Dynamic tension digital measurement method for frame-type wire and cable stranding machine

Abstract

The present application relates to the field of measurement, specifically to a kind of dynamic tension digital measurement method of frame type electric wire cable stranding machine, comprising: original tension signal is collected to carry out low-frequency trend separation, obtain basic tension component and detrended alternating tension signal;Using multiple signal classification algorithm to the spectrum estimation of detrended alternating tension signal, determine the number of variational mode decomposition and initial center frequency;In decomposition iteration, the deviation and change rate of residual noise evaluation index dynamically adjust penalty factor to convergence;Convert main rotating speed into target frequency reference and select the most matched target mode component, carry out hilbert transform to calculate tension fluctuation amplitude, extract the mean of basic tension component, obtain final measurement result in combination with preset sensor calibration coefficient.The present application improves the decomposition stability of non-stationary tension signal, enhances the accuracy and anti-interference ability of tension measurement under complex working conditions.

Description

Technical Field

[0001] This invention relates to the field of measurement technology. More specifically, this invention relates to a dynamic tension digital measurement method for a frame-type wire and cable stranding machine. Background Technology

[0002] Frame-type stranding machines are widely used in the stranding process of wires and cables. The stability of the tension in their unwinding device directly affects the final forming quality, mechanical properties, and electrical properties of the cable. If the unwinding tension is too high, the wire is prone to excessive stretching and deformation, or even breakage. If the unwinding tension is too low, it will lead to a loose internal structure and insufficient stranding tightness in the cable. In actual high-speed continuous production environments, the operation of equipment is accompanied by and superimposed multi-source complex noise, including mechanical vibration, motor electromagnetic interference, transmission impact, guide wheel friction, and periodic structural disturbances. This industrial environment causes the raw signal output by the tension sensor to exhibit extremely strong noise interference, non-stationarity, and high multi-frequency coupling characteristics. This constitutes the core obstacle and root cause of the problem in accurately measuring dynamic tension.

[0003] Currently, by deploying sensing components at key stress points such as the wire guide rollers to collect electrical signals, frequency domain transformation or time-frequency decomposition of the original electrical signals containing interference is performed according to conventional signal analysis theory. Interference is then stripped away using a pre-defined static bandpass filter or a traditional mode decomposition algorithm with a fixed penalty factor and a preset number of decomposition layers. Based on these techniques, feature extraction of signals with relatively simple spectral morphology can be achieved, and conventional background noise with frequency bands significantly exceeding the mechanical operating frequency can be eliminated to a certain extent, thus initially separating the tension signal containing the basic trend.

[0004] However, the aforementioned technical solutions can, to some extent, address basic tension monitoring in low-speed and single-condition scenarios. But when faced with complex conditions involving high-speed rotation and multi-source frequency coupling, existing technologies rely on fixed decomposition parameter configurations or historical experience thresholds. The system cannot adaptively adjust based on the non-stationary noise state and frequency composition of the current equipment, making it prone to severe mode aliasing under strong interference. This causes the weak, effective tension fluctuations that truly characterize the periodic force on the equipment to be directly masked by broadband noise. Traditional processing architectures are often isolated from the actual mechanical operation of the equipment, lacking a physical frequency anchoring mechanism tied to the real-time rotational speed of the stranding machine spindle. This results in the algorithm's inability to accurately separate the periodic tension fluctuation components induced by the equipment's rotation from the macroscopic basic tension trend components within the complex frequency domain. This disconnect between theory and physical state causes both the final output basic tension value and dynamic tension fluctuation amplitude to deviate from the true physical force state, significantly reducing measurement accuracy. Summary of the Invention

[0005] To address the technical problems in existing technologies, such as modal aliasing due to reliance on fixed decomposition parameters and inaccurate separation of periodic tension fluctuations from the base tension due to lack of physical frequency anchoring, this invention provides a dynamic tension digital measurement method for a frame-type wire and cable stranding machine.

[0006] This invention provides a dynamic tension digital measurement method for a frame-type wire and cable stranding machine, comprising: S1, acquiring the noisy raw tension signal output by the tension sensor of the wire-laying device, separating the low-frequency trend of the noisy raw tension signal to obtain the basic tension component and the detrended AC tension signal, and using a multiple signal classification algorithm to estimate the spectrum of the detrended AC tension signal to determine the number of decompositions and the initial center frequency of the variational mode decomposition; S2, in the variational mode decomposition iteration, calculating the residual noise evaluation index of the residual signal after each update of the intrinsic mode function components, and based on the residual noise evaluation index and the target... The penalty factor is dynamically adjusted based on the deviation and rate of change of the noise evaluation index. The iteration terminates when the update amount of the intrinsic mode function component, the residual noise evaluation index, and the rate of change of the index all meet the preset convergence conditions. S3: The main speed of the stranding machine is obtained and converted into the target frequency reference by the rotation frequency multiple. The component with the smallest distance between the center frequency and the target frequency reference is selected as the target mode component. The Hilbert transform of the target mode component is used to extract the envelope to obtain the tension fluctuation amplitude. The mean value of the basic tension component is extracted as the basic tension value. The final basic tension measurement value and the final tension fluctuation amplitude are obtained by combining the preset sensor calibration coefficients.

[0007] This invention improves the decomposition stability of non-stationary tension signals by performing spectral estimation on detrended AC tension signals and dynamically adjusting the penalty factor of variational mode decomposition based on residual noise evaluation indicators. By using the target frequency reference mapped by the main rotation speed of the equipment for mode screening, the risk of mode aliasing is effectively reduced. It can accurately distinguish between the periodic tension fluctuation components caused by equipment rotation and the basic tension trend components, thereby improving the accuracy and anti-interference capability of dynamic tension measurement under complex multi-source noise conditions.

[0008] Preferably, the noisy raw tension signal output by the tension sensor of the wire-laying device is obtained by: sensing the change in mechanical tension force through a resistance strain gauge unit attached to the end-stage guide wheel of the wire-laying reel to generate a continuous analog voltage signal; using an anti-aliasing filter circuit composed of a low-pass filter to filter out high-frequency noise from the continuous analog voltage signal, wherein the cutoff frequency of the anti-aliasing filter circuit is higher than the upper limit of the target frequency reference; and using an analog-to-digital converter to discretize the filtered signal at a preset sampling rate to generate a one-dimensional time-series digital signal array as the noisy raw tension signal.

[0009] This invention constructs an anti-aliasing filter circuit by setting a low-pass filter with a cutoff frequency higher than the upper limit of the target frequency reference. This circuit can cut off useless signals in the high-frequency resonance band of the mechanical system and effectively retain the main low-frequency and mid-low-frequency characteristics of the tension, thereby reducing the spectral aliasing effect in the subsequent discretization digital sampling process.

[0010] Preferably, the step of separating the low-frequency trend of the noisy original tension signal to obtain the basic tension component and the detrended AC tension signal includes: calculating the target frequency reference based on the main rotation speed of the stranding machine and the inherent mechanical vibration frequency doubling coefficient; setting the low-frequency trend separation cutoff frequency to a preset ratio lower than the target frequency reference and not higher than the preset lower limit of the main AC mode frequency band; performing zero-phase low-pass filtering on the noisy original tension signal using the low-frequency trend separation cutoff frequency to obtain the basic tension component; and subtracting the basic tension component from the noisy original tension signal to obtain the detrended AC tension signal.

[0011] This invention sets the low-frequency trend separation cutoff frequency to a specific ratio lower than the target frequency reference and performs zero-phase low-pass filtering, which can completely preserve the slow change trend of the base tension, reduce the risk of the target mechanical periodic fluctuation component being incorrectly filtered out or incorporated into the base tension component, and improve the accuracy of tension component separation.

[0012] Preferably, the step of using a multiple signal classification algorithm to estimate the spectrum of the detrended AC tension signal and determine the number of decompositions and the initial center frequency of the variational mode decomposition includes: calculating the autocorrelation matrix of the detrended AC tension signal; performing eigenvalue decomposition on the autocorrelation matrix to divide the eigenvector space into a signal subspace and a noise subspace; constructing a frequency steering vector orthogonal to the noise subspace; calculating the spatial spectrum estimation function; scanning and drawing a spatial spectrum map within the target frequency band at a preset frequency step size; extracting extreme points in the spatial spectrum map whose amplitude is greater than a preset energy threshold; sequentially setting the frequency values ​​corresponding to the extreme points as the initial center frequency of each mode; and assigning the total number of extreme points to the variational mode decomposition algorithm as the number of decompositions.

[0013] This invention employs a multiple signal classification algorithm to extract extreme points in the spatial spectrum to determine the initial center frequency and the number of decompositions, providing more reasonable initial physical parameters for subsequent variational mode decomposition. This effectively reduces the risk of mode aliasing caused by traditional fixed empirical parameter settings and improves the energy concentration in the target tension fluctuation frequency band.

[0014] Preferably, in the variational mode decomposition iteration, the residual noise evaluation index of the residual signal is calculated after each update of the intrinsic mode function components, including: subtracting the reconstructed sum of each current intrinsic mode function component from the detrended AC tension signal to obtain the current residual signal, and performing mean and standard deviation normalization processing on the current residual signal to obtain the dimensionless residual signal; calculating the Gaussianity deviation of the amplitude distribution, the residual intensity of the time autocorrelation, and the flatness deviation of the power spectrum based on the dimensionless residual signal, and weighting and fusing the three to obtain the residual noise evaluation index.

[0015] Preferably, the step of dynamically adjusting the penalty factor based on the deviation and rate of change between the residual noise evaluation index and the target noise evaluation index includes: calculating the rate of change based on the residual noise evaluation index of the current iteration and the previous iteration; using a proportional-derivative controller, with the deviation between the residual noise evaluation index and the target noise evaluation index as the proportional term and the rate of change as the derivative term, calculating the penalty factor compensation amount in combination with a preset dimension conversion coefficient, updating the penalty factor for the next iteration cycle using the compensation amount, and applying preset upper and lower limit constraints during the update.

[0016] This invention introduces a closed-loop dynamic adjustment mechanism for the penalty factor based on residual noise evaluation index and proportional-derivative controller, and applies upper and lower limit clamping protection for parameters during updates to prevent parameter out-of-bounds errors that could lead to algorithm divergence, thereby improving the convergence stability and adaptability in the decomposition process of different non-stationary tension conditions.

[0017] Preferably, the step of calculating the Gaussianity deviation of the amplitude distribution, the residual intensity of the time-series autocorrelation, and the flatness deviation of the power spectrum based on the dimensionless residual signal includes: generating the actual probability density distribution curve of the dimensionless residual signal using a kernel density estimation algorithm; calculating the divergence between the actual probability density distribution curve and the standard normal probability density distribution curve to obtain the Gaussianity deviation of the amplitude distribution; calculating the average absolute value of the autocorrelation coefficients of each order within a preset lag order range of the dimensionless residual signal as the residual intensity of the time-series autocorrelation; calculating the flatness of the power spectrum of the dimensionless residual signal; and obtaining the flatness deviation of the power spectrum based on the flatness of the power spectrum.

[0018] Preferably, the step of obtaining the main rotation speed of the stranding machine and converting it into a target frequency reference by rotating frequency multiplication includes: acquiring the actual operating speed of the main shaft of the stranding machine under the current working conditions in real time by using a rotary encoder installed on the main shaft of the stranding machine; converting the actual operating speed of the main shaft from revolutions per minute to revolutions per second, and then multiplying it by the inherent mechanical vibration frequency multiplication coefficient related to the structure of the wire feeding device to obtain the target frequency reference.

[0019] Preferably, the step of selecting the component with the smallest distance between its center frequency and the target frequency reference as the target mode component includes: performing a Fast Fourier Transform on each intrinsic mode function component output by variational mode decomposition to obtain the discrete spectrum sequence of each intrinsic mode function component; searching for the main peak with the largest amplitude in the discrete spectrum sequence of each intrinsic mode function component, and taking the frequency value corresponding to the largest main peak as the actual center frequency of the corresponding intrinsic mode function component; calculating the absolute difference between the actual center frequency of each intrinsic mode function component and the target frequency reference; and taking the intrinsic mode function component with the smallest absolute difference as the target mode component representing tension fluctuation.

[0020] Preferably, the step of extracting the envelope of the target modal components using Hilbert transform to obtain the tension fluctuation amplitude, extracting the mean of the basic tension components as the basic tension value, and combining the preset sensor calibration coefficients to obtain the final basic tension measurement value and the final tension fluctuation amplitude includes: performing a Hilbert transform on the target modal components, constructing an analytical signal by combining orthogonal Hilbert transform signals, and calculating the complex modulus to obtain the instantaneous envelope signal; calculating the time arithmetic mean of the instantaneous envelope signal and the basic tension components respectively within the steady-state measurement time window as the tension fluctuation amplitude and the basic tension value; and substituting the basic tension value and the tension fluctuation amplitude into a dimension conversion function containing preset sensor calibration coefficients to convert the voltage dimension into the mechanical dimension of the final basic tension measurement value and the final tension fluctuation amplitude, respectively.

[0021] The technical solution of the present invention has the following beneficial technical effects:

[0022] This invention employs a multiple signal classification algorithm to estimate the spectrum of detrended AC tension signals to determine the initial center frequency and the number of decompositions, thus solving the problem of setting separation parameters and ensuring the reliability of extracting the basic tension components.

[0023] This invention utilizes the deviation and rate of change of the residual noise evaluation index to control the dynamic adjustment process of the penalty factor, thereby prompting the update of the intrinsic mode function components to meet the preset convergence conditions and terminating the variational mode decomposition iteration, thus improving the iterative stability when extracting intrinsic mode function components from noisy original tension signals.

[0024] This invention converts the main rotation speed of the stranding machine into a target frequency reference, selects the component with the best matching center frequency based on the principle of minimum frequency distance, and establishes the reference conditions for separating the target modal component and the basic tension component.

[0025] This invention extracts numerical values ​​from the input and target modal components of the basic tension measurement branch and converts them to obtain the final basic tension measurement value and the final tension fluctuation amplitude. This eliminates the interference of multi-frequency coupling noise and improves the extraction accuracy of the final basic tension measurement value and the final tension fluctuation amplitude. Attached Figure Description

[0026] Figure 1 This is a flowchart of a dynamic tension digital measurement method for a frame-type wire and cable stranding machine; Figure 2 This is a timing diagram of the original tension signal; Figure 3 This is a schematic diagram illustrating the spatial spectrum estimation results of the detrended AC tension signal. Detailed Implementation

[0027] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0028] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0029] This invention discloses a method for digitally measuring the dynamic tension of a frame-type wire and cable stranding machine, referring to... Figure 1 This includes the following steps: S1. Acquire tension signals and determine decomposition parameters. Specifically, acquire the noisy raw tension signal output by the tension sensor of the wire feeding device during the operation of the stranding machine, perform low-frequency trend separation on the noisy raw tension signal to obtain the basic tension component, and subtract the basic tension component from the noisy raw tension signal to obtain the detrended AC tension signal; use a multiple signal classification algorithm to perform spectral estimation on the detrended AC tension signal to determine the number of decompositions and the initial center frequency of the variational mode decomposition.

[0030] The analog voltage signal from the tension sensor is read in real time at a set sampling rate using a data acquisition card and converted into a one-dimensional digital signal sequence. Low-frequency trend separation is performed on the one-dimensional digital signal sequence to obtain the basic tension component. The detrended AC tension signal is obtained by subtracting the basic tension component from the one-dimensional digital signal sequence. Low-frequency trend separation is achieved using a zero-phase low-pass filter. Specifically, the target frequency reference is calculated based on the main speed of the stranding machine and the inherent mechanical vibration frequency harmonics. The low-frequency trend separation cutoff frequency is set to be a certain percentage lower than the target frequency reference. The relationship between the cutoff frequency and the target frequency reference is as follows: ; in, This is the cutoff frequency for low-frequency trend separation; The target frequency reference.

[0031] The cutoff frequency is taken as a certain proportion of the target frequency reference. The proportion is determined based on the fact that the bandwidth of the base tension change under steady-state conditions is much lower than the target mechanical periodic fluctuation frequency. This is to preserve the slow trend of the base tension and avoid the periodic fluctuation component from being mixed into the base tension component, and the cutoff frequency is not higher than the lower limit of the preset main AC mode frequency band. When the main speed is 1200 r / min and the inherent mechanical vibration frequency doubling factor is 4, the target frequency reference is 80 Hz, and the cutoff frequency can be set to 1 Hz. Subsequently, zero-phase low-pass filtering is performed on the noisy original tension signal to obtain the base tension component that changes slowly over time; then, the base tension component is subtracted from the noisy original tension signal to obtain the detrended AC tension signal. Zero-phase low-pass filtering can preserve the slow trend of the base tension change and reduce the risk of the target mechanical periodic fluctuation component being incorporated into the base tension component.

[0032] Construct the autocorrelation matrix of the detrended AC tension signal and perform eigenvalue decomposition on the autocorrelation matrix. Sort the obtained eigenvalues ​​in descending order. Based on the set noise subspace dimension, separate the eigenvectors corresponding to the signal subspace and the noise subspace. Use the noise subspace eigenvectors to construct the spatial spectrum estimation function for a multiple signal classification algorithm. Traverse the set frequency range and calculate the spatial spectrum value of the spatial spectrum estimation function at each frequency point. Perform peak search on all spatial spectrum values ​​to obtain a series of extreme points. Calculate the spectral energy corresponding to each extreme point, extract the extreme points in the spatial spectrum whose amplitude is greater than a preset energy threshold, and assign the total number of extreme points to the decomposition quantity K of the variational mode decomposition algorithm. Extract the frequency values ​​corresponding to the extreme points sequentially and use them as the initial center frequencies of the K intrinsic mode function components.

[0033] In an optional embodiment, a strain gauge unit consisting of four foil strain gauges with a resistance of 350Ω and a sensitivity coefficient of 2 is mounted on the bearing shaft surface of the final guide roller of the wire feeding spool using a full-bridge configuration. When the stranding machine operates at a set linear speed, such as 30m / min, the guide roller is subjected to a mechanical tension force with an amplitude ranging from 10N to 200N. The strain gauge bridge outputs a continuous analog voltage signal in the millivolt range, which is then conditioned to a continuous analog voltage signal in the 0V to 5V range by a differential amplifier circuit. To prevent spectral aliasing during subsequent digital sampling, an anti-aliasing filter circuit is constructed using a fourth-order Butterworth active low-pass filter with a cutoff frequency set to 500Hz.

[0034] The cutoff frequency is set based on the fact that the upper limit of the target frequency reference during normal operation of the stranding machine is usually no more than 150Hz. High-frequency resonance in mechanical transmission is mostly distributed above 1000Hz; therefore, 500Hz can cut off high-frequency interference while retaining the main low-frequency and mid-low-frequency characteristics of the tension. Using a signal acquisition card with an integrated 24-bit analog-to-digital converter, the filtered 0V to 5V analog signal is continuously and uninterruptedly discretized at a preset sampling rate of 2000Hz under software-driven control. The sampling clock is provided by a temperature-compensated crystal oscillator. The sampling rate is set with a margin of more than 5 times that of the Nyquist sampling theorem, based on the upper limit of the maximum target frequency reference corresponding to the highest operating speed of the stranding machine, balancing signal detail preservation and data volume control. Each acquisition trigger lasts for a 2-second analysis window, generating a one-dimensional time-series digital signal array containing 4000 discrete floating-point voltage amplitude data points. The data type is double-precision floating-point, with values ​​ranging from 0.1V to 4.9V. Using a 1×4000-dimensional digital signal array as the noisy raw tension signal for subsequent algorithm analysis and processing can preserve the dynamic details of tension and reduce the impact of sampling aliasing.

[0035] Reference Figure 2 The original tension signal exhibits continuous amplitude fluctuations and random spikes over time. The overall waveform oscillates around a slowly fluctuating baseline, indicating that the sensor output signal simultaneously contains basic tension trend components, periodic tension fluctuation components, and broadband noise, exhibiting strong noise and non-stationary characteristics. This demonstrates that directly averaging the original signal or using a single filter is insufficient to accurately separate the basic tension and tension fluctuations, thus necessitating a process of separating the trend before performing mode decomposition.

[0036] In an optional embodiment, an observation data matrix is ​​constructed based on a detrended AC tension signal sequence containing 4000 sampling points. The autocorrelation lag length is set to 1000, and then an autocorrelation matrix with a dimension of 1000×1000 is calculated using the sample autocovariance method. The lag length of the autocorrelation matrix is ​​determined based on the analysis window length and the signal correlation time. In this embodiment, the analysis window contains 4000 sampling points, and the lag length is selected as 1000, which is 1 / 4 of the window length. Singular value decomposition or eigenvalue decomposition is performed on the autocorrelation matrix to obtain 1000 non-negative eigenvalues ​​arranged in descending order and their corresponding eigenvectors. The dimension M of the signal features is determined using the minimum representation length criterion. Typically, M is between 3 and 7; here, it is assumed that M is 4. The eigenvectors corresponding to the first four largest eigenvalues ​​are used to span the signal subspace, and the remaining 996 are distributed in a 1×10⁻⁶ space. -5 Up to 1×10 -6 The eigenvector matrix corresponding to the noise basis eigenvalues ​​within the range constitutes the noise subspace.

[0037] A frequency steering vector of length 1000 is constructed using a complex exponential sequence. Based on the approximate orthogonality between the steering vector and the noise subspace, a spatial spectrum estimation function is constructed. During peak search, a target frequency band matching the actual mechanical operation is set to 0.5Hz to 300Hz, and a dense calculation scan is performed with a preset frequency step size of 0.5Hz to plot a one-dimensional spatial spectrum representing the energy distribution of each frequency component. The frequency step size is determined based on the minimum frequency interval between adjacent mechanical harmonics at the lowest operating speed of the equipment. In this embodiment, the minimum operating speed of the stranding machine spindle is 300r / min, and the interval between adjacent harmonics is 5Hz. The step size is selected to not exceed 1 / 10 of the minimum frequency interval; in this embodiment, the preset frequency step size is 0.5Hz to ensure sufficient frequency resolution. A preset energy threshold is set to 10% of the global maximum amplitude of the spatial spectrum. Assuming the global maximum peak amplitude is 1500, the threshold is automatically set to 150. The preset energy threshold is determined based on the statistical characteristics of the peak amplitude of the equipment's background noise. Specifically, the process involves: acquiring background noise signals while the equipment is off, performing multi-signal classification spectral estimation, and calculating the maximum amplitude of the noise spectrum peaks. Twice this maximum amplitude is used as the discrimination threshold. This threshold is calculated to be approximately 10% of the global maximum amplitude. The spectrum is traversed, and local peaks with amplitudes greater than 150 are identified as extrema. Assuming the algorithm locates four extrema, corresponding to frequencies of 1.5Hz, 25Hz, 48Hz, and 120Hz, the frequency parameters corresponding to these extrema are sequentially set as the initial center frequency vector parameters of each eigenmode function component in the variational mode decomposition algorithm. The total number of extrema, 4, is assigned to the decomposition quantity K of the variational mode decomposition algorithm.

[0038] Reference Figure 3 The spatial spectrum exhibits sharp spectral peaks at several discrete frequencies, with the amplitude of each peak significantly higher than the spectral base of other frequency bands. The frequencies corresponding to the spectral peaks fall near 1.5Hz, 25Hz, 48Hz, and 120Hz, respectively. This indicates that the multiple signal classification algorithm can distinguish independent frequency components in a strong noise background. Based on the number and frequency positions of spectral peaks above the energy threshold, it provides the number of decompositions and the initial center frequency for variational mode decomposition, thereby reducing the risk of mode aliasing.

[0039] S2. Iterative decomposition and dynamic adjustment of the penalty factor. Specifically, in the variational mode decomposition iteration, after each update of the intrinsic mode function components, the residual noise evaluation index of the current residual signal is calculated. The penalty factor is dynamically adjusted according to the deviation and rate of change between the residual noise evaluation index and the target noise evaluation index. When the update amount of the intrinsic mode function components, the residual noise evaluation index, and the rate of change all meet the preset convergence conditions, the variational mode decomposition iteration is terminated.

[0040] The penalty factor and Lagrange multipliers of the variational mode decomposition are initialized, and the intrinsic mode function components and center frequencies corresponding to the detrended AC tension signal are updated iteratively in the frequency domain according to the Wiener filtering principle. All intrinsic mode function components obtained in the current iteration are summed, and the summation result is subtracted from the detrended AC tension signal to obtain the current residual signal sequence. The current residual signal sequence is normalized by removing the mean and standard deviation to obtain the dimensionless residual signal. Based on the dimensionless residual signal, the Gaussianity deviation of the amplitude distribution, the residual intensity of the time-series autocorrelation, and the flatness deviation of the power spectrum are calculated, and the three are weighted and fused to obtain the residual noise evaluation index. The difference between the current residual noise evaluation index and the preset target noise evaluation index is calculated as the index deviation value, and the difference between the current residual noise evaluation index and the residual noise evaluation index of the previous iteration is calculated as the index change rate. A proportional-derivative controller is constructed, and the index deviation value and index change rate are input into the proportional and derivative terms, respectively. The compensation amount of the penalty factor is calculated, and the current penalty factor is added to the compensation amount to complete the update of the penalty factor and use it for the next iteration.

[0041] Calculate the L2 norm of the difference between each intrinsic mode function component in the current iteration and the previous iteration, and sum the L2 norms of all components to obtain the sum of the L2 norms of the update quantity. Divide the sum of the L2 norms of the update quantity by the sum of the L2 norms of each intrinsic mode function component in the previous iteration to obtain the normalized update quantity. The normalized update quantity is used to measure the degree of convergence by referring to the known convergence criteria of variational mode decomposition, with the relative L2 norm of the update quantities of the mode components in two adjacent iterations. If the normalized update quantity is less than a set first threshold, the residual noise evaluation index is less than a set second threshold, and the absolute value of the rate of change of the residual noise evaluation index is less than a set third threshold, then the loop is terminated and the iteration ends; otherwise, the frequency domain update step of the intrinsic mode function components is returned. The first threshold is preferably 1 × 10⁻⁶. -3 The second threshold is preferably 0.03, and the third threshold is preferably 1×10. -3The first threshold is determined based on the relative change of modal components between two adjacent iterations. Variational mode decomposition is performed on representative tension signals, and the normalized update amount after each iteration is recorded. A decay curve is plotted, and 1 / 5 of the update amount when the curve enters a flat section is selected as the first threshold. The second threshold is determined based on the amplitude distribution, time autocorrelation, and power spectrum flatness statistics of the residual signal under stable operating conditions. Twenty representative tension signal samples are collected under stable operation of the stranding machine and normal tension fluctuations. Variational mode decomposition is performed until iterative stability is achieved, and the residual noise evaluation index of the final residual signal of each sample is calculated. The upper bound of the 95% confidence interval of all index values ​​is taken as the second threshold. The third threshold is determined based on the convergence change amplitude of the residual noise evaluation index in continuous iterations. On samples under the same stable operating conditions, the absolute value of the rate of change of the residual noise evaluation index at the end of each iteration is calculated, and the upper bound of the 95% confidence interval of all samples is taken as the third threshold.

[0042] In an optional embodiment, at the end of each iteration of the variational mode decomposition, the reconstructed sum of each modal component is subtracted from the detrended AC tension signal to obtain the residual signal. The residual signal is then normalized by removing the mean and standard deviation to obtain a dimensionless residual signal. The standard Gaussian kernel function is called, and the smoothing bandwidth parameter h is set using Scott's rule of thumb, resulting in h=0.15. The kernel density estimation algorithm is used to generate the actual probability density distribution curve of the current residual signal containing 1000 discrete points within the dimensionless amplitude domain [-5,5] with a step size of 0.01. The KL divergence between the actual probability density distribution curve and the standard normal probability density distribution curve is calculated to obtain the Gaussianity deviation of the amplitude distribution. .

[0043] KL divergence is a well-known method for measuring the degree of difference between two probability distributions. The closer the residual signal is to Gaussian white noise, the smaller the KL divergence between the actual probability density distribution and the standard normal distribution. Therefore, KL divergence is used to characterize the Gaussianity deviation of the amplitude distribution. Further, the autocorrelation coefficients of the dimensionless residual signal within a preset lag order range are calculated, and the average absolute value of each autocorrelation coefficient is taken as the residual intensity of the time-series autocorrelation. .

[0044] The autocorrelation coefficients of each order approach 0. The autocorrelation coefficients increase when the residual signal contains unseparated periodic components. Therefore, the average absolute value of each order autocorrelation coefficient is used to characterize the residual strength of the time-series autocorrelation. The power spectral density of the dimensionless residual signal is calculated, and the power spectral flatness is obtained from the ratio of the geometric mean to the arithmetic mean of the power spectral density. Subtract the power spectral flatness from 1 As a deviation from power spectrum flatness .

[0045] Power spectral flatness is defined as the ratio of the geometric mean to the arithmetic mean of the power spectrum. It is a well-known measure of spectral whiteness. Power spectral flatness approaches 1. Power spectral flatness decreases when lumped frequency components are present. Therefore, subtracting the power spectral flatness from 1 characterizes the degree of residual periodic components in the residual signal. A residual noise evaluation index is constructed by weighted fusion of the three indices. The residual noise evaluation index is calculated as follows: ; in, As an evaluation index for residual noise; This indicates Gaussian bias in the amplitude distribution. The residual strength is the time-series autocorrelation value. This is due to power spectrum flatness deviation; , , The weights are non-negative and the sum of the three is 1.

[0046] Linear weighted fusion is a well-known method for comprehensive evaluation of multiple indicators. This method, based on this, jointly characterizes noise purity using residual features from three dimensions: amplitude distribution, temporal correlation, and power spectrum. As an example, we can take... =0.4、 =0.3、 =0.3. Assume the result obtained in the t-th iteration is... =0.085、 =0.03、 =0.12, then =0.079; The residual noise evaluation index for the (t-1)th iteration is 0.092, and the index change rate is -0.013 when the iteration step size is set to unit 1.

[0047] An incremental proportional-derivative controller is constructed to implement the penalty factor. The target noise evaluation index, reflecting the residual noise state, is set to 0.02. The current index deviation is 0.059. A set proportional gain is selected. =5000 and differential gain =1500, and with a preset dimension conversion coefficient β=1, the penalty factor compensation is calculated using the control law equation, which is as follows: ; in, This is the amount of compensation for the penalty factor; For proportional gain; The deviation between the residual noise evaluation index and the target noise evaluation index; This is the differential gain; The rate of change of the indicator; These are the preset dimension conversion coefficients.

[0048] With the stranding machine stopped and the pay-off reel unloaded, 10 sets of pure background noise signals were continuously collected. For each set of signals, the Gaussianity deviation of the amplitude distribution, the residual intensity of the time-series autocorrelation, and the flatness deviation of the power spectrum were calculated. =0.4、 =0.3、 =0.3 weighted fusion, and the arithmetic mean of 10 sets of measurement results is taken as the target noise evaluation index. On typical tension signal samples, the proportional-derivative controller is run with different proportional gain values. The settling time and overshoot of the residual noise evaluation index from the initial value to the target value are observed, and the gain value with fast settling speed and no significant overshoot is selected. After the proportional gain is calibrated, the derivative gain is gradually increased, and the oscillation suppression effect in the system step response is observed. A value that makes the system damping moderate is selected. The preset dimension conversion coefficient is determined according to the proportional relationship between the dimension of the penalty factor and the dimension of the controller output. It is usually set to 1 so that the compensation amount directly reflects the calculation result of the control law.

[0049] Proportional-derivative (PD) control is a well-known method for process parameter feedback adjustment. This method uses the residual noise evaluation index deviation as the proportional term and the index change rate as the derivative term to apply online compensation to the penalty factor. Substituting the deviation (0.059) and the index change rate (-0.013) into the control law equation, the compensation amount is calculated. According to the update formula, the current penalty factor (1200) is added to the compensation amount to obtain the initial update value (1475.5). To prevent parameter out-of-bounds errors that could lead to algorithm divergence, threshold clamping protection is implemented. The lower limit of the penalty factor is set at 100, and the upper limit at 5000. If the calculated value is lower than the lower limit, it is overridden to 100; if the calculated value is higher than the upper limit, it is overridden to 5000. Otherwise, 1475.5 is used as the new penalty factor parameter and input into the (t+1)th variational mode decomposition iteration. The dynamic adjustment mechanism of the penalty factor improves the parameter adaptability and decomposition stability of the algorithm under different tension conditions. The upper and lower limits of the penalty factor are determined based on sensitivity experiments using typical working condition samples. Specifically, variational mode decomposition is performed on representative tension signal samples with different penalty factors. The mode separation effect and iterative convergence speed are observed, and the minimum and maximum effective values ​​that ensure stable convergence and do not cause mode aliasing are selected as constraint boundaries.

[0050] S3. Filter target modes and output measurement results. Specifically, obtain the main rotational speed of the stranding machine and convert it to the target frequency reference according to the rotational frequency harmonics. Traverse all intrinsic mode function components, and select the component whose center frequency best matches the target frequency reference based on the principle of minimum frequency distance as the target mode component. Use the basic tension component as the input of the basic tension measurement branch, perform Hilbert transform on the target mode component to extract the envelope to calculate the tension fluctuation amplitude, extract the mean value of the basic tension component as the basic tension value, and combine it with the preset sensor calibration coefficients to obtain the final basic tension measurement value and the final tension fluctuation amplitude.

[0051] The pulse signal from the spindle encoder of the stranding machine is read by the programmable logic controller (PLC), and the industrial communication library is used to parse the pulse signal into a rotational speed per minute (RPM). The RPM value is divided by 60 and multiplied by the inherent mechanical vibration frequency doubling coefficient related to the structure of the wire feeding device to obtain the target frequency reference. An empty list is created to store the frequency difference. The absolute value of the difference between the center frequency of each intrinsic mode function (IMF) component output after the iteration terminates and the target frequency reference is calculated and appended to the empty list. The index corresponding to the smallest absolute value in the list is obtained, and the corresponding IMF component is extracted based on the index as the target mode component. At the same time, the basic tension component obtained by low-frequency trend separation is called and used as the input of the basic tension measurement branch. An analytical signal is constructed for the target mode component, the magnitude sequence of the analytical signal is obtained, the amplitude envelope of the target mode component is extracted, and the arithmetic mean of the amplitude envelope sequence is calculated as the tension fluctuation amplitude. The arithmetic mean of the basic tension component sequence is calculated as the basic tension value. The sensor sensitivity coefficient and zero offset obtained through linear fitting experiments are read from the system configuration file. The basic tension value is multiplied by the sensor sensitivity coefficient and then the zero offset is added to calculate the final basic tension measurement value. The tension fluctuation amplitude is multiplied by the sensor sensitivity coefficient to calculate the final tension fluctuation amplitude.

[0052] An incremental photoelectric rotary encoder with a resolution of 1024 pulses per revolution is coaxially installed at the tail end of the main shaft of the stranding machine. When the equipment is running, the matching programmable logic controller (PLC) or digital signal processor uses a high-speed hardware counter module to set a fixed 50ms speed measurement time window and continuously reads the pulse count value output by the encoder using the M / T speed measurement algorithm. After conversion using the gearbox reduction ratio and transmission chain coefficient, the actual operating speed N of the main shaft under the current working condition is calculated, and the real-time reading value is 1200 r / min. After obtaining the main shaft speed, the inherent mechanical vibration frequency multiplication factor C, which is strongly correlated with the mechanical topology of the current pay-off device, is extracted. In a frame-type stranding machine, high-frequency tension disturbances usually originate from the periodic rotational cutting and pulling of the mechanical forks, guide wheel groups, or the wire guide arms of the stranding cage on the pay-off reel. If four guide fork structures are evenly distributed on the equipment, each rotation of the main shaft will induce four tension impact pulsations; based on this, the frequency multiplication factor C is calibrated to 4. The frequency conversion subroutine is executed, and the target frequency reference is calculated as follows: ; in, As the target frequency reference; The actual operating speed of the main spindle; This is the inherent mechanical vibration harmonic factor. Dividing the rotational speed by 60 converts revolutions per minute to revolutions per second, which is the fundamental frequency of rotation. Multiplying this by the inherent mechanical vibration harmonic factor yields the characteristic frequency of the mechanical periodic disturbance, a well-known approach to frequency conversion in rotating machinery. Substituting the real-time rotational speed of 1200 r / min and the coefficient 4, the target frequency reference is found to be 80 Hz.

[0053] In an optional embodiment, for the K one-dimensional eigenmode function component sequences, each with a length of 4000 points, output after variational mode decomposition, assuming that they currently contain four components from IMF1 to IMF4, the data processor adds a Hanning window to each component array to suppress spectral energy leakage and calls a radix-2 fast Fourier transform algorithm based on a butterfly arithmetic architecture. With a system sampling rate of 2000Hz, the fast Fourier transform calculation generates four sets of discrete amplitude spectrum sequences with a single-sideband range of 0 to 1000Hz and a frequency resolution of 0.5Hz. Within the data space of each discrete spectrum sequence, an extremum point scanning pointer is run to find the location of the highest global spectral line amplitude, and the discrete frequency value corresponding to the amplitude peak is extracted as its corresponding actual center frequency. For example, the actual center frequencies of IMF1, IMF2, IMF3, and IMF4 are measured to be 1.2Hz, 24.5Hz, 79.5Hz, and 121Hz, respectively. Using the previously calculated 80Hz target frequency benchmark as a reference anchor point, the absolute difference between the actual center frequency of each intrinsic mode function component and the 80Hz benchmark point is calculated using the absolute value function. The absolute differences between each component and the target frequency benchmark are 78.8Hz, 55.5Hz, 0.5Hz, and 41Hz, respectively. The differences are quickly sorted and indexed to locate the element with the smallest difference, 0.5Hz, which corresponds to the IMF3 component. Therefore, the logical judgment algorithm specifies IMF3 as the target mode component representing the tension periodic fluctuation caused by mechanical rotation. The basic tension component is obtained from the low-frequency trend separation step.

[0054] In an optional embodiment, for the target modal component sequence x(t) with a data length of 4000 points obtained after filtering, such as representing the IMF3 component of 80Hz tension fluctuation, the digital signal processor performs a Hilbert transform based on discrete convolution on the sequence, causing all positive frequency components in the sequence to shift by -90° and negative frequency components to shift by +90°, thereby outputting a Hilbert transform signal array y(t) orthogonal to the original sequence. Using the complex plane construction rule, with x(t) as the real part and y(t) as the imaginary part, a complex analytic signal sequence is synthesized, which is represented as follows: ; in, It is a complex analytic signal; The target modal component sequence; For Hilbert transform signal arrays; It is the imaginary unit.

[0055] The analytic signal is constructed by using the original signal as the real part and the Hilbert transform signal as the imaginary part. This is a well-known method for envelope demodulation, and the complex modulus is the instantaneous envelope. The Euclidean norm complex modulus is calculated point-by-point for the complex analytic signal, generating an array of 4000 non-negative real numbers for the instantaneous envelope signal, used to characterize the time-domain envelope trajectory of the outer edge amplitude of the high-frequency tension oscillation. To filter out random spikes and obtain stable evaluation quantities, a 1-second steady-state measurement time window is extracted from the system operation. This corresponds to 2000 discrete sampling points that are stable in the middle of the time series. The instantaneous envelope amplitudes within the time window are summed and divided by 2000, yielding a time arithmetic mean of 0.45V. This voltage scalar is then solidified as the tension fluctuation amplitude for the current period. The basic tension assessment branch is executed synchronously. The basic tension component obtained through low-frequency trend separation is extracted. Within the aligned 1-second steady-state measurement time window, the arithmetic mean of 2000 sampling points is calculated to obtain a mean value of 2.5V representing the macroscopic tension bearing baseline, which is used as the basic tension value at the voltage level. The sensor linear model parameters, pre-stored in ROM and fitted using the static weight calibration method, are retrieved. These parameters include the sensor sensitivity coefficient of 50N / V and the zero-point drift compensation constant of -10N. The voltage parameters are substituted into the dimensional conversion function for mathematical calculation. The final basic tension measurement value and the final tension fluctuation amplitude are calculated as follows: ; ; in, This is the final measured value of the basic tension. This refers to the sensor sensitivity coefficient; The base tension value; This is the zero-point offset; This represents the final tension fluctuation amplitude. This represents the amplitude of tension fluctuation.

[0056] A linear model of the sensor was obtained by fitting the static weight calibration method. The base tension, being a DC component, requires the addition of a zero-point offset. Tension fluctuation, being an AC component without DC bias, is calculated without adding the zero-point offset. The base tension value of 2.5V is mapped to a final measured base tension value of 115N under mechanical dimensions. For the AC fluctuation, no zero-point offset is added, and the final tension fluctuation amplitude is calculated to be 22.5N. Thus, the voltage array is analytically restored to a practically measurable engineering mechanical index, meeting the needs of actual industrial field condition monitoring.

[0057] The experimental verification was conducted using a frame-type stranding machine with an operating speed set at 30 m / min, and the tension force borne by the wire feeding device ranged from 10 N to 200 N. The test system acquired the original tension signal at a sampling rate of 2000 Hz, and used the output of a high-sensitivity reference tension sensor as the baseline true value. The experiment was set up with four test groups: Group 1 was the complete scheme group using multi-signal classification spectrum estimation, adaptive adjustment of variational mode decomposition parameters, and main speed frequency doubling mode screening; Group 2 was the constant mode decomposition control group using fixed decomposition parameters; Group 3 was the ablation experimental group with a dynamic adjustment strategy to remove the penalty factor; and Group 4 was the ablation experimental group with the multi-signal classification spectrum estimation step removed. Performance evaluation indicators included signal-to-noise ratio improvement, mode aliasing rate, and relative error in tension fluctuation amplitude measurement.

[0058] Experimental results show that the signal-to-noise ratio (SNR) improvement of Group 1 (complete scheme) was 15.8 dB, the modal aliasing rate was 1.2%, and the relative error of tension fluctuation amplitude measurement was 1.5%; the SNR improvement of Group 2 (constant mode decomposition control group) was 8.4 dB, the modal aliasing rate was 16.5%, and the relative error of measurement was 9.2%; the SNR improvement of Group 3 (ablation experiment group with dynamic adjustment strategy to remove penalty factor) was 11.6 dB, the modal aliasing rate was 4.8%, and the relative error of measurement was 4.5%; and the SNR improvement of Group 4 (ablation experiment group with removal of multiple signal classification spectrum estimation step) was 10.2 dB, the modal aliasing rate was 9.4%, and the relative error of measurement was 6.1%.

[0059] The experimental results show that, under the experimental conditions, the complete scheme reduced the relative error of tension fluctuation amplitude measurement from 9.2% to 1.5% compared to the control group with constant mode decomposition, a reduction of 7.7 percentage points. This indicates that the complete scheme improved the accuracy of tension fluctuation amplitude measurement in the test scenario. The results of the third group of ablation experiments show that the introduction of a dynamic adjustment strategy for the penalty factor based on residual noise evaluation indicators and a proportional-derivative controller improved the convergence stability during the non-stationary tension signal decomposition process. The results of the fourth group of ablation experiments show that the multiple signal classification spectrum estimation step provides a more reasonable initial center frequency and number of decompositions for variational mode decomposition, thereby reducing the risk of mode aliasing and improving the energy concentration of the target tension fluctuation frequency band. Overall, the complete scheme exhibits good anti-noise interference capability and stability in tension fluctuation component extraction under experimental conditions.

[0060] This invention also discloses a dynamic tension digital measurement system for a frame-type wire and cable stranding machine, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, a dynamic tension digital measurement method for a frame-type wire and cable stranding machine according to the present invention is implemented.

[0061] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for digitally measuring the dynamic tension of a frame-type wire and cable stranding machine, characterized in that, include: S1. Collect the noisy raw tension signal output by the tension sensor of the wire laying device, separate the low-frequency trend of the noisy raw tension signal to obtain the basic tension component and the detrended AC tension signal, and use a multiple signal classification algorithm to estimate the spectrum of the detrended AC tension signal to determine the number of decompositions and the initial center frequency of the variational mode decomposition. S2. In the variational mode decomposition iteration, after each update of the intrinsic mode function components, the residual noise evaluation index of the residual signal is calculated. The penalty factor is dynamically adjusted according to the deviation and rate of change between the residual noise evaluation index and the target noise evaluation index. The iteration is terminated when the update amount of the intrinsic mode function components, the residual noise evaluation index and the rate of change of the index all meet the preset convergence conditions. S3. Obtain the main speed of the stranding machine and convert it to the target frequency reference by the rotation frequency multiplier. Select the component with the smallest distance between the center frequency and the target frequency reference as the target modal component. Extract the envelope of the target modal component by Hilbert transformation to obtain the tension fluctuation amplitude. Extract the mean value of the basic tension component as the basic tension value. Combine the preset sensor calibration coefficients to obtain the final basic tension measurement value and the final tension fluctuation amplitude.

2. The dynamic tension digital measurement method for a frame-type wire and cable stranding machine according to claim 1, characterized in that, The noisy raw tension signal output by the tension sensor of the wire feeding device is obtained by: sensing the change in mechanical tension force through a resistance strain gauge unit attached to the wire guide wheel at the end of the wire feeding reel, generating a continuous analog voltage signal; using an anti-aliasing filter circuit composed of a low-pass filter to filter out high-frequency noise from the continuous analog voltage signal, wherein the cutoff frequency of the anti-aliasing filter circuit is higher than the upper limit of the target frequency reference; and using an analog-to-digital converter to discretize the filtered signal at a preset sampling rate to generate a one-dimensional time-series digital signal array as the noisy raw tension signal.

3. The dynamic tension digital measurement method for a frame-type wire and cable stranding machine according to claim 1, characterized in that, The process of separating the low-frequency trend of the noisy original tension signal to obtain the basic tension component and the detrended AC tension signal includes: calculating the target frequency reference based on the main rotation speed of the stranding machine and the inherent mechanical vibration frequency doubling coefficient; setting the low-frequency trend separation cutoff frequency to a preset ratio lower than the target frequency reference and not higher than the preset lower limit of the main AC mode frequency band; performing zero-phase low-pass filtering on the noisy original tension signal using the low-frequency trend separation cutoff frequency to obtain the basic tension component; and subtracting the basic tension component from the noisy original tension signal to obtain the detrended AC tension signal.

4. The dynamic tension digital measurement method for a frame-type wire and cable stranding machine according to claim 1, characterized in that, The method of using a multiple signal classification algorithm to estimate the spectrum of the detrended AC tension signal and determine the number of decompositions and the initial center frequency of the variational mode decomposition includes: calculating the autocorrelation matrix of the detrended AC tension signal; performing eigenvalue decomposition on the autocorrelation matrix to divide the eigenvector space into a signal subspace and a noise subspace; constructing a frequency steering vector orthogonal to the noise subspace; calculating the spatial spectrum estimation function; scanning and drawing a spatial spectrum map within the target frequency band at a preset frequency step size; extracting extreme points in the spatial spectrum map whose amplitude is greater than a preset energy threshold; setting the frequency values ​​corresponding to the extreme points sequentially as the initial center frequency of each mode; and assigning the total number of extreme points to the variational mode decomposition algorithm as the number of decompositions.

5. The dynamic tension digital measurement method for a frame-type wire and cable stranding machine according to claim 1, characterized in that, In the variational mode decomposition iteration, after each update of the intrinsic mode function components, the residual noise evaluation index of the residual signal is calculated, including: subtracting the reconstructed sum of each current intrinsic mode function component from the detrended AC tension signal to obtain the current residual signal, and performing mean and standard deviation normalization on the current residual signal to obtain the dimensionless residual signal; calculating the Gaussianity deviation of the amplitude distribution, the residual intensity of the time autocorrelation, and the flatness deviation of the power spectrum based on the dimensionless residual signal, and weighting and fusing the three to obtain the residual noise evaluation index.

6. The dynamic tension digital measurement method for a frame-type wire and cable stranding machine according to claim 5, characterized in that, The step of dynamically adjusting the penalty factor based on the deviation and rate of change between the residual noise evaluation index and the target noise evaluation index includes: calculating the rate of change of the residual noise evaluation index between the current iteration and the previous iteration; using a proportional-derivative controller, with the deviation between the residual noise evaluation index and the target noise evaluation index as the proportional term and the rate of change of the index as the derivative term, calculating the penalty factor compensation amount in combination with a preset dimension conversion coefficient, updating the penalty factor for the next iteration cycle using the compensation amount, and applying preset upper and lower limit constraints during the update.

7. The dynamic tension digital measurement method for a frame-type wire and cable stranding machine according to claim 5, characterized in that, The calculation of the Gaussian deviation of amplitude distribution, the residual intensity of time-series autocorrelation, and the flatness deviation of power spectrum based on the dimensionless residual signal includes: generating the actual probability density distribution curve of the dimensionless residual signal using a kernel density estimation algorithm; calculating the divergence between the actual probability density distribution curve and the standard normal probability density distribution curve to obtain the Gaussian deviation of amplitude distribution; calculating the average absolute value of each order autocorrelation coefficient of the dimensionless residual signal within a preset lag order range as the residual intensity of time-series autocorrelation; calculating the flatness of the power spectrum of the dimensionless residual signal; and obtaining the flatness deviation of power spectrum based on the flatness of power spectrum.

8. The dynamic tension digital measurement method for a frame-type wire and cable stranding machine according to claim 1, characterized in that, The process of obtaining the main rotation speed of the stranding machine and converting it into a target frequency reference by rotating frequency multiplication includes: acquiring the actual operating speed of the main shaft of the stranding machine under the current working conditions in real time by using a rotary encoder installed on the main shaft; converting the actual operating speed of the main shaft from revolutions per minute to revolutions per second, and then multiplying it by the inherent mechanical vibration frequency multiplication factor related to the structure of the wire feeding device to obtain the target frequency reference.

9. The dynamic tension digital measurement method for a frame-type wire and cable stranding machine according to claim 1, characterized in that, The selection of the component with the smallest distance between its center frequency and the target frequency reference as the target mode component includes: performing a Fast Fourier Transform on each intrinsic mode function component output by variational mode decomposition to obtain the discrete spectrum sequence of each intrinsic mode function component; searching for the main peak with the largest amplitude in the discrete spectrum sequence of each intrinsic mode function component, and taking the frequency value corresponding to the largest main peak as the actual center frequency of the corresponding intrinsic mode function component; calculating the absolute difference between the actual center frequency of each intrinsic mode function component and the target frequency reference; and taking the intrinsic mode function component with the smallest absolute difference as the target mode component representing tension fluctuation.

10. The dynamic tension digital measurement method for a frame-type wire and cable stranding machine according to claim 1, characterized in that, The process of extracting the envelope of the target modal components using Hilbert transform to obtain the tension fluctuation amplitude, extracting the mean of the basic tension components as the basic tension value, and combining them with preset sensor calibration coefficients to obtain the final basic tension measurement value and the final tension fluctuation amplitude includes: performing a Hilbert transform on the target modal components, constructing an analytical signal by combining orthogonal Hilbert transform signals, and calculating the complex modulus to obtain the instantaneous envelope signal; calculating the time arithmetic mean of the instantaneous envelope signal and the basic tension components within the steady-state measurement time window, respectively, as the tension fluctuation amplitude and the basic tension value; and substituting the basic tension value and the tension fluctuation amplitude into a dimension conversion function containing preset sensor calibration coefficients to convert them from the voltage dimension to the mechanical dimension, respectively, to obtain the final basic tension measurement value and the final tension fluctuation amplitude.

Images