Cable wrapping tension drift prediction method and system based on machine learning model

Abstract

The application provides a cable wrapping tension drift prediction method and system based on a machine learning model, and is applied to the technical field of tension drift prediction; through analyzing the cable wrapping tension demand influence mechanism, quantifying the influence degree of the change of the winding diameter and the change of the traction line speed on the cable wrapping tension demand, when the judgment result is implementation, based on the output quantitative influence degree result, decoupling and drift identification analysis of the cable wrapping tension signal level are carried out, when the judgment result is non-implementation, the received multi-source cable wrapping time sequence data is taken as a prediction benchmark, and the machine learning model is used to predict the cable wrapping tension drift trend, so as to solve the problem that in the process of predicting the tension drift in the cable wrapping process, the tension signal is usually superimposed with inertial impact and control lag under the conditions of rapid change of the winding diameter, frequent acceleration and deceleration of the speed, short-time dramatic fluctuation, and the coupling of the drift trend and the transient overshoot / undershoot.

Description

Technical Field

[0001] This invention relates to the field of tension drift prediction technology, and in particular to a method and system for predicting cable wrapping tension drift based on a machine learning model. Background Technology

[0002] To ensure efficient and stable data transmission, high-speed communication cables, employing advanced materials and designs, possess characteristics such as low attenuation, high bandwidth, and strong anti-interference capabilities. These enable them to maintain signal clarity over long distances, meeting the high demands of modern communication equipment for data transmission speed and quality. Currently, the world is in a phase of rapid digital transformation, with 5G and even more advanced communication technologies becoming increasingly widespread. 5G networks, with their high speed, low latency, and large capacity, require high-quality high-speed communication cables as the physical transmission medium to guarantee stable and efficient signal transmission.

[0003] The existing cable production process mainly includes the following stages: raw material inspection and pretreatment, conductor drawing / stranding, insulation extrusion and cooling sizing, (shielding / filling) cabling, inner and outer sheath extrusion, wrapping, armoring, braiding and other structural processing, online testing and electrical performance testing. Cable wrapping is a key process that determines the density and geometric stability of the structural layers. Tension drift (including zero-point drift, proportional drift, and trend drift) directly leads to quality defects such as uneven wrapping tightness, lap ratio fluctuations, wrinkling, voids, misalignment, and localized stress concentration, thus affecting insulation protection and durability reliability. Therefore, it is crucial to investigate cable wrapping tension drift. Predictive analysis has become an important foundation and urgent need for achieving proactive process quality control and closed-loop adaptive adjustment. Existing cable wrapping machines typically include the following general units: wire feeding device: wire feeding reel, wire feeding frame, active wire feeding motor (or brake), tension adjustment mechanism (swinging arm / tension arm, etc.), wrapping execution device: wrapping head (ring / planetary / oscillating type), wrapping tape supply mechanism, etc., traction device: traction wheel or track traction, wire take-up device: take-up reel / winding machine, etc., based on the cable wrapping machine, detection and control device: tension detection, speed / length measurement, roll diameter detection, temperature and humidity detection, controller (motion controller) and human-machine interface, etc. The existing prediction process is as follows: During the cable wrapping production process, multi-source cable wrapping timing data, such as tension, traction or unwinding speed, roll diameter change, temperature and humidity, are simultaneously collected by acquisition equipment. Specifically: a tension sensor is installed on the exit path of the wrapping tape at the wire release end to obtain the real-time tension of the wrapping tape; encoders are installed on the drive shaft ends of the traction device and the wire release device to obtain the traction line speed and the wire release speed, respectively; a distance-measuring roll diameter sensor or a vision measurement device is installed on the side of the wire release reel (and / or the take-up reel) to obtain the roll diameter change; and temperature and humidity sensors are installed near the wrapping head to obtain environmental and thermal drift information. The signals from all the above sensors are then unified. Timestamp alignment and synchronous sampling form multi-source time-series data for monitoring and analysis of the wrapping process. After denoising, time-series alignment, and anomaly removal of the tension signal, a feature sequence reflecting the evolution of the working condition (such as short-time mean, variance, slope, frequency domain energy, and process parameter combination) is constructed. A regression-type machine learning model (such as random forest) is used to learn the mapping relationship between the working condition window and the tension drift in the future time period. The predicted value and confidence level of the tension drift in the next few seconds or several revolutions are output online. The prediction results are used for feedforward compensation (adjusting the traction, braking, or wrapping execution amount) and threshold warning, so as to achieve early correction and stable control before the tension exceeds the tolerance.

[0004] Technical Issues: Tension is a key process parameter determining the density and structural stability of cable wrapping. Tension drift directly leads to quality defects such as uneven wrapping tightness, overlapping rate fluctuations, wrinkling, voids, misalignment, and localized stress concentration. To meet the high demands for precision wrapping quality and consistency control, existing methods typically employ constant tension control based on real-time feedback from tension sensors, supplemented by empirical threshold alarms or periodic manual calibration. However, under conditions of rapid changes in coil diameter and frequent acceleration / deceleration of speed (traction line speed), the coil diameter sensor and visual measurement device... Rapid changes in the detected roll diameter cause a dynamic difference between the pay-off linear velocity corresponding to the encoder at the drive shaft end of the pay-off device and the traction linear velocity corresponding to the encoder at the drive shaft end of the traction device, resulting in rapid time-varying linear velocity and equivalent inertia. At the same time, the tension sensor and its measurement link arranged on the tape exit path at the pay-off end of the wrapping tape inevitably accumulate noise and may produce zero-point or gain offsets, causing the controller to lag or deviate in adjusting the active pay-off motor / brake actuator. This makes it easy for tension overshoot or undershoot to occur during constant tension control, thereby causing predicted wrapping quality fluctuations and an increase in the defect rate.

[0005] As can be seen from the above, when predicting the tension drift during cable wrapping, there is usually a problem that the tension signal is superimposed with inertial impact and control lag under the conditions of rapid change of roll diameter and frequent acceleration and deceleration, resulting in short-term violent fluctuations. The drift trend and transient overshoot / undershoot are coupled and difficult to distinguish. Summary of the Invention

[0006] This invention provides a method and system for predicting cable wrapping tension drift based on a machine learning model. The technical solution provided by this application is as follows:

[0007] Firstly, a method for predicting cable wrapping tension drift based on a machine learning model is provided. The specific implementation of this method is as follows: S1, analyze the influence mechanism of cable wrapping tension demand, quantify the influence of changes in coil diameter and traction line speed on cable wrapping tension demand, obtain the corresponding quantified influence results, and obtain the decision result on whether to implement decoupling and drift identification analysis at the cable wrapping tension signal level; S2, if the decision result is to implement, based on the output quantified influence results, perform decoupling and drift identification analysis at the cable wrapping tension signal level to obtain data on the quantified influence of cable wrapping tension signal on cable wrapping tension demand; if the decision result is not to implement, predict the cable wrapping tension drift trend based on a machine learning model; S3, using the received multi-source cable wrapping time-series data as a prediction benchmark, predict the cable wrapping tension drift trend based on a machine learning model.

[0008] Secondly, a cable wrapping tension drift prediction system based on a machine learning model is provided, including: a tension demand influence analysis module, used to analyze the influence mechanism of cable wrapping tension demand, quantify the influence of changes in coil diameter and traction line speed on cable wrapping tension demand, obtain the corresponding quantitative influence results, and obtain the decision result of whether to implement cable wrapping tension signal level decoupling and drift identification analysis; a tension signal decoupling and drift decision module, used to perform cable wrapping tension signal level decoupling and drift identification analysis based on the output quantitative influence results when the decision result is to implement, obtain data on the influence of cable wrapping tension signal on cable wrapping tension demand, and predict the cable wrapping tension drift trend based on a machine learning model when the decision result is not to implement; and a machine learning and tension drift prediction module, used to predict the cable wrapping tension drift trend based on a machine learning model using received multi-source cable wrapping time series data as a prediction benchmark.

[0009] The beneficial effects of the technical solutions provided in the embodiments of the present invention include at least the following:

[0010] 1. By analyzing the influence mechanism of cable wrapping tension demand, the influence of changes in coil diameter and traction speed on cable wrapping tension demand is quantified, yielding corresponding quantitative influence results and determining whether to implement decoupling and drift identification analysis at the cable wrapping tension signal level. This helps to accurately locate and quantify the influence of tension demand disturbance factors, clarify the necessity of decoupling identification, and avoid invalid analysis and redundant operations. When the decision is to implement, decoupling and drift identification analysis at the cable wrapping tension signal level is performed based on the output quantitative influence results, obtaining data on the influence of cable wrapping tension demand on the quantified cable wrapping tension signal. When the decision is not to implement, the cable wrapping tension drift trend is predicted based on a machine learning model, which helps to achieve differentiated adaptation between tension signal processing and drift prediction. Using the received multi-source cable wrapping time-series data as a prediction benchmark, the cable wrapping tension drift trend is predicted based on a machine learning model, which helps to achieve forward-looking prediction of cable wrapping tension drift and avoid the risk of drift accumulation in advance.

[0011] 2. For multi-source cable wrapping timing data corresponding to the qualified wrapping state range, a machine learning model is used to predict the cable wrapping tension drift trend. For multi-source cable wrapping timing data corresponding to the abnormal wrapping state range, adaptive adjustments are made based on the impact of tension demand. Compared with the shortcomings of existing technologies, such as the inability to adapt to the differences in tension demand under different working conditions, resulting in untimely tension drift control under abnormal working conditions, excessive redundant calculations under qualified working conditions, and insufficient tension control accuracy, this technology helps to achieve accurate differentiation and differentiated adaptation of wrapping working conditions. This makes tension prediction more in line with actual wrapping needs, thereby improving the accuracy and efficiency of cable wrapping tension control, reducing the impact of tension drift on wrapping quality, and ensuring the consistency of wrapping quality.

[0012] 3. By performing decoupling and drift identification analysis at the tension signal level, and triggering the cable wrapping tension signal adjustment mechanism when the decoupling-identification anomaly value of the tension signal exceeds the defined decoupling identification anomaly threshold, and conversely, predicting the cable wrapping tension drift trend based on a machine learning model, this method overcomes the shortcomings of existing technologies, which are prone to drift misjudgment, missed judgment, or ineffective adjustment, resulting in poor tension control stability and large fluctuations in wrapping quality. This method helps to accurately separate the drift component and the effective component in the tension signal, accurately identify the degree of drift anomaly, and achieve differentiated triggering for drift prevention and prediction. In this way, it improves the accuracy of tension drift judgment and the reliability of tension control, effectively suppresses the tension drift trend, and ensures the stability of cable wrapping tension.

[0013] 4. By using the received multi-source cable wrapping timing data as input to a preset machine learning model, the model outputs tension drift prediction data including tension drift risk value, drift direction, and warning level. When tension drift level one is detected, the model outputs the tension adjustment amount and adjustment step size and synchronously feeds them back to the controller and tension adjustment mechanism. Compared with the shortcomings of existing technologies that cannot achieve coordinated linkage between drift warning and real-time adjustment, leading to drift expansion and wrapping quality defects, this model helps to achieve coordinated linkage of accurate graded warning, direction prediction, and real-time adjustment of tension drift, improves the timeliness and accuracy of tension drift prevention and control, thereby avoiding tension drift accumulation, reducing cable wrapping forming defects, and improving the stability and wrapping efficiency of equipment operation.

[0014] 5. In the event of a short-term loss of the cable wrapping tension signal, an adjustment mechanism for the cable wrapping tension signal is triggered. When the signal-to-noise ratio (SNR) of the cable wrapping tension signal is not within the predefined SNR range, abnormal signal replacement processing is performed based on the neighborhood averaging method. Compared to existing technologies that can only simply filter or discard abnormal data, which can easily lead to the loss of effective signals, incomplete interference filtering, or distorted data completion, affecting the accuracy of subsequent tension drift identification and prediction, this method helps to achieve accurate processing of abnormal signals. While eliminating interference, it retains the effective features of the tension signal and completes invalid data, ensuring the integrity and purity of the tension signal. This improves the accuracy of subsequent tension signal decoupling, drift identification, and prediction, providing reliable signal support for tension adjustment and further ensuring the effectiveness of cable wrapping tension control. Attached Figure Description

[0015] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the overall process of the cable wrapping tension drift prediction method based on a machine learning model provided in this embodiment of the invention;

[0017] Figure 2 This is a schematic diagram illustrating the division of abnormal wrapping states provided in an embodiment of the present invention;

[0018] Figure 3 This is a schematic diagram of the adaptive adjustment of tension demand influence provided in an embodiment of the present invention;

[0019] Figure 4 This is a schematic diagram of the cable wrapping tension drift prediction system based on a machine learning model provided in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0021] Before providing a detailed explanation of the embodiments of this application, the application scenarios of these embodiments will be described first.

[0022] Example 1: This embodiment of the invention provides a method for predicting cable wrapping tension drift based on a machine learning model. For example... Figure 1 The diagram shown illustrates the overall process of a cable wrapping tension drift prediction method based on a machine learning model. The processing flow of this method may include the following steps:

[0023] S1, Tension Demand Impact Analysis: During the cable wrapping process, an analysis of the impact mechanism on cable wrapping tension demand is conducted. The influence of changes in coil diameter and traction line speed on cable wrapping tension demand is quantified, yielding corresponding quantitative impact results. A decision is made regarding whether to implement decoupling and drift identification analysis at the cable wrapping tension signal level. This tension demand impact analysis helps to accurately identify the core disturbance factors affecting tension demand caused by changes in coil diameter and traction line speed, quantifies their impact, clarifies the necessity of implementing tension signal decoupling and drift identification, avoids invalid analysis and redundant operations, and thus provides targeted data support and decision-making basis for subsequent tension signal processing and drift prediction.

[0024] S2, Tension Signal Decoupling and Drift Decision: When the decision result is to implement the action, based on the quantified impact of the output, decoupling and drift identification analysis are performed at the cable wrapping tension signal level to quantify the degree of influence of the cable wrapping tension signal on the cable wrapping tension demand. This yields data on the quantified influence of the cable wrapping tension signal on the cable wrapping tension demand. When the decision result is not to implement the action, the cable wrapping tension drift trend is predicted based on a machine learning model. By performing tension signal decoupling and drift decision, it is helpful to separate the effective component and irrelevant interference component in the tension signal that is dominated by the tension demand, accurately identify tension drift characteristics, determine the drift anomaly level, and clarify whether the tension adjustment mechanism needs to be triggered, thus avoiding wrapping quality defects caused by drift misjudgment or omission.

[0025] S3, Machine Learning and Tension Drift Prediction: Based on the received multi-source cable wrapping time-series data as the prediction benchmark, the machine learning model predicts the cable wrapping tension drift trend. The multi-source cable wrapping time-series data represents the data used to reflect the cable wrapping state during the cable wrapping process. By performing machine learning and tension drift prediction, it is helpful to rely on the multi-source wrapping time-series data to predict the tension drift trend in advance, accurately output key information such as drift amplitude and drift rate, realize the forward prevention and control of tension drift, and reserve sufficient response time for tension adjustment.

[0026] To ensure that the cable wrapping tension drift prediction method based on machine learning models accurately adapts to actual wrapping conditions, and to unify the benchmark standards of input and output parameters of various models, thereby improving the consistency and reliability of threshold determination, parameter matching, and drift prediction, and avoiding prediction errors caused by inconsistent parameter benchmarks, the embodiments of this invention require the prior construction of a database covering defined decoupling identification anomaly thresholds, etc. This database is used to store various benchmark parameters and threshold parameters adapted to wrapping conditions, providing data support for the training, prediction, and triggering of various adjustment mechanisms of the machine learning model.

[0027] In this embodiment, by linking the impact analysis of tension demand, decoupling and drift judgment of tension signals, and machine learning and tension drift prediction, it is helpful to construct a closed-loop control logic for impact analysis, decoupling judgment and prediction and prevention, so as to realize the synergistic linkage of tension demand disturbance identification, accurate tension drift judgment and forward-looking prediction, thereby effectively suppressing the drift trend of cable wrapping tension signal, reducing the amplitude of tension fluctuation, and ensuring the stability of cable wrapping forming quality.

[0028] The cable provided by this invention has three main characteristics. First, regarding material selection, the cable uses four materials: conductor, insulation, shielding, and outer protective layer. The cable developed in this project has high requirements for electrical performance, high and low temperature resistance, and lightweight design; therefore, material selection is crucial. The conductor is preferably made of thickened silver-plated soft copper wire, with a silver layer thickness consistency error not exceeding 0.0002 mm and a plating surface roughness lower than Ra2.0. The insulation material is preferably high-quality domestic F46, significantly improving the cable's temperature resistance and mechanical strength. The shielding layer uses ultra-thin aluminum foil Mylar, providing good shielding while reducing cable weight. The sheath is wrapped with a single layer of ultra-thin PET (Polyethylene Terephthalate) film, achieving the goal of lightweighting the cable.

[0029] Secondly, the cable structure design preferably adopts a double-line parallel placement with an external leakage wire in the middle of the parallel lines to reduce the cable volume. The ultra-thin shield and sheath design also greatly reduces the weight of the cable and the space occupied, bringing great convenience to the overall design and use of more precision equipment and flight devices.

[0030] Third, in the cable manufacturing process, the conductor is made of extremely fine silver-plated copper wire stranded together, providing a more flexible foundation for the cable and facilitating installation and laying within equipment; the stranding process uses a high-precision constant tension stranding machine, and the shielding layer is produced using a high-precision constant tension active release horizontal wrapping machine, strictly controlling the consistency of wrapping tension, speed, and accuracy. Ultra-thin aluminum foil Mylar is used as the shielding material, with the aluminum side facing inward and in contact with the ground wire. The wrapping angle is 55°~60°, the overlap rate is ≥25%, the pitch is 3mm, and the tension is 120±5 grams. Ultra-thin PET is used as the sheath material, with a wrapping angle of 55°~60°, an overlap rate of ≥25%, a pitch of 3.8mm, and a tension of 135±5 grams. After wrapping, it is sintered and sealed at high temperature of 145℃~150℃, with a traction speed of 4.5~5.0 meters / min.

[0031] Taking the production process of communication cables used in airborne devices (such as drones and aircraft) as an example, it typically places high demands on the cables for high bandwidth, low latency, impedance consistency, strong electromagnetic shielding, and bending life. For instance, a certain type of airborne high-speed differential communication cable may be designed with the following requirements: differential characteristic impedance of 100Ω±10Ω, insertion loss of no more than 0.6dB / m at 1GHz, and shielding effectiveness of no less than 60dB (100MHz~1GHz). When conditions such as rapid changes in roll diameter and frequent starts / stops / acceleration / deceleration occur, the supply tension and overlap position of the wrapping tape or shielding tape will rapidly fluctuate with transient acceleration and changes in roll diameter, causing the wrapping layer to alternate between tightness and looseness and instantaneous shift in overlap rate over a short distance. For example (illustrative example), setting a constant tension target of 20N and a target overlap rate of 30%±2%, during the wrapping process where the online speed gradually decreases from 0→2.5m / s, the acceleration is approximately 1.2m / s², and the roll diameter gradually decreases from 360mm to 140mm: During the start-up, stop, or acceleration phases, the tension may briefly surge to 28–30N (overshoot +40%–+50%), and during the deceleration phase, it may drop to 14–16N (undershoot -20%–-30%), lasting for 0.3–1.5s. Furthermore, when the overlap rate and the equivalent thickness of the wrapping layer fluctuate as described above over a short distance, it can easily lead to a decrease in shielding continuity, inconsistencies between local thickness and impedance, and a reduction in fatigue life. For example, localized voids / offsets in the cladding can cause discontinuities in shielding coverage, potentially leading to a 5–15 dB decrease in shielding effectiveness in certain frequency bands; deteriorating impedance consistency can cause the differential impedance to shift from the target 100Ω to 110Ω or 90Ω (exceeding the ±10Ω design tolerance), which may manifest as increased echo / crosstalk on 2–5Gbps links, and a decrease in the bit error rate. The magnitude rose to This can lead to risks such as increased error rates in communication links, insufficient anti-interference margins, or intermittent failures during service.

[0032] Furthermore, the specific process for analyzing the influence mechanism of cable wrapping tension demand is as follows: Based on the data acquisition equipment arranged at preset acquisition points on the wrapping machine, data reflecting abnormal wrapping tension of multi-source cables is collected; using the collected data as reference values ​​for classifying abnormal wrapping states, the abnormal wrapping states are divided into abnormal wrapping state intervals and qualified wrapping state intervals; the acquisition equipment is used to synchronously acquire quality characterization parameters during the cable wrapping process, including tension sensors, roll diameter sensors, temperature and humidity sensors, etc.; the abnormal wrapping state interval represents the acquisition time period corresponding to when the data reflecting abnormal wrapping tension of multi-source cables meets the abnormal wrapping state conditions; the qualified wrapping state interval represents the acquisition time period corresponding to when the data reflecting abnormal wrapping tension of multi-source cables does not meet the abnormal wrapping state conditions; the abnormal wrapping state conditions, i.e., the roll diameter change rate is not within the preset roll diameter change rate threshold range set by the personnel, or the traction line speed change rate is not within the preset roll diameter change rate threshold range set by the personnel, correspond to the acquisition time period.

[0033] The key addition is that, based on the collected data as reference values ​​for classifying abnormal wrapping states, abnormal wrapping states are divided into abnormal wrapping state intervals and qualified wrapping state intervals. The specific process is as follows: The ratio of the absolute value of the difference between the initial and final roll diameters monitored by the roll diameter sensor within a preset acquisition time interval to the corresponding duration of the preset acquisition time interval is used as the roll diameter change rate; the ratio of the absolute value of the difference between the initial and final traction linear speeds monitored by the encoder at the drive shaft end of the traction wheel or track within a preset acquisition time interval to the corresponding duration of the preset acquisition time interval is used as the traction linear speed change rate; in the state determination of the wrapping process, the roll diameter change rate and the traction linear speed change rate are used as the basic reference conditions for roll diameter change and traction linear speed change, respectively, to accurately determine the abnormal wrapping state.

[0034] like Figure 2The diagram illustrates the process of classifying abnormal cable wrapping states according to an embodiment of this invention: The roll diameter change rate is set as a key parameter reflecting the dynamic changes in roll diameter, and the traction line speed change rate is set as a key parameter reflecting the changes in traction line speed. When both the roll diameter change rate and the traction line speed change rate are detected within a preset threshold range set by a pre-defined personnel, the corresponding preset acquisition time interval is determined and recorded as a qualified wrapping state interval; conversely, the corresponding preset acquisition time interval is determined and recorded as an abnormal wrapping state interval. For the multi-source cable wrapping time sequence data corresponding to the qualified wrapping state interval, a machine learning model is used to predict the cable wrapping tension drift trend. For the multi-source cable wrapping time sequence data corresponding to the abnormal wrapping state interval, adaptive adjustments based on tension demand are implemented. These adaptive adjustments are used to adapt to rapid changes in roll diameter and frequent acceleration / deceleration of traction line speed, offset tension disturbances caused by mismatch between time-varying inertia and line speed, and mitigate tension drift trends in advance.

[0035] like Figure 3 The diagram shown is a schematic of the adaptive adjustment of tension demand influence provided in the embodiment of the present invention: the abnormal winding state interval is sequence extracted to obtain the abnormal winding sequence, the abnormal winding sliding window is filtered based on the extracted abnormal winding sequence, the sliding step size, the sliding window length and the sampling period are obtained for filtering, and the parameters corresponding to the wire feeding motor are set after the filtering operation.

[0036] Specifically, the adaptive adjustment process based on tension demand is as follows: Sequence extraction is performed on the abnormal wrapping state interval to obtain an abnormal wrapping sequence. Sequence extraction involves locating abrupt change points in the roll diameter change rate and traction line speed change rate within the abnormal wrapping state interval using an event segmentation algorithm, and then slicing the sequence according to the start and end boundaries of these abrupt change points. Specifically, the abnormal wrapping state interval is recorded as an abnormal event. Using the abnormal wrapping state interval as the core interval, a preset buffer length is extended before the start time and after the end time of this core interval to form an abnormal wrapping sequence. The abnormal wrapping sequence includes a roll diameter sequence (a time-series sequence composed of the roll diameters collected or obtained according to the chronological order of the cable wrapping time process) and a traction line speed sequence (a time-series sequence composed of the traction line speeds collected or obtained according to the chronological order of the cable wrapping time process). An abnormal wrapping sliding window filter is then applied based on the extracted abnormal wrapping sequence.

[0037] It should be added that the embodiments of this invention provide a series of models for outputting fixed parameters, including a preset wrapping sliding filter parameter model, a pre-constructed output model for the speed adjustment value of the wire feeding motor, a gain-pulse matching model, a preset machine learning model, a preset drift prediction adjustment model, a signal-to-noise ratio-gain pulse matching model, etc. In practical applications, it is only necessary to input the corresponding input parameters to the corresponding preset wrapping sliding filter parameter model, the pre-constructed output model for the speed adjustment value of the wire feeding motor, the gain-pulse matching model, the preset machine learning model, the preset drift prediction adjustment model, the signal-to-noise ratio-gain pulse matching model, etc.

[0038] The input parameters include the abnormal wrapping sequence and the combination of the sampling frequency corresponding to the abnormal wrapping sequence, time-varying feature components, tension signal decoupling-identification of abnormal values, multi-source cable wrapping timing data, tension drift prediction data, tension signal decoupling-identification of abnormal values, and the signal-to-noise ratio corresponding to the cable wrapping tension signal, etc. The output parameters include the appropriate combination of sliding step size, sliding window length and sampling period, cable laying motor speed, the combination of the signal gain amplification factor of the tension sensor and the pulse width threshold corresponding to the cable wrapping tension signal, tension drift prediction data, tension adjustment amount and adjustment step size, the signal gain amplification factor of the tension sensor and the pulse width threshold corresponding to the cable wrapping tension signal, etc.

[0039] To improve the generalization ability and prediction accuracy of the output models for each parameter, and to reduce the tension drift judgment error caused by signal noise, operating condition disturbances, and parameter mismatch, thereby achieving accurate prediction of cable wrapping tension drift, it is necessary to collect training data in advance. This training data collection includes: first, collecting training data from historical time periods, including abnormal wrapping sequences, the sampling frequency and preset sliding step size corresponding to the abnormal wrapping sequences, combinations of sliding window length and sampling period, combinations of time-varying feature components and preset cable-laying motor speeds, tension signal decoupling and abnormal value identification, combinations of preset tension sensor signal gain amplification factor and pulse width threshold corresponding to the cable wrapping tension signal, combinations of multi-source cable wrapping time-series data and preset tension drift prediction data, combinations of tension drift prediction data and preset tension adjustment amount and adjustment step size, combinations of tension signal decoupling and abnormal value identification, signal-to-noise ratio corresponding to the cable wrapping tension signal, and combinations of preset tension sensor signal gain amplification factor and pulse width threshold corresponding to the cable wrapping tension signal.

[0040] Based on machine learning algorithms, such as random forests, the collected training data is cleaned, denoised, and normalized. Feature vectors such as tension are constructed and the training set and validation set are divided. Within the preset hyperparameter space, parameters such as the number of trees, maximum depth, minimum number of leaf node samples, and feature sampling ratio are optimized and the model is trained. Then, based on the validation set, error evaluation and overfitting verification are performed, and the optimal model parameters are solidified to finally obtain the desired model.

[0041] Anomaly wrapping sliding window filtering refers to filtering anomaly wrapping sequences based on the adjustment parameters output by a preset wrapping sliding filter parameter model. The specific process is as follows: The sampling frequencies corresponding to the anomaly wrapping sequences and those monitored by a multi-channel high-speed data acquisition instrument are input to the preset wrapping sliding filter parameter model, which is used to obtain the matching relationship between the sliding step size, sliding window length, and sampling period. The model outputs configurable sliding step size, sliding window length, and sampling period. Filtering is performed based on a sliding weighted average algorithm. The sliding step size controls the movement interval of the filtering window on the time sequence, and the sliding window length limits the single weighted average. The number of sampling points covered is calculated, and the sampling period is used to calibrate the time base of the time series data within the filtering window. The coordinated configuration of the three can adapt to the sequence characteristics under different abnormal winding conditions, improve the smoothing effect of filtering on abnormal fluctuations in roll diameter and traction line speed, and retain effective time series features. After completing the abnormal winding sliding window filtering operation for abnormal winding conditions, based on the filtering results, the extracted time-varying feature components are input into the pre-constructed output model of the wire feeding motor speed adjustment value, and the wire feeding motor speed for the next preset time period is output. This is used to set the parameters corresponding to the wire feeding motor, and after the setting is completed, the decoupling and drift identification analysis mode at the tension signal level is switched.

[0042] In this embodiment, data reflecting multi-source cable wrapping tension anomalies, namely the roll diameter change rate and the traction line speed change rate, are obtained by analyzing the influence mechanism of cable wrapping tension demand. When both the roll diameter change rate and the traction line speed change rate are within a preset roll diameter change rate threshold range, the cable wrapping tension drift trend is predicted based on a machine learning model; otherwise, adaptive adjustments based on the influence of tension demand are adopted. This helps to accurately determine the stability of cable wrapping conditions, distinguish between normal and abnormal conditions, avoid invalid tension adjustment operations and redundant drift prediction calculations, and proactively prevent the accumulation of tension drift caused by untimely adjustments due to abnormal conditions. This effectively ensures the stability of cable wrapping tension and reduces cable forming defects caused by tension drift during the wrapping process. By directly judging the roll diameter change rate and the traction line speed change rate to determine whether to adopt adaptive adjustments based on the influence of tension demand, it helps to identify potential risks of tension drift in advance, shorten the response time of tension adjustment, thereby reducing the impact of tension drift on wrapping quality and improving the real-time performance and reliability of tension control.

[0043] By adopting an adaptive adjustment based on the influence of tension demand, abnormal wrapping sequences are first obtained. The extracted abnormal wrapping sequences are then filtered using an abnormal wrapping sliding window. This filtering is based on a sliding weighted average algorithm and the obtained sliding step size, sliding window length, and sampling period. Compared to existing methods that typically suffer from fixed filtering parameters, inability to adapt to dynamic changes in abnormal wrapping sequences, loss of effective working condition features during filtering, or incomplete interference removal, this method helps to accurately filter out interference components in abnormal wrapping sequences while preserving effective time-varying features of the roll diameter and traction line speed. It avoids the impact of filtering distortion on subsequent tension adjustments, thereby improving the accuracy of tension control under abnormal working conditions. After filtering, the corresponding speed of the pay-off motor is set, and after setting, the method switches to a decoupling and drift identification analysis mode at the tension signal level. This helps improve the synergy between tension adjustment and signal decoupling, making tension control more closely aligned with actual wrapping requirements. It accurately identifies the demand-driven and drift components in the tension signal, enabling targeted tension drift prevention and adjustment.

[0044] By analyzing the influence mechanism of cable wrapping tension demand, predicting the cable wrapping tension drift trend based on machine learning models, and taking adaptive adjustments based on the influence of tension demand, the interrelationship and interaction among these three aspects help improve the overall integrity and synergy of cable wrapping tension control. This enables closed-loop linkage of working condition identification, drift prediction, and adaptive adjustment, thereby facilitating accurate prediction and effective suppression of cable wrapping tension drift and ensuring the consistency and stability of cable wrapping quality.

[0045] Furthermore, the decoupling and drift identification analysis at the tension signal level specifically involves: acquiring parameters that quantify the decoupling and identification status of the cable wrapping tension signal, i.e., the tension signal decoupling-identification anomaly value. The obtained parameters are used as the criteria for classifying the anomaly level of the cable wrapping tension signal to determine whether the adjustment mechanism of the cable wrapping tension signal is triggered. The tension signal decoupling-identification anomaly value is represented by the ratio of the spectral energy (in (Newtons squared per second)) of the cable wrapping tension signal monitored by the tension sensor and spectrum analyzer in the preset low-frequency band to the total spectral energy of the cable wrapping tension signal. The larger the tension signal decoupling-identification anomaly value, the higher the proportion of low-frequency components in the tension signal, and the more likely the tension baseline is to show a slow and continuous drift trend. This indicates that the drift components caused by time-varying roll diameter, frictional heat drift, or sensor zero / gain offset are more significant, thus resulting in a higher risk of tension anomaly drift.

[0046] To accurately separate the effective component from the drift and interference components in the cable wrapping tension signal, objectively quantify the degree of tension signal anomaly, clarify the criteria for classifying tension signal anomaly levels, and rationally determine the triggering timing of the adjustment mechanism to avoid false or missed triggering of adjustment operations, thus ensuring the accuracy and reliability of tension signal processing and drift identification, the specific process for determining whether to trigger the cable wrapping tension signal adjustment mechanism is as follows: The decoupling-identification anomaly value of the tension signal is compared with a defined decoupling-identification anomaly threshold, and the comparison result is output. The defined decoupling-identification anomaly threshold is represented by the average value of the decoupling-identification anomaly values ​​of the tension signal over a historical time period. When the tension... When the signal decoupling-identification anomaly value exceeds the defined decoupling-identification anomaly threshold, the output comparison result triggers the cable wrapping tension signal adjustment mechanism. This mechanism corrects the drift component deviation in the cable wrapping tension signal, suppresses tension fluctuations, and compensates for tension offsets caused by time-varying disturbances in the coil diameter and traction line speed. When the tension signal decoupling-identification anomaly value is not greater than the defined decoupling-identification anomaly threshold, the output comparison result predicts the cable wrapping tension drift trend based on a machine learning model. The comparison result includes both the triggering of the cable wrapping tension signal adjustment mechanism and the prediction of the cable wrapping tension drift trend based on a machine learning model.

[0047] It should be added that the specific process of triggering the adjustment mechanism of the cable wrapping tension signal is as follows: The identified tension signal decoupling-identification anomaly value that is greater than the defined decoupling identification anomaly threshold is imported as an input parameter into the gain-pulse matching model; the gain-pulse matching model outputs the signal gain amplification factor of the tension sensor and the pulse width threshold corresponding to the cable wrapping tension signal based on the input value; according to the output result of the model, the signal gain amplification factor of the tension sensor and the pulse width threshold corresponding to the cable wrapping tension signal are configured accordingly for the next preset time period; after the configuration operation is completed, the cable wrapping tension drift trend is predicted based on the machine learning model.

[0048] Based on the model's output, the signal gain amplification factor of the tension sensor and the pulse width threshold corresponding to the cable wrapping tension signal are configured accordingly for the next preset time period. This helps to improve the acquisition accuracy and recognition of the tension sensor signal, accurately filter out spike pulse interference in the tension signal, reduce the adverse effects of signal distortion on tension drift identification and prediction, optimize the effectiveness and reliability of the tension signal, thereby improving the accuracy of cable wrapping tension drift prediction and the response efficiency of the tension signal adjustment mechanism, ensuring the pertinence and rationality of tension adjustment, and further suppressing the tension drift trend.

[0049] In this embodiment, decoupling and drift identification analysis at the tension signal level is performed to obtain the decoupling-identification anomaly value of the tension signal. When the decoupling-identification anomaly value of the tension signal exceeds the defined decoupling identification anomaly threshold, the adjustment mechanism of the cable wrapping tension signal is triggered. Conversely, the cable wrapping tension drift trend is predicted based on a machine learning model. This helps to accurately separate the drift component and transient fluctuation component in the tension signal, accurately identify the degree of tension drift anomaly, avoid drift misjudgment, missed judgment and invalid adjustment operation, and achieve targeted prevention and forward prediction of tension drift. This ensures the stability and accuracy of the cable wrapping tension signal, reduces cable wrapping forming defects caused by tension drift, and improves the consistency of wrapping quality.

[0050] Furthermore, the prediction of cable wrapping tension drift trend based on a machine learning model means that, based on the acquired multi-source cable wrapping time-series data, a preset machine learning model is used to output tension drift prediction data within a prediction time window. The process of predicting cable wrapping tension drift trend based on a machine learning model is as follows: The received multi-source cable wrapping time-series data (such as tension, traction, or unwinding speed) is used as the input of the preset machine learning model, and the output includes tension drift prediction data including tension drift risk value, drift direction, and warning level. The tension drift risk value represents the probability that the wrapping tension will drift within the preset prediction time window. The drift direction includes positive drift and negative drift. Among them, positive drift indicates that the wrapping tension is increasing relative to the preset steady-state reference tension, and negative drift indicates that the wrapping tension is decreasing relative to the preset steady-state reference tension. The warning level includes tension drift level one and tension drift level two, which are used to quantify the risk level of tension drift during the cable wrapping process. The risk level of tension drift level one is higher than that of tension drift level two.

[0051] The method for predicting cable wrapping tension drift based on a machine learning model also includes: when tension drift level 1 is detected, the corresponding tension drift prediction data is input into a preset drift prediction and adjustment model, and the tension adjustment amount and adjustment step size are output and simultaneously fed back to the controller and tension adjustment mechanism (gallery arm / tension arm, etc.) as the tension adjustment amount (the adjustment amount directly output based on the model's predicted future tension deviation, such as tension adjustment amount: +200N, adjustment step size: 50N each time) and adjustment step size, to adjust the tension for the next time period; when tension drift level 2 is detected, a cable wrapping tension qualified prompt is sent, and the cable wrapping process continues to be monitored; tension drift level 1 indicates the warning level corresponding to when the output tension drift risk value is greater than the predefined tension drift risk value; tension drift level 2 indicates the warning level corresponding to when the output tension drift risk value is not greater than the predefined tension drift risk value, wherein the predefined tension drift risk value is represented by the average tension drift risk value over a historical time period.

[0052] like Figure 4 The diagram shows a schematic of the structure of a cable wrapping tension drift prediction system based on a machine learning model provided in an embodiment of this invention. It includes: a tension demand influence analysis module, used to analyze the influence mechanism of cable wrapping tension demand during the cable wrapping process, quantify the influence of changes in roll diameter and traction line speed on the cable wrapping tension demand, obtain the corresponding quantified influence results, and obtain a decision result on whether to implement decoupling and drift identification analysis at the cable wrapping tension signal level; a tension signal decoupling and drift decision module, used to perform decoupling and drift identification analysis at the cable wrapping tension signal level to quantify the influence of the cable wrapping tension signal on the cable wrapping tension demand when the decision result is to implement it, based on the output quantified influence results, to obtain data on the quantified influence of the cable wrapping tension signal on the cable wrapping tension demand, and to predict the cable wrapping tension drift trend based on a machine learning model when the decision result is not to implement it; and a machine learning and tension drift prediction module, used to predict the cable wrapping tension drift trend based on a machine learning model using received multi-source cable wrapping time-series data as a prediction benchmark, where the multi-source cable wrapping time-series data represents data reflecting the cable wrapping state during the cable wrapping process.

[0053] In this embodiment, a preset machine learning model is used to output tension drift prediction data within a prediction time window, obtaining tension drift risk value, drift direction, and warning level. Simultaneously, upon detecting level one tension drift, the output tension adjustment amount and adjustment step size are fed back to the controller and tension adjustment mechanism, helping to improve the real-time performance and accuracy of tension drift control, achieving rapid response and precise adaptation of tension adjustment, preventing further expansion of tension drift, and reducing the impact of drift accumulation on cable wrapping quality. Upon detecting level two tension drift, a cable wrapping tension qualification notification is sent. This technology helps improve the efficiency of cable wrapping condition monitoring, reduces redundant adjustment operations when there is no drift risk, lowers equipment energy consumption and adjustment losses, and ensures the smoothness of the wrapping process. Compared with existing technologies, which have shortcomings such as no graded warning, disconnect between warning and adjustment, inability to differentiate processing according to drift level, and easy over-adjustment or missed drift risk, this technology helps to achieve graded warning and differentiated processing of tension drift, accurately match the prevention and control needs of different drift levels, improve the accuracy of tension drift warning and the synergy of tension adjustment, and further ensure the stability of cable wrapping tension and the consistency of wrapping quality.

[0054] Example 2, based on the method of Example 1, aims to accurately monitor the purity and effectiveness of the cable wrapping tension signal, promptly identify signal anomalies caused by short-term signal loss, clarify the triggering conditions for abnormal signal replacement, and avoid interference from abnormal signals on subsequent tension drift identification, prediction, and adjustment operations. The specific process for triggering the adjustment mechanism of the cable wrapping tension signal is as follows: The signal-to-noise ratio (SNR) of the cable wrapping tension signal at a preset time point is monitored using a tension sensor, spectrum analyzer, and industrial control computer (signal processing unit). Whether the SNR exceeds a predefined tension signal SNR range is used as the criterion for determining whether to perform abnormal signal replacement. When the SNR of the cable wrapping tension signal is not within the predefined tension signal SNR range set by the pre-defined personnel, the corresponding cable wrapping tension signal is defined as... Abnormal signals are replaced using a neighborhood averaging method. For example, the average of the tension signal sampling points adjacent to the abnormal signal point is calculated, and this average is used to replace the abnormal signal, thereby eliminating spike pulse interference in the tension signal. When the signal-to-noise ratio (SNR) of the cable wrapping tension signal is within a predefined SNR range, the abnormal value of the tension signal and the SNR of the cable wrapping tension signal are decoupled and identified, and then input into the SNR-gain pulse matching model. The model outputs the signal gain amplification factor of the tension sensor and the pulse width threshold of the cable wrapping tension signal. The model then configures the signal gain amplification factor of the tension sensor and the pulse width threshold of the cable wrapping tension signal for the next preset time period, and predicts the cable wrapping tension drift trend based on a machine learning model after configuration.

[0055] In this embodiment, when there is a short-term loss of sampling in the cable wrapping tension signal (such as the number of lost sampling points being greater than a preset loss threshold, where the number of lost sampling points is obtained by counting the difference between the number of valid sampling points actually collected within the preset sampling time window and the preset number of sampling points using a counter), the cable wrapping tension signal adjustment mechanism is triggered. First, the signal-to-noise ratio (SNR) corresponding to the cable wrapping tension signal is obtained. If the SNR corresponding to the cable wrapping tension signal is not within the predefined tension signal SNR range, the corresponding cable wrapping tension signal is defined as an abnormal signal. Abnormal signal replacement processing is performed based on the neighborhood averaging method, which helps improve the integrity and purity of the cable wrapping tension signal, eliminates spike pulse interference, compensates for signal loss caused by short-term sampling loss, and avoids interference from abnormal signals on the tension drift identification, prediction, and adjustment mechanisms. This ensures the accuracy of tension signal decoupling and drift identification analysis, providing reliable signal support for subsequent tension drift prediction and adjustment operations.

[0056] When the signal-to-noise ratio (SNR) of the cable wrapping tension signal is within the predefined SNR range, relevant configurations based on the signal gain amplification factor of the output tension sensor and the pulse width threshold of the cable wrapping tension signal help improve the sensitivity and anti-interference capability of the tension sensor signal acquisition, accurately filter out residual slight spike pulse interference, optimize the amplitude identification and effectiveness of the tension signal, thereby improving the accuracy of tension drift prediction and the rationality of the tension adjustment mechanism response, further suppressing the tension drift trend, and ensuring the stability of the cable wrapping tension.

[0057] The above-disclosed embodiments are merely some examples of the present invention and should not be construed as limiting the scope of the present invention. Therefore, any equivalent variations made in accordance with the claims of the present invention are still within the scope of the present invention.

Claims

1. A method for predicting cable wrapping tension drift based on a machine learning model, characterized in that, The method includes: S1, conduct an analysis of the influence mechanism of cable wrapping tension demand, quantify the influence of changes in coil diameter and traction speed on cable wrapping tension demand, obtain the corresponding quantitative influence results, and obtain the decision results on whether to implement decoupling and drift identification analysis at the cable wrapping tension signal level. S2, when the decision is to implement, based on the result of the quantitative impact of the output, decoupling and drift identification analysis are performed at the level of cable wrapping tension signal to obtain data on the influence of the quantified cable wrapping tension signal on the cable wrapping tension demand. When the decision is not to implement, the drift trend of cable wrapping tension is predicted based on the machine learning model. S3, based on the received multi-source cable wrapping timing data as the prediction benchmark, predict the cable wrapping tension drift trend based on the machine learning model; The decoupling and drift identification analysis at the tension signal level is specifically represented as follows: Parameters for the decoupling and identification of the cable wrapping tension signal are obtained and used as the criteria for classifying the abnormal level of the cable wrapping tension signal to determine whether the adjustment mechanism of the cable wrapping tension signal is triggered. The specific process of the adjustment mechanism for the trigger cable wrapping tension signal is as follows: The identified tension signal decoupling-identification anomaly values ​​that exceed the defined decoupling identification anomaly threshold are imported as input parameters into the gain-pulse matching model; The gain-pulse matching model is based on the input value, the signal gain amplification factor of the output tension sensor and the pulse width threshold corresponding to the cable wrapping tension signal; Based on the model's output, the signal gain amplification factor of the tension sensor and the pulse width threshold corresponding to the cable wrapping tension signal are configured accordingly. After the configuration is completed, the cable wrapping tension drift trend is predicted based on the machine learning model. The prediction of cable wrapping tension drift trend based on machine learning model means that, based on the acquired multi-source cable wrapping time series data, a preset machine learning model is used to output the tension drift prediction data within the prediction time window. The process of predicting the cable wrapping tension drift trend based on a machine learning model is as follows: The received multi-source cable wrapping time series data is used as the input of a preset machine learning model, and the output includes tension drift prediction data including tension drift risk value, drift direction and warning level. The tension drift risk value represents the probability that the wrapping tension will drift within a preset prediction time window; The drift direction includes positive drift and negative drift; The warning levels include tension drift level one and tension drift level two; The method for predicting cable wrapping tension drift trend based on machine learning models also includes: When tension drift level 1 is detected, the corresponding tension drift prediction data is input into the preset drift prediction adjustment model, and the tension adjustment amount and adjustment step size are output and synchronously fed back to the controller and tension adjustment mechanism as the tension adjustment amount and adjustment step size corresponding to the prediction time window, so as to adjust the tension in the next time period. When tension drift level 2 is detected, a cable wrapping tension qualified indication is sent; The tension drift level 1 indicates the warning level corresponding to when the output tension drift risk value is greater than the predefined tension drift risk value; The tension drift level 2 indicates the warning level corresponding to when the output tension drift risk value is not greater than the predefined tension drift risk value.

2. The cable wrapping tension drift prediction method based on a machine learning model as described in claim 1, characterized in that, The specific process for analyzing the influence mechanism of cable wrapping tension demand is as follows: Based on the data acquisition equipment arranged at the preset acquisition points of the cable wrapping machine, data reflecting abnormal cable wrapping tension from multiple sources is collected. Based on the collected data as reference values ​​for classifying abnormal wrapping states, the abnormal wrapping states are divided into abnormal wrapping state intervals and qualified wrapping state intervals. The abnormal wrapping state interval represents the time period during which the data reflecting the abnormal wrapping tension of multi-source cables meets the abnormal wrapping state conditions. The wrapped cable qualified status interval represents the time period during which the data reflecting the abnormal wrapping tension of the multi-source cable does not meet the abnormal wrapping status conditions.

3. The cable wrapping tension drift prediction method based on a machine learning model as described in claim 2, characterized in that, The collected data is used as a reference value for classifying abnormal wrapping states. The abnormal wrapping states are then divided into abnormal wrapping state intervals and qualified wrapping state intervals. The specific process is as follows: The ratio of the absolute value of the difference between the initial and final roll diameters monitored within the preset acquisition time interval to the corresponding duration of the preset acquisition time interval is used as the roll diameter change rate. The ratio of the absolute value of the difference between the initial and final traction line velocities monitored within a preset acquisition time interval to the corresponding duration of the preset acquisition time interval is used as the traction line velocity change rate. In the determination of the state of the wrapping process, the roll diameter change rate and the traction line speed change rate are used as the basic roll diameter change reference conditions and traction line speed change reference conditions, respectively, to carry out accurate determination of the abnormal state of the wrapping. The roll diameter change rate is set as a key parameter reflecting the dynamic change characteristics of the roll diameter, and the traction line speed change rate is set as a key parameter reflecting the change characteristics of the traction line speed. When the roll diameter change rate is detected to be within the preset roll diameter change rate threshold range, and the traction line speed change rate is detected to be within the preset roll diameter change rate threshold range, the corresponding preset acquisition time interval is determined and recorded as the wrapping qualified state interval; otherwise, the corresponding preset acquisition time interval is determined and recorded as the wrapping abnormal state interval. For the multi-source cable wrapping timing data corresponding to the qualified wrapping state interval, the cable wrapping tension drift trend is predicted based on the machine learning model. For the multi-source cable wrapping timing data corresponding to the abnormal wrapping state interval, an adaptive adjustment based on the influence of tension demand is adopted.

4. The cable wrapping tension drift prediction method based on a machine learning model as described in claim 3, characterized in that, The specific process for adaptive adjustment based on the influence of tension demand is as follows: Sequence extraction is performed on the abnormal wrapping state interval to obtain the abnormal wrapping sequence; The sequence extraction refers to locating the abrupt change points of the roll diameter change rate and traction line speed change rate within the abnormal state interval of the wrapping based on the event segmentation algorithm, and extracting slices according to the start and end boundaries of the abrupt change points. Anomaly wrapping sliding window filtering is performed based on the extracted anomaly wrapping sequence; The abnormal wrapping sliding window filtering refers to the filtering process applied to the abnormal wrapping sequence based on the adjustment parameters output by the preset wrapping sliding filter parameter model. The specific process is as follows: The abnormal wrapping sequence and the sampling frequency corresponding to the abnormal wrapping sequence are input into the preset wrapping sliding filter parameter model used to obtain the matching relationship of sliding step size, sliding window length and sampling period. The output is a configurable sliding step size, sliding window length and sampling period, and filtering is performed based on the sliding weighted average algorithm. After completing the abnormal wrapping sliding window filtering operation, based on the filtering results, the extracted time-varying feature components are input into the pre-constructed output model of the wire feeding motor speed adjustment value, and the wire feeding motor speed is output. This is used to set the corresponding parameters of the wire feeding motor, and after the setting is completed, the decoupling and drift identification analysis mode at the tension signal level is switched.

5. The cable wrapping tension drift prediction method based on a machine learning model as described in claim 1, characterized in that, The process for determining whether the cable wrapping tension signal adjustment mechanism is triggered is as follows: The abnormal value of the tension signal is decoupled and identified, and compared with the defined decoupling and abnormal identification threshold. The comparison result is then output. When the decoupling-identification anomaly value of the tension signal exceeds the defined decoupling-identification anomaly threshold, the output comparison result triggers the adjustment mechanism of the cable wrapping tension signal. When the decoupling-identification anomaly value of the tension signal is not greater than the defined decoupling-identification anomaly threshold, the output comparison result is a prediction of the cable wrapping tension drift trend based on a machine learning model. The comparison results include the adjustment mechanism that triggers the cable wrapping tension signal and the prediction of cable wrapping tension drift trend based on machine learning models.

6. The cable wrapping tension drift prediction method based on a machine learning model as described in claim 5, characterized in that, The specific process of the adjustment mechanism for the trigger cable wrapping tension signal is as follows: The signal-to-noise ratio of the cable wrapping tension signal at a preset time point is monitored, and whether the signal-to-noise ratio exceeds the predefined tension signal signal-to-noise ratio range is used as the criterion to determine whether to replace the abnormal signal. When the signal-to-noise ratio of the cable wrapping tension signal is not within the predefined tension signal signal-to-noise ratio range, the corresponding cable wrapping tension signal is defined as an abnormal signal, and abnormal signal replacement processing is performed based on the neighborhood mean method. When the signal-to-noise ratio (SNR) of the cable wrapping tension signal is within the predefined SNR range, the tension signal is decoupled and abnormal values ​​are identified, and the SNR of the cable wrapping tension signal is input to the SNR-gain pulse matching model. The output is the signal gain amplification factor of the tension sensor and the pulse width threshold of the cable wrapping tension signal. The signal gain amplification factor of the tension sensor and the pulse width threshold of the cable wrapping tension signal are configured accordingly. After the configuration is completed, the cable wrapping tension drift trend is predicted based on the machine learning model.

7. A cable wrapping tension drift prediction system based on a machine learning model, used to implement the cable wrapping tension drift prediction method based on a machine learning model as described in any one of claims 1-6, characterized in that, include: The tension demand impact analysis module is used to analyze the impact mechanism of cable wrapping tension demand, quantify the impact of changes in coil diameter and traction line speed on cable wrapping tension demand, obtain the corresponding quantitative impact results, and obtain the decision results of whether to implement decoupling and drift identification analysis at the cable wrapping tension signal level. The tension signal decoupling and drift decision module is used to perform decoupling and drift identification analysis on the cable wrapping tension signal level based on the output quantified influence results when the decision result is to implement it. This results in data on the quantified influence of cable wrapping tension signal on cable wrapping tension demand. When the decision result is not to implement it, the module predicts the cable wrapping tension drift trend based on a machine learning model. The machine learning and tension drift prediction module is used to predict the cable wrapping tension drift trend based on the received multi-source cable wrapping time series data as a prediction benchmark and a machine learning model.

Images