressure transmitter are arranged along the cable extrusion channel. The thermocouple array is used to collect real-time melt temperature gradient data, and the pressure transmitter is used to collect melt pressure distribution data.

10. The cable insulation eccentricity compensation control method based on laser detection according to claim 3, characterized in that, The torque sensor output of the traction machine drive end of the cable insulation extrusion production line is obtained, the longitudinal force characteristics under the current production line operation state are calculated, and the traction tension data characterizing the cable forming resistance is generated.

Images