A lateral pressure encoder early warning method for cable laying

Abstract

The application relates to a cable laying side pressure encoder early warning method, a multi-axis pressure sensing array is arranged at a bridge corner key point of a cable laying path, three-dimensional dynamic side pressure analog signals applied when a cable passes through are collected in real time; the analog signals are input into an analog-digital conversion module, a dynamic calibration algorithm is adopted for nonlinear correction, and a digitalized real-time side pressure value sequence with a range of 0-500N continuously adjustable is generated; through a multi-thread fusion processing module, the side pressure value sequence, laying length measurement values and speed control instructions are combined to construct a dynamic pressure-length-speed three-dimensional relationship matrix; spectrum feature analysis is performed on the three-dimensional relationship matrix, when it is detected that the energy of an abnormal frequency component exceeds a preset risk threshold, a spatial positioning hierarchical early warning signal is triggered, and a spectrum diagnosis report containing a failure point spatial coordinate is generated.

Description

Technical Field

[0001] This application relates to the field of electrical digital data processing technology, specifically to a method for early warning of side pressure encoders in cable laying. Background Technology

[0002] In March 2025, during the landscape improvement project of Chongbu Village and Xingang Village in Mingcheng Town, it was found that in the existing technology of cable laying equipment control and monitoring systems, incremental encoders are used to indirectly monitor and control factors related to cable laying side pressure. This is mainly reflected in the fact that the control of cable laying speed is crucial for the control of side pressure. The existing technology has the following defects:

[0003] It cannot directly measure the actual lateral pressure, but can only indirectly infer the pressure through speed and bending angle, and cannot monitor the dynamic lateral pressure value at key locations (cable tray corners) in real time; the control precision is insufficient. If the speed is too fast, it will cause a sudden increase in lateral pressure (>500N), and if it is too slow, it will increase friction and indirectly raise the pressure, which can easily cause cable wear or equipment failure; it lacks the ability to locate abnormalities, which means that it cannot identify the abnormal pressure frequency components and their spatial location, making it difficult to provide timely warnings and diagnose failure points. Summary of the Invention

[0004] In order to solve the problems existing in the prior art, the purpose of this application is to provide a side pressure encoder early warning method for cable laying.

[0005] To address the shortcomings of the existing technology, the present invention proposes the following solution: deploy a multi-axis pressure sensor array at the corner of the cable tray to acquire three-dimensional dynamic side pressure signals in real time; generate a 0-500N digital pressure sequence through a dynamic calibration algorithm, and output the laying length using an incremental encoder (±0.1%FS accuracy); construct a three-dimensional matrix of pressure-length-speed and perform spectrum analysis; when the abnormal frequency energy exceeds the threshold, trigger a spatial positioning hierarchical early warning and generate a spectrum diagnostic report containing the coordinates of the failure point.

[0006] The side pressure encoder early warning method for cable laying described in this application includes the following steps:

[0007] S101. Deploy a multi-axis pressure sensor array at key points of the cable tray corners along the cable laying path to collect real-time three-dimensional dynamic lateral pressure simulation signals applied when the cable passes through.

[0008] S102. Input the analog signal into the analog-to-digital conversion module, use the dynamic calibration algorithm to perform nonlinear correction, and generate a digital real-time side pressure value sequence with a range of 0-500N that is continuously adjustable.

[0009] S103. The cable displacement pulse signal is acquired in real time by an incremental encoder. Combined with a temperature-compensated calibration algorithm, the pulse counting is jittered and the cumulative error is compensated to output a laying length measurement value accurate to ±0.1%FS.

[0010] S104. Based on the preset time-varying pressure threshold model, a sliding window is used to monitor the real-time side pressure value sequence: when the pressure value continuously exceeds the upper limit threshold, a closed-loop deceleration control command is generated to drive the servo motor; when the pressure value continues to be lower than the lower limit threshold, a fuzzy acceleration control command is generated to optimize the tension balance.

[0011] S105. The multi-threaded fusion processing module combines the side pressure value sequence, laying length measurement value and speed control command to construct a dynamic pressure-length-speed three-dimensional relationship matrix.

[0012] S106. Perform spectral feature analysis on the three-dimensional relationship matrix. When the energy of abnormal frequency components exceeds the preset risk threshold, trigger a spatial positioning graded early warning signal and generate a spectral diagnostic report containing the spatial coordinates of the failure point.

[0013] Preferably, in step S101, at the corner of the cable tray, a multi-axis pressure sensor array is used to collect the three-dimensional dynamic lateral pressure signal generated by the cable passing through in real time, forming an original pressure dataset. When the signal strength exceeds a preset threshold, the system immediately uses Fast Fourier Transform to perform frequency domain decomposition to obtain a frequency domain feature set. Subsequently, the principal component analysis algorithm is used to extract the main pressure distribution patterns from the frequency domain features, generating a dimensionality-reduced feature vector set. If the variance ratio of the principal components meets the standard, the feature vectors are smoothed and optimized using Kalman filtering to establish a reliable pressure distribution model. Based on this model, the key dynamic mechanical parameters of the cable passing through are calculated to form a mechanical behavior dataset. The system further uses time series analysis to predict the pressure change trend, obtains a pressure prediction sequence, and accurately identifies the pressure distribution anomalies at the corner, ultimately obtaining an anomaly detection report.

[0014] Preferably, in step S102, the analog-to-digital conversion module samples the analog signal at a fixed frequency to generate a digital side pressure sequence. If the sequence has abnormal fluctuations, a mean-mode filtering algorithm is immediately applied for smoothing to obtain a smoothed side pressure sequence. Subsequently, based on the smoothed sequence, periodic components are extracted through time series decomposition to form a periodic feature sequence. When the fluctuation amplitude of this sequence exceeds a preset threshold, the system automatically calls an autoregressive model to predict future pressure changes and outputs a pressure change prediction sequence. Then, the sliding window method is used to calculate the local pressure change rate of the prediction sequence to generate a pressure change rate sequence. Once a sudden change point is detected in the change rate sequence, the anomaly detection algorithm immediately locates its position to form an anomaly point distribution sequence. Finally, a cluster analysis method is applied to classify the anomalies in the distribution sequence to obtain an anomaly point classification set.

[0015] Preferably, in step S103, the system acquires cable displacement pulse signals in real time at a fixed sampling frequency using an incremental encoder, generating a pulse counting sequence. If an abnormal jump is detected in the sequence, a mean filtering algorithm is immediately applied for smoothing to obtain a smooth pulse sequence. Subsequently, based on the smooth pulse sequence and combined with real-time ambient temperature data, a temperature compensation algorithm is used to correct the counting deviation caused by temperature changes, generating a more accurate correction pulse sequence. Then, a sliding window method is applied to the correction pulse sequence to calculate the local displacement change rate, forming a displacement change rate sequence. When a sudden change point appears in the sequence, the anomaly detection algorithm automatically identifies its location and outputs a sudden change point distribution sequence. Further, a cluster analysis method is used to classify the sudden change points in this distribution sequence to obtain an anomaly classification set. Finally, the system determines the type in the anomaly classification set according to a preset threshold to obtain a cable laying length measurement value sequence.

[0016] Preferably, in step S104, the side pressure value is collected in real time at a fixed frequency to form a pressure sequence. Then, the sliding window method is applied to perform segmented analysis on the sequence, calculate the average pressure within the window, and generate a pressure average sequence. Next, this sequence is compared with a preset time-varying pressure threshold model (including upper and lower thresholds) to determine the triggering condition. When the average pressure continuously exceeds the upper threshold, the system immediately uses the proportional-integral-derivative (PID) control algorithm to generate a closed-loop deceleration command sequence.

[0017] If the mean value remains below the lower threshold, a fuzzy control algorithm is automatically applied to generate an acceleration command sequence. Finally, the deceleration or acceleration command sequence drives the servo motor to adjust the running speed or optimize the tension balance, respectively, to obtain a speed adjustment sequence or a tension adjustment sequence.

[0018] Preferably, in step S105, the sensor array collects the side pressure sequence, laying length measurement value, and speed control command in real time. After multi-threaded fusion processing, an original data set is generated. Then, parallel processing is used to synchronously collect and denoise the data to obtain a denoised data sequence. Next, data standardization is performed and statistical features are calculated to form a standardized data sequence. Based on this, a three-dimensional relationship matrix of pressure-length-speed is constructed, and principal component analysis is used to extract dynamic correlation features to obtain a set of feature vectors. If the contribution rate of the dynamic feature variance of this set exceeds a threshold, a dynamic control parameter sequence is generated accordingly. Finally, the control parameters drive the servo motor to adjust the operating state, output an optimized speed command sequence, and synchronously update the speed components in the three-dimensional matrix.

[0019] Preferably, in step S106, the spectral feature set is converted to the time domain using inverse fast Fourier transform to generate a time-domain signal sequence. Subsequently, the periodicity of the sequence is calculated to form a periodic feature vector. When the amplitude of the vector exceeds a preset fluctuation threshold, the system immediately performs spatial clustering on the coordinates of the failure points in the three-dimensional relation matrix to obtain a cluster center set. Then, based on this cluster center set, the K-means algorithm is used to divide the failure region and generate a region division result. Further, boundary features are extracted from the division result and the geometric center of each region is calculated to obtain a center coordinate set. Then, the spatial structure components of the three-dimensional relation matrix are updated using the center coordinate set to form an updated matrix. Finally, anomaly pattern matching is performed on the updated matrix. If a preset anomaly pattern is matched, an adjustment signal sequence is generated.

[0020] The side pressure encoder early warning method for cable laying described in this application has the advantage of addressing the problems of equipment failure and low efficiency caused by abnormal side pressure, length measurement error and tension imbalance during cable laying. By deploying a multi-axis pressure sensor array at the corner of the cable tray, the three-dimensional dynamic side pressure signal is collected in real time. A high-precision side pressure value sequence is generated through analog-to-digital conversion and dynamic calibration algorithm. At the same time, an incremental encoder combined with a temperature compensation calibration algorithm is used to output a laying length measurement value accurate to ±0.1%FS.

[0021] This invention monitors the side pressure sequence through a sliding window, generates closed-loop deceleration or fuzzy acceleration control commands based on a time-varying pressure threshold model, optimizes servo motor drive and tension balance, constructs a three-dimensional relationship matrix of pressure-length-velocity through multi-threaded fusion processing, performs spectral feature analysis, detects abnormal frequency components and triggers spatial positioning hierarchical early warning, and generates a diagnostic report containing the coordinates of the failure point.

[0022] This invention enables dynamic monitoring and intelligent control of the entire cable laying process, significantly improving laying accuracy, equipment stability, and operating efficiency. Attached Figure Description

[0023] Figure 1 This application describes a method for early warning of cable laying using a side pressure encoder. Figure 1 ;

[0024] Figure 2 This application describes a method for early warning of cable laying using a side pressure encoder. Figure 2 . Detailed Implementation

[0025] like Figures 1-2 As shown, the side pressure encoder early warning method for cable laying described in this application includes the following steps:

[0026] like Figures 1-2 As shown in Figure S101, a multi-axis pressure sensor array is deployed at the key points of the cable tray corners along the cable laying path to collect the three-dimensional dynamic lateral pressure simulation signal applied when the cable passes through in real time.

[0027] Further, in step S101, the three-dimensional dynamic lateral pressure signal at the corner of the cable tray is acquired by a multi-axis pressure sensor array to generate the original pressure dataset;

[0028] If the signal strength of the original pressure dataset exceeds the preset threshold, the fast Fourier transform algorithm is used to decompose the signal in the frequency domain to obtain the frequency domain feature set.

[0029] Based on the frequency domain feature set, the principal component analysis algorithm is used to extract the main pressure distribution patterns and generate a dimensionality-reduced feature vector set.

[0030] If the proportion of principal component variance in the feature vector set is greater than a preset threshold, the feature vector set is smoothed by the Kalman filter algorithm to obtain the optimized pressure distribution model.

[0031] Based on the optimized pressure distribution model, the dynamic mechanical parameters when the cable passes through are calculated, and a mechanical behavior dataset is generated.

[0032] Using a mechanical behavior dataset, time series analysis is employed to predict the pressure change trend when a cable passes through, resulting in a pressure prediction sequence.

[0033] Based on the pressure prediction sequence, identify the abnormal pressure distribution points at the corners of the cable tray and generate an anomaly detection report.

[0034] Specifically, in step S101, when deploying a multi-axis pressure sensor array at the key point of the cable tray corner in the cable laying path, the cable tray corner is first geometrically modeled using high-precision 3D modeling software to accurately calculate the radius of curvature at the corner, for example, set to 0.5 meters, and generate grid data containing the corner coordinates (x, y, z) with a resolution of 0.01 meters.

[0035] Subsequently, a multi-axis pressure sensor array was deployed at the key corner point. A triaxial pressure sensor (model MPU-9250) was selected, with each sensor covering an area of ​​10 cm × 10 cm. The array was arranged in a 5 × 5 matrix, with a total of 25 sensors. The sampling frequency was 100 Hz, and the three-dimensional dynamic side pressure simulation signal was collected in real time when the cable passed through.

[0036] Sensor data is transmitted to edge computing nodes via the I2C protocol, with a data packet size of 128 bytes per transmission.

[0037] The edge node uses an embedded processor (Raspberry Pi 4) to run a real-time signal processing algorithm. First, the acquired raw pressure signal is denoised using a Kalman filter algorithm. The process noise covariance Q is set to 0.01 and the measurement noise covariance R is set to 0.05. After filtering out high-frequency noise, smooth three-dimensional pressure components (Fx, Fy, Fz) are obtained.

[0038] Next, the total pressure is calculated using a vector synthesis algorithm. Fx, Fy, and Fz represent the triaxial pressure components (unit: N), and the peak pressure is extracted. For example, the maximum value is 500 N, which occurs at the point of maximum curvature at the corner.

[0039] Time series analysis of pressure data was performed, and the signal was converted into the frequency domain using Fast Fourier Transform (FFT) to identify the main vibration frequency (10Hz) and determine whether cable slippage caused abnormal vibration.

[0040] If the frequency exceeds the preset threshold (15Hz), an alarm signal is triggered and sent to the cloud management system via the MQTT protocol;

[0041] The cloud-based system uses a machine learning model (Support Vector Machine, SVM) to train historical pressure data. Input features include pressure peak, frequency, and angular curvature. The output is a cable wear risk level (low, medium, high), with an accuracy of 95%.

[0042] Finally, the system generates a real-time pressure distribution heat map with a resolution of 0.02 meters, which is stored in a cloud database for subsequent optimization of cable laying paths;

[0043] The entire process is automated, with data processing latency of less than 50 milliseconds, ensuring real-time performance and reliability.

[0044] In one embodiment, the Kalman filter state equation can be expressed as:

[0045] Equations of state: ;

[0046] Measurement equation: ;

[0047] The physical meaning of the characters in the above calculation formula:

[0048] xk: State (displacement) at time k, v: Cable speed (unit: m / s), Δt: Sampling interval (unit: s, Δt=0.01 s in the example), n: Measurement noise.

[0049] like Figures 1-2 As shown in step S102, the analog signal is input into the analog-to-digital conversion module, and a dynamic calibration algorithm is used for nonlinear correction to generate a digital real-time side pressure value sequence with a range of 0-500N that is continuously adjustable.

[0050] Further, in step S102, the analog signal is sampled by the analog-to-digital conversion module, and a fixed sampling frequency is used to obtain the digital side pressure sequence;

[0051] Based on the digitized side pressure sequence, if there are abnormal fluctuations in the sequence, the sequence is smoothed by the median filtering algorithm to obtain a smoothed side pressure sequence.

[0052] A smoothed lateral pressure sequence was used, and periodic components were extracted using time series decomposition methods to obtain a periodic feature sequence.

[0053] Based on the periodic characteristic sequence, if the fluctuation amplitude of the periodic characteristic sequence exceeds the preset threshold, the sequence is predicted by an autoregressive model to obtain the pressure change prediction sequence.

[0054] By using the pressure change prediction sequence, the local pressure change rate is calculated using the sliding window method to obtain the pressure change rate sequence;

[0055] Based on the pressure change rate sequence, if there are abrupt changes in the sequence, the location of the abrupt changes is identified by an anomaly detection algorithm, and an anomaly distribution sequence is obtained.

[0056] By using the distribution sequence of outliers, cluster analysis is employed to classify the outliers, resulting in a set of outlier classifications.

[0057] Specifically, in step S102, the analog signal collected at the corner of the cable tray along the cable laying path is processed by an analog-to-digital converter module. A high-precision analog-to-digital converter (ADS1256, 24-bit resolution) is selected to convert the analog voltage signal (range 0-5V) of the triaxial pressure sensor into a digital signal. The sampling rate is 200Hz and the range is set to 0-500N.

[0058] To ensure data accuracy, a dynamic calibration algorithm is used to perform nonlinear correction on the analog-to-digital conversion results. The algorithm is based on a polynomial fitting model, assuming the correction function is y=ax³+bx²+cx+d, where the coefficients are a=0.0001, b=0.02, c=1.5, and d=0.1, y: the corrected pressure value (unit: N), and x: the original analog-to-digital conversion value;

[0059] The sensor output is fitted using the least squares method to correct nonlinear errors, with the error range controlled within ±0.5N.

[0060] The corrected digital signal is transmitted to the edge computing unit (NVIDIA Jetson Nano) via the SPI protocol, with a data packet size of 256 bytes per transmission.

[0061] The edge computing unit runs a real-time processing program to smooth the corrected pressure data sequence. It uses a sliding window averaging algorithm with a window size of 10 samples to reduce random noise interference and obtain a continuous pressure value sequence (e.g., a peak pressure of 480N occurs when the cable sliding speed is 0.2m / s at a corner).

[0062] Subsequently, the system extracts features from the pressure sequence and calculates the pressure change rate using the formula ΔP / Δt, where Δt = 0.01 seconds. The maximum change rate (50 N / s) is extracted as the dynamic load feature.

[0063] Feature data is uploaded to the cloud analysis platform via the LoRa protocol. The cloud runs time series analysis algorithms and uses an autoregressive model (ARIMA, p=2, d=1, q=1) to predict the short-term trend of the stress series. The prediction window is 1 second and the prediction error is less than 5N.

[0064] The analysis results are stored in a distributed database (MongoDB) and indexed by timestamps and location coordinates (0.02 meters resolution) to facilitate dynamic adjustment of the subsequent cable laying path;

[0065] The entire processing flow is implemented through automated scripts, and the data processing latency is controlled within 30 milliseconds to ensure real-time performance.

[0066] like Figures 1-2 As shown in Figure S103, the cable displacement pulse signal is acquired in real time through an incremental encoder, and the pulse counting is eliminated by a temperature-compensated calibration algorithm to compensate for jitter error and cumulative error, and the laying length measurement value is output with an accuracy of ±0.1%FS.

[0067] Furthermore, in step S103, the cable displacement pulse signal is acquired in real time by an incremental encoder, and a pulse counting sequence is obtained by using a fixed sampling frequency;

[0068] If there are abnormal jumps in the pulse counting sequence, the sequence is smoothed by the median filtering algorithm to obtain a smooth pulse sequence;

[0069] Based on the smoothed pulse sequence, a temperature compensation algorithm is used in conjunction with ambient temperature data to correct the counting deviation caused by temperature, resulting in a corrected pulse sequence.

[0070] By correcting the pulse sequence, the local displacement change rate is calculated using the sliding window method to obtain the displacement change rate sequence.

[0071] If there are abrupt changes in the displacement change rate sequence, the location of the abrupt changes is identified by an anomaly detection algorithm to obtain the abrupt change point distribution sequence.

[0072] Based on the distribution sequence of mutation points, cluster analysis is used to classify the mutation points and obtain an anomaly classification set;

[0073] By using an anomaly classification set and a preset threshold to determine the anomaly type, a sequence of laying length measurement values ​​is generated.

[0074] Specifically, in step S103, during the cable laying process, a high-resolution incremental encoder (HEDL-5540, 1000 pulses / revolution) is used to collect the cable displacement pulse signal in real time. The encoder is installed on the traction wheel axle, the wheel circumference is 0.314 meters, and 1000 pulses are generated per revolution. When the cable moving speed is 0.5 meters / second, the pulse frequency is 1592Hz.

[0075] The pulse signal is acquired by a high-speed counting module (based on an STM32F4 microcontroller). The counter records the number of pulses at a sampling rate of 100kHz to generate the original displacement data.

[0076] To eliminate jitter error, a Kalman filter algorithm is used to smooth the pulse count. The filter state equation is x(k)=x(k-1)+v*Δt, and the measurement equation is z(k)=x(k)+n, where v is the cable speed, Δt=0.01 seconds, noise covariance Q=0.01, R=0.1. After filtering, the pulse count jitter error is reduced to ±2 pulses.

[0077] To compensate for the cumulative error caused by temperature, a temperature-compensated calibration algorithm is introduced. The temperature at the traction wheel (25℃ to 40℃) is measured based on an ambient temperature sensor (DS18B20, resolution 0.0625℃). The linear correction model L=Lo*(1+α*ΔT) is used, where L is the corrected length, Lo is the original count length, α=0.000011 / ℃ is the coefficient of thermal expansion, and ΔT is the temperature change. The corrected length error is controlled within ±0.1%FS (full scale 100 meters).

[0078] The corrected displacement data is transmitted to the edge computing unit (Raspberry Pi 4) via the MQTT protocol. The data packet size is 128 bytes and the transmission frequency is 100Hz.

[0079] The algorithm for accumulating edge cell running length is as follows: Wherein, S: cumulative laying length (unit: m), n: pulse count, C: circumference of traction wheel (unit: m, C=0.314 m in the example), N: number of encoder pulses per revolution (N=1000 in the example).

[0080] The formula is S=∑(n*0.314 / 1000), which calculates the real-time laying length (50.125 meters).

[0081] To ensure data reliability, the edge unit performs anomaly detection on the length sequence using the Z-score method with a threshold of 3σ, and removes abnormal pulses (count jumps caused by sudden noise).

[0082] The processed length data is uploaded to a cloud database (InfluxDB), indexed by timestamp (1 millisecond resolution) and laying location (0.05 meter resolution), for use by the path planning system.

[0083] The entire process is automated using Python scripts, with data processing latency controlled to within 20 milliseconds.

[0084] like Figures 1-2 As shown in S104, based on a preset time-varying pressure threshold model, a sliding window is used to monitor the real-time side pressure value sequence:

[0085] When the pressure value continuously exceeds the upper limit threshold, a closed-loop deceleration control command is generated to drive the servo motor.

[0086] When the pressure value remains below the lower threshold, a fuzzy acceleration control command is generated to optimize the tension balance.

[0087] Further, in step S104, the side pressure value sequence is collected in real time by the sensor, and a pressure sequence is obtained by using a fixed sampling frequency;

[0088] The pressure sequence is segmented using the sliding window method, and the average pressure within the window is calculated to obtain the average pressure sequence.

[0089] Based on the average pressure sequence and combined with the preset time-varying pressure threshold model, the average value is compared with the upper and lower thresholds to determine the triggering conditions.

[0090] If the mean value continuously exceeds the upper limit threshold, the proportional-integral-derivative control algorithm is used to generate closed-loop deceleration control commands, resulting in a deceleration command sequence.

[0091] By using a deceleration command sequence, the servo motor is driven to adjust its operating speed, resulting in an adjusted speed sequence.

[0092] If the mean value remains below the lower threshold, a fuzzy control algorithm is used to generate acceleration control commands, resulting in an acceleration command sequence.

[0093] Based on the acceleration command sequence, the servo motor is driven to optimize the tension balance, resulting in a tension adjustment sequence.

[0094] Specifically, in step S104, during the cable laying process, in order to monitor the side pressure of the traction system in real time and dynamically adjust the traction speed, a high-precision pressure sensor (MPX5700AP, range 0-700kPa, resolution 0.1kPa) is used to collect the side pressure value. The sensor is installed on the side wall of the traction wheel, the sampling frequency is 200Hz, and a real-time pressure sequence is generated.

[0095] For example, the cable laying speed is 0.8 meters per second, and the lateral pressure fluctuation during normal operation is 200-300 kPa;

[0096] The pressure data is converted into a digital signal at a sampling rate of 100kHz through an analog-to-digital conversion module (based on ADS1256, 24-bit resolution) and stored as a time series.

[0097] To achieve sliding window monitoring, a sliding window of length 50 (time span 0.25 seconds) was used to calculate the mean of the pressure sequence, using the formula P. avg =∑P i / 50, where P i The pressure value within the window;

[0098] If the average value exceeds the preset upper limit threshold of 350 kPa for 5 consecutive times (based on historical data analysis), the system generates a closed-loop deceleration command. The servo motor (model SGM7J, rated speed 3000 rpm) is adjusted through a PID controller (proportional coefficient Kp=0.5, integral coefficient Ki=0.1, derivative coefficient Kd=0.05) to reduce the target speed to 0.6 m / s. The calculation formula is: u(t)=Kp*e(t)+Ki*∫e(t)dt+Kd*de(t) / dt, where, u(t): control output (speed command), e(t) is the pressure deviation, Kp, Ki, Kd: PID coefficients (in this example, Kp=0.5, Ki=0.1, Kd=0.05);

[0099] Conversely, if the mean value is below the lower limit threshold of 150 kPa for 5 consecutive times, the system triggers fuzzy acceleration control based on fuzzy logic rules (input is pressure deviation and deviation change rate, output is speed increment). For example, when the deviation is -50 kPa and the change rate is -10 kPa / s, the speed increment is +0.1 m / s to optimize tension balance.

[0100] The fuzzy rule table is designed based on experience and contains 25 rules. It uses triangular membership functions and defuzzifies the rules by weighted average method after inference.

[0101] The processed instructions are transmitted to the motor driver via the CAN bus (speed 1Mbps), with a data packet size of 64 bytes and a delay controlled within 10 milliseconds;

[0102] The stress sequence was simultaneously uploaded to the edge computing unit (Jetson Nano), and anomaly detection was performed using a Python script. An anomaly removal algorithm based on mean and standard deviation was used, with a threshold of 2.5σ, to remove mutation data (instantaneous jump to 500kPa).

[0103] The final data is stored in a local SQLite database with timestamps (1 millisecond resolution) for subsequent analysis.

[0104] like Figures 1-2 As shown in S105, the multi-threaded fusion processing module constructs a dynamic pressure-length-speed three-dimensional relationship matrix by combining the side pressure value sequence, laying length measurement value, and speed control command.

[0105] Furthermore, in step S105, the side pressure value sequence, laying length measurement value and speed control command are collected in real time by the sensor array, and the original data set is obtained by using a multi-threaded fusion processing module;

[0106] Parallel processing methods are used to synchronously acquire and denoise the original dataset, generating a denoised data sequence.

[0107] Based on the denoised data sequence, perform data standardization, calculate the statistical characteristics of each sequence, and obtain the standardized data sequence;

[0108] By standardizing the data sequence, a three-dimensional relationship matrix of pressure-length-velocity is constructed, and the dynamic correlation features are extracted using the principal component analysis algorithm to obtain a set of feature vectors.

[0109] If the variance contribution rate of the dynamically associated features in the feature vector set exceeds a preset threshold, then dynamic control parameters are generated based on the feature vector set to obtain a control parameter sequence.

[0110] Based on the control parameter sequence, the operating state of the servo motor is adjusted to generate an optimized speed command sequence;

[0111] By optimizing the velocity command sequence, the velocity components in the three-dimensional relation matrix are updated, resulting in the updated three-dimensional relation matrix.

[0112] Specifically, in step S105, during the cable laying process, in order to construct a dynamic pressure-length-speed three-dimensional relationship matrix, a multi-threaded fusion processing module is used to process the side pressure value sequence, laying length measurement value and speed control command.

[0113] First, the lateral pressure data is collected by a high-precision pressure sensor (range 0-600kPa, resolution 0.05kPa) at a sampling frequency of 250Hz to generate a time series. During normal operation, the pressure fluctuation is between 180-280kPa.

[0114] The laying length is measured in real time by an encoder (resolution 0.01 meters), with a sampling frequency of 100 Hz, and the cable advance distance is recorded. For example, when the current laying speed is 0.9 meters / second, the length sequence is updated every 0.01 seconds.

[0115] Speed ​​control commands are generated by the servo motor driver and are based on voltage signals (range 0-10V).

[0116] The multi-threaded processing module uses three threads to process in parallel: Thread 1 performs a Fast Fourier Transform (FFT) on the stress sequence to extract frequency features, as shown in the formula. N: Number of sampling points (N=64 in the example), x(n): Time-domain pressure sequence, X(k): Frequency-domain component, calculate the dominant frequency component (e.g., 10Hz) to analyze the periodicity of pressure fluctuations;

[0117] Thread 2 applies a Kalman filter to the length sequence, with the state equation being x. k =x (k-1) +v k *Δt, noise covariance Q=0.01, smooths length data, eliminates measurement jitter, for example, reduces noise interference from 0.03 meters to 0.01 meters;

[0118] Thread 3 processes instructions using a weighted moving average algorithm, with the formula V. avg =0.6*V k +0.3*V (k-1) +0.1*V (k-2) Smooth speed fluctuations, for example, adjust the speed from 0.92 m / s to 0.90 m / s;

[0119] The fusion module maps the three data points to a three-dimensional matrix with dimensions [100, 100, 50], corresponding to pressure (0-600 kPa), length (0-1000 m), and speed (0-2 m / s), respectively.

[0120] The matrix elements are calculated using a three-dimensional interpolation algorithm (linear interpolation). For example, the interpolation result for the point (250 kPa, 500 m, 0.9 m / s) is a control weight of 0.85.

[0121] Anomaly detection was performed using K-means clustering with K=3, and data that deviated from the cluster center by two standard deviations (pressure sudden change to 450 kPa) were removed.

[0122] The processing results are transmitted to the cloud via the MQTT protocol (2Mbps rate), with a data packet size of 128 bytes and a latency controlled within 5 milliseconds. The data is then stored in a MongoDB database with a timestamp resolution of 0.5 milliseconds.

[0123] like Figures 1-2 As shown in step S106, spectral feature analysis is performed on the three-dimensional relationship matrix. When the energy of an abnormal frequency component exceeds a preset risk threshold, a spatial positioning graded early warning signal is triggered and a spectral diagnostic report containing the spatial coordinates of the failure point is generated.

[0124] Furthermore, in step S106, the spectral feature set is transformed in the time domain by using the Fast Fourier Transform algorithm to obtain the time domain signal sequence;

[0125] The periodic characteristics of the signal are calculated from the time-domain signal sequence, and a periodic feature vector is generated.

[0126] If the amplitude of the periodic eigenvector exceeds the preset fluctuation threshold, spatial clustering is performed on the coordinates of the failure points of the three-dimensional relation matrix to obtain the cluster center set.

[0127] Based on the cluster center set, the K-means algorithm is used to divide the failure point region and generate the region division result;

[0128] Extract boundary features from the region division results, calculate the geometric center of each region, and obtain the center coordinate set;

[0129] By using the central coordinate set, the spatial dimension components of the three-dimensional relation matrix are updated to generate an update matrix;

[0130] Perform anomaly pattern matching on the update matrix. If a preset anomaly pattern is matched, generate an adjustment signal sequence.

[0131] Specifically, in step S106, during cable laying, to perform spectral feature analysis of the three-dimensional relationship matrix and generate a spectral diagnostic report containing the spatial coordinates of the failure point, the three-dimensional matrix (dimensions [100, 100, 50], corresponding to pressure 0-600 kPa, length 0-1000 meters, and speed 0-2 m / s) is first subjected to spectral analysis. Discrete wavelet transform (DWT) is used to decompose the signal in the pressure dimension. Daubechies wavelet (db4) is selected, decomposed into 4 levels, extract frequency components, and calculate the energy of each level. The formula is as follows: , where c j (k) represents the wavelet coefficients of the j-th layer, E j The energy of the j-th wavelet layer, for example, the wavelet coefficients of the 4th layer are calculated for a pressure sequence (sampling frequency 200Hz, length 1024 points). The main frequency energy is concentrated in 2-8Hz, and the normal energy range is 500-800 units.

[0132] If the energy in a certain frequency band (12Hz) reaches 1200 units, exceeding the preset risk threshold of 1000 units, an anomaly detection will be triggered.

[0133] Anomaly detection uses the Support Vector Machine (SVM) algorithm with a radial basis function (RBF) kernel function, parameters C=1.0, γ=0.01. The pressure, length, and velocity data in the three-dimensional matrix are projected into a high-dimensional space to classify anomaly points. For example, if the coordinates (300kPa, 600m, 1.2m / s) deviate from the normal cluster, it is marked as a failure point.

[0134] Spatial positioning is achieved through a 3D grid search algorithm with a grid resolution of [0.5 kPa, 0.1 m, 0.01 m / s]. The coordinates of the location failure point are determined, for example (300.2 kPa, 600.1 m, 1.21 m / s), and the Euclidean distance between it and the normal data is calculated using the following formula:

[0135] Where d is the distance between the failure point and the normal cluster, p,l,v are the pressure (kPa), length (m), and velocity (m / s) of the failure point, and p0,l0,v0 are the coordinates of the center of the normal cluster. The distance value is 15.3, which exceeds the threshold of 10, triggering a graded early warning signal.

[0136] The warning signal is divided into three levels: 10-15 is low level, 15-20 is medium level, and >20 is high level. This example is a medium level warning.

[0137] The spectrum diagnostic report is generated in JSON format, including the coordinates of the failure point, the spectrum energy (12Hz, 1200 units), and the warning level (medium). It is transmitted to the cloud analysis platform via the Kafka protocol (throughput 5Mbps, data packet 256 bytes) and stored in the Elasticsearch database. The index timestamp accuracy is 1 millisecond to ensure real-time performance and traceability.

[0138] 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 application.

Claims

1. A method of early warning of a side pressure encoder of a cable laying, characterized in that, The method comprises the following steps: S101, deploying a multi-axis pressure sensing array at the bridge corner key point of the cable laying path, and collecting three-dimensional dynamic side pressure simulation signals applied by the cable in real time, and specifically comprising: Generating an original pressure data set through the multi-axis pressure sensing array; If the signal strength of the original pressure data set exceeds a preset threshold, frequency domain decomposition is performed to obtain a frequency domain feature set; Based on the frequency domain feature set, a pressure distribution pattern is extracted to generate a dimension reduction feature vector set; If the principal component variance proportion of the feature vector set is greater than a preset threshold, smoothing processing is performed to obtain an optimized pressure distribution model; Based on the optimized pressure distribution model, dynamic mechanical parameters are calculated to generate a mechanical behavior data set; Using a time series analysis method to predict the pressure change trend to obtain a pressure prediction sequence; According to the pressure prediction sequence, pressure distribution abnormal points are determined and an abnormality detection report is generated; S102, inputting the simulation signal into an analog-to-digital conversion module, and using a dynamic calibration algorithm to perform non-linear correction to generate a digital real-time side pressure value sequence with a range of 0-500N continuously adjustable; S103, acquiring cable displacement pulse signals in real time through an incremental encoder, and combining a temperature compensation type calibration algorithm to eliminate jitter error and compensate cumulative error for pulse counting, and outputting a laying length measurement value with an accuracy of ±0.1%FS; S104, based on a preset time-varying pressure threshold model, performing sliding window monitoring on the real-time side pressure value sequence: when the pressure value continuously exceeds the upper threshold, a closed-loop deceleration control instruction is generated to drive the servo motor; when the pressure value continuously falls below the lower threshold, a fuzzy acceleration control instruction is generated to optimize tension balance; S105, through a multi-thread fusion processing module, the side pressure value sequence, the laying length measurement value and the speed control instruction are combined to construct a dynamic pressure-length-speed three-dimensional relationship matrix; S106, performing spectral feature analysis on the three-dimensional relationship matrix, and when an abnormal frequency component energy exceeding a preset risk threshold is detected, triggering a spatial positioning hierarchical early warning signal and generating a spectral diagnosis report containing the spatial coordinates of the failure point.

2. The method of claim 1, wherein the method further comprises: The step S102 comprises: Sampling the analog signal through the analog-to-digital conversion module to obtain a digital side pressure sequence; If there is abnormal fluctuation in the sequence, smoothing processing is performed to obtain a smoothed side pressure sequence; Based on the smoothed side pressure sequence, periodic components are extracted to obtain a periodic feature sequence; If the fluctuation amplitude of the periodic feature sequence exceeds a preset threshold, a pressure change prediction sequence is predicted through an autoregressive model; Using a sliding window method to calculate the local pressure change rate to obtain a pressure change rate sequence; If there is a mutation point in the change rate sequence, an abnormal point distribution sequence is obtained by identifying the mutation point position through an abnormality detection algorithm; Classifying the abnormal points to obtain an abnormal point classification set.

3. The method of claim 1, wherein the method further comprises: The step S103 comprises: Acquiring cable displacement pulse signals through an incremental encoder to obtain a pulse count sequence; If there is an abnormal jump in the sequence, smoothing processing is performed to obtain a smoothed pulse sequence; Based on the smoothed pulse sequence and combined with environmental temperature data, a temperature compensation algorithm is used to correct the count deviation to obtain a corrected pulse sequence; Calculating the local displacement change rate of the corrected pulse sequence to obtain a displacement change rate sequence; If there is a mutation point in the displacement change rate sequence, a mutation point distribution sequence is obtained by identifying the mutation point position through an anomaly detection algorithm; The mutation points are classified to obtain an anomaly classification set; According to a preset threshold, the type of the anomaly classification set is judged to obtain a laying length measurement value sequence.

4. The method of claim 1, wherein, The step S105 comprises: A sensor array is used to collect a lateral pressure sequence, a laying length measurement value and a speed control instruction in real time, and a multi-thread fusion processing is performed to generate an original data set; A parallel processing method is used to perform synchronous collection and denoising processing on the original data set to obtain a denoised data sequence.

5. The lateral pressure encoder pre-alarm method for cable laying according to claim 1, characterized in that, The step S106 comprises: A time domain conversion is performed on the spectrum feature set to obtain a time domain signal sequence; A periodic feature vector is generated by calculating a signal periodicity feature; If the amplitude of the periodic feature vector exceeds a preset fluctuation threshold, spatial clustering is performed on the failure point coordinates of the three-dimensional relationship matrix to obtain a clustering center set.

Images