A method for quantitatively detecting damage of a steel wire rope

Abstract

The application provides a steel wire rope damage quantitative detection method, the method eliminates abnormal signal points by setting a threshold value for signals collected by a steel wire rope detection device, adopts a detrending method to eliminate baseline drift, then carries out noise reduction processing through a self-adaptive wavelet denoising algorithm, optimizes parameters of SVR through a heuristic optimization algorithm, and finally determines an SVR quantitative analysis model. The application carries out nondestructive testing on the steel wire rope through the construction of a steel wire rope damage quantitative detection method, and improves the detection precision.

Description

Technical Field

[0001] This invention relates to wire rope damage detection technology, specifically to a quantitative detection method for wire rope damage based on pretreatment and SVR. Background Technology

[0002] Steel wire ropes are commonly used for lifting and traction of heavy objects. If a steel wire rope breaks, the impact force is extremely strong, which can easily cause personal injury and property damage. However, replacing steel wire ropes too frequently will result in significant waste and economic losses. Therefore, the ability to accurately detect the damage to steel wire ropes and thus predict their service life has important social and economic benefits.

[0003] Electromagnetic testing is currently the most mature, reliable, and widely used non-destructive testing method for steel wire ropes, offering advantages such as low cost and ease of implementation. Metal magnetic memory testing technology is a rapid non-destructive testing method that utilizes the magnetic memory effect of metals to detect stress concentration areas in components. It overcomes the shortcomings of traditional non-destructive testing, enabling the diagnosis of stress concentration zones within ferromagnetic metal components, including microscopic defects, early failures, and damage, thus preventing sudden fatigue damage.

[0004] Based on the fundamental principle of the magnetic memory effect in metals, this method records the distribution of magnetic field strength components perpendicular to the surface of a metal component along a certain direction. The magnetic memory signal is primarily influenced by stress and defects, allowing for the diagnosis and evaluation of stress concentration and the presence of microscopic defects, thus preventing sudden fatigue damage. In the use of wire ropes, magnetic memory detection can be used to inspect them, enabling real-time detection of damage. While Hall effect sensors can acquire leakage magnetic signals at defects, these signals are affected by magnetization, lift-off distance, wire diameter, and fracture width. Furthermore, sensors are inevitably susceptible to noise interference. Therefore, effective algorithms are needed to analyze and process the leakage magnetic signals to improve the accuracy of wire rope defect detection. Summary of the Invention

[0005] To address the above problems, this invention provides a quantitative detection method for steel wire rope damage. It employs the WT-GSNDT steel wire rope detection device, which can acquire, record, and store the leakage magnetic information of the steel wire rope in real time with high speed and high sensitivity. It also uses machine learning algorithms to identify local defects and displays the detection results in real time through a graphical interface.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a method for quantitative detection of wire rope damage, characterized by comprising the following steps:

[0007] Step 1: Extract the magnetic memory signals of different damage conditions of the wire rope detected by the wire rope detection device;

[0008] Step 2: Eliminate abnormal signal points by setting threshold values;

[0009] Step 3: Fit a curve to the signal data and use a detrending method to eliminate baseline drift;

[0010] Step 4: Perform noise reduction processing on the detrended signal data;

[0011] Step 5: Create a dataset of steel wire rope damage samples, with each sample containing signal data and a label;

[0012] Step 6: Divide the sample set into a training set, a validation set, and a test set;

[0013] Step 7: Construct an SVR quantitative analysis model, using the training set as input, and employ the GAPSO algorithm to optimize the penalty parameter c and kernel function parameter g in the SVR, outputting quantitative detection results;

[0014] Step 8: Calculate the model accuracy;

[0015] Step 9: Validation set validation, adjust model parameters, and determine the SVR quantitative analysis model;

[0016] Step 10: Input the test set into the SVR quantitative analysis model to obtain the quantitative detection results and calculate the detection accuracy;

[0017] Step 11: Use a wire rope detection device to collect real-time leakage magnetic signal data of the wire rope. Perform the abnormal signal point elimination, baseline drift elimination, and noise reduction processing steps 2-4 on the collected signal data. Use the processed data as input and perform quantitative analysis using the SVR quantitative analysis model to output the damage status of the wire rope.

[0018] Preferably, in step 2, for short interference pulse signals, outliers, and spurious signals, a threshold value T is set, and the outliers are smoothed using a moving window, employing the following method:

[0019]

[0020] Among them, the spatial domain signal sequence is {x m}, m=0,1,2,Λ, First, determine whether the signal point is a singular point. For signal points that satisfy t(x)≥T, use the moving window smoothing method with a window width of (2k+1). Calculate the average value of k points before and after the center point m of the window, t(x)=|x m -[x m-k +x m-k+1 +Λ+x m-1 +x m+1 +Λ+x m+k-1 +x m+k / 2k|, T is preset by the amplitude of the signal change and experience, and the amplitude of the signal change is the derivative of the signal data.

[0021] Preferably, in step 3, polynomial fitting is performed on a number of signal data to obtain a fitting curve of the data trend. The specific steps for obtaining the fitting curve are as follows:

[0022] In the current problem, m-degree polynomial fitting is adopted. Let f(x) = a1x + a2x 2 + Λ + a m x m

[0023] where, a k is an undetermined coefficient (k = 1, 2, Λ, m, m << n), and the fitting criterion is to minimize the sum of the squares of the distances between the signal data y i (i = 1, 2, Λ, n) and f(x i ). f(x i ) is the extended extension of f(x), and y i is the actual value of the i-th sample, that is, to make reach the minimum. The selection of the fitting order m is determined according to the existing signal data. Specifically, m is taken in sequence starting from 3, and the appropriate value of m is determined according to the fitting situation, so that the fitting curve can just reflect the data trend.

[0024] After obtaining the data trend curve f(x), the trend line f(x) is subtracted from the original data according to the detrending method to eliminate the baseline drift.

[0025] Preferably, in step 4, an adaptive wavelet denoising algorithm is adopted. The appropriate wavelet basis and the decomposition level J are selected according to the data characteristics. Based on the results of wavelet transform of each layer, the optimal threshold is found by using the Stein Unbiased Risk Estimation (SURE) optimization algorithm to realize the adaptive change of the threshold.

[0026] Preferably, in step 6, according to the SPXY method, the distance between samples is calculated as d xy (i, j),

[0027]

[0028] where, the distance between samples calculated with signal data as the characteristic parameter is d x (i, j), and the distance between samples calculated with the label value as the characteristic parameter is d y (i, j).

[0029] Preferably, in step 7, the values ​​of the penalty parameter c and the kernel function parameter g in the SVR are used as the positions of the particles. The particle swarm optimization algorithm is optimized using the idea of ​​genetic algorithm. By crossbreeding and mutation of particles, the diversity of the population is increased, avoiding premature convergence of the particle swarm optimization algorithm and avoiding getting trapped in local optima. The algorithm steps are as follows:

[0030] (1) Initialize the random positions and velocities of the particles in the particle swarm, so that the particles are dispersed throughout the space. The position of the i-th particle corresponds to the values ​​of parameters c and g, x. i =(c i ,g i The velocity of position change of the i-th particle is v. i =(v i1 ,v i2 Generally speaking, the number of particles ranges from 20 to 40.

[0031] (2) Use the root mean square error of the model training set as the fitness value, and calculate the fitness of each particle according to the fitness function;

[0032] (3) Compare the fitness value of each particle with the best historical position of this particle (denoted as gbest). If the current fitness value is better, then the fitness value is taken as the best current position gbest. Otherwise, gbest remains unchanged. Compare the fitness value of each particle with the best historical position of all these particles (denoted as zbest). If the current fitness value is better, then the fitness value is taken as the best current position zbest. Otherwise, zbest remains unchanged.

[0033] (4) Sort the particles according to their fitness, hybridize them, calculate the position and velocity of the offspring particles, and compare the fitness of the offspring particles with that of the parent particles. If the offspring's fitness is better, replace the parent's velocity and position with the offspring's velocity and position. Calculate the velocity and position of the offspring particles according to the following:

[0034] childv=(v1+v2)×|v1| / |v1+v2|

[0035] childx=rand()×x1+(1-rand())×x2

[0036] Where childv is the velocity of the offspring particle, v1 and v2 are the velocities of the selected particles to be hybridized, x1 and x2 are the positions of the selected particles to be hybridized, and rand() is a random number between (0 and 1).

[0037] (5) Select a particle to mutate, calculate the position and velocity of the mutated particle, compare the fitness of the mutated particle with that of the original particle, and if the fitness of the mutated particle is better, replace the original particle's position with the mutated particle; calculate the position of the mutated particle according to the following:

[0038] mutation = x3 × (1 + rand())

[0039] Where x3 is the position of the selected particle to be mutated;

[0040] (6) Calculate the fitness of the particles after hybridization and mutation. Compare the fitness value of each particle with the fitness value of the best position found by all particles (denoted as zbest, zbest is the best value among gpbest, which is the global best position of the entire population in one iteration). If the current fitness value is better, then this position is taken as the global best position zbest of all particles. Otherwise, zbest remains unchanged.

[0041] (7) Update the particle's velocity and position according to the following formula:

[0042] v id (new) = w × v id (old)+c1×rand()×(g id -x id )+c2×rand()×(z id -x id )

[0043] x id (new) = x id (old)+μ×v id (new)

[0044] Among them, v id Let x be the particle velocity. id For the particle position, g id For the individual's optimal position gbest, z id Let z be the global optimal position, d = 1, 2; w be the inertia weight; rand() be a random number between (0, 1); μ be the constraint factor, used to control the speed weights; c1 and c2 are learning factors, also known as acceleration constants, calculated as follows:

[0045] c1 = c 1s +(c 1e -c 1s )×iter / iter max

[0046] c2 = c 2s +(c 2e -c2s )×iter / iter max

[0047] Among them, c 1s and c 2s Let c1 and c2 be the initial values ​​of the learning factors, respectively. 1e and c 2e These are the final values ​​of iterations c1 and c2, respectively, where iter is the current iteration number. max To determine the maximum number of iterations, the range of c1 is (1, 2.5), and the range of c2 is (1.5, 2.75).

[0048] (8) Repeat steps (2) to (7) until the termination condition is met, i.e., the set number of iterations is reached. Obtain the optimized parameters c and g, and determine the SVR model.

[0049] Preferably, the accuracy calculation in step 8 includes root mean square error, correlation coefficient, and accuracy rate.

[0050] Root mean square error (RMSE)

[0051]

[0052] Among them, y i Let ypred be the actual value of the i-th sample. i This is the predicted value for the i-th sample. When n is the number of samples in the training set, the root mean square error of the training set is calculated. When n is the number of samples in the validation set, the root mean square error of the validation set is calculated. The smaller the root mean square error of the validation set, the stronger the model's ability to predict external samples.

[0053] Coefficient of determination (R) 2 Or the correlation coefficient (R)

[0054]

[0055] Among them, y i Let ypred be the actual value of the i-th sample. i Let y be the predicted value of the i-th sample, y be the average of the actual values ​​of all samples, and n be the number of samples in the calibration set or prediction set.

[0056] Accuracy

[0057]

[0058] Among them, TP (True Positive): the number of positive classes predicted as positive; FN (False Negative): the number of negative classes predicted as negative; FP (False Positive): the number of negative classes predicted as positive (Type I error); TN (True Negative): the number of positive classes predicted as negative (Type II error).

[0059] Preferably, in step 1, one data point is collected every 10ms, and each data point has 4000 data points, for a total of 600 data points are collected.

[0060] This invention provides a method for quantitative detection of wire rope damage, which has the following beneficial effects:

[0061] 1. During data preprocessing, noise interference is reduced by eliminating abnormal signal points, eliminating baseline drift, and denoising.

[0062] 2. A quantitative analysis model for SVR (Support Vector Regression) was constructed, and the parameters in SVR were optimized using the GAPSO (Genetic Particle Swarm Optimization) algorithm, which effectively improved the model's detection accuracy.

[0063] 3. Compared to neural network algorithms, SVR can better avoid overfitting, and the quantitative detection results are more reliable;

[0064] 4. This invention enables quantitative detection of damage to steel wire ropes during use, with accurate and efficient calculations. Attached Figure Description

[0065] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0066] Figure 1 A flowchart for a quantitative detection method for wire rope damage;

[0067] Figure 2 Flowchart for optimizing SVR model parameters for GAPSO;

[0068] Figure 3 The waveform of the magnetic memory signal before noise reduction processing is shown.

[0069] Figure 4 This is a waveform diagram of the magnetic memory signal after noise reduction processing. Detailed Implementation

[0070] like Figure 1 As shown, the quantitative detection method for wire rope damage includes the following steps:

[0071] Step 1: Extract the magnetic memory signals of different damage conditions of the wire rope detected by the wire rope detection device. Among them, the damage range of the wire rope is 0-25%, and there are 25 wire ropes in total. For the 10 wire ropes with a damage range of 0-10%, 36 data are randomly collected for each wire rope. For the 15 wire ropes with a damage range of 10-25%, 16 data are randomly collected for each wire rope. One data is collected every 10 ms, and each data collects 4000 points, with a total of 600 data collected; the wire rope detection device uses the WT-GSNDT wire rope detection device.

[0072] Step 2: Eliminate abnormal signal points by setting a threshold value;

[0073] For singular points such as short interference pulse signals, wild points, and stray signals, set a threshold value T, and perform moving window smoothing on the abnormal points. The following method is adopted:

[0074]

[0075] Among them, the spatial domain signal sequence is {x m}, m = 0, 1, 2, Λ. First, judge whether the signal point is a singular point. For the signal points that meet t(x)≥T, use the moving window smoothing method. The window width is (2k + 1), and calculate the average value of k points before and after the center point m of the window. t(x)=|x m -[x m-k +x m-k+1 +Λ+x m-1 +x m+1 +Λ+x m+k-1 +x m+k / 2k|, and T is preset by the amplitude of the signal change and experience. The amplitude of the signal change is the derivative of the signal data.

[0076] Step 3: Fit a curve for the signal data and use the detrending method to eliminate the baseline drift;

[0077] Perform polynomial fitting on 600 signal data to obtain the fitting curve of the data trend. The specific steps to obtain the fitting curve are as follows:

[0078] In the current problem, use the m-th order polynomial fitting, and let f(x)=a1x + a2x 2 +Λ+a m x m

[0079] Among them, a k is the undetermined coefficient (k = 1, 2, Λ, m, m << n). The fitting criterion is to make the sum of the squares of the distances between the signal data y i (i = 1, 2, Λ, n) and f(x i ) the smallest, f(x i) is an extension of f(x), y i For the actual value of the i-th sample, that is, to obtain To achieve a minimum, the number of fitting iterations m is determined based on the existing signal data. Specifically, m is selected sequentially starting from 3, and a suitable value for m is determined based on the fitting results, so that the fitted curve perfectly reflects the data trend.

[0080] After obtaining the data trend curve f(x), the trend line f(x) is subtracted from the original data using the detrending method, thereby eliminating baseline drift.

[0081] Step 4: Perform noise reduction processing on the detrended signal data;

[0082] An adaptive wavelet denoising algorithm is adopted. Based on the data characteristics, an appropriate wavelet basis and decomposition level J are selected. Based on the results of wavelet transform at each level, the optimal threshold is found using the Stein unbiased estimation (SURE) optimization algorithm to achieve adaptive change of the threshold.

[0083] Step 5: Establish a dataset of steel wire rope damage samples, where each sample contains preprocessed signal data and a label;

[0084] For each data point, mark the problematic signal peak and the damage value of the wire rope.

[0085] Step 6: Divide the sample set into a training set, a validation set, and a test set;

[0086] According to the SPXY method, the distance between samples is calculated as d. xy (i,j).

[0087]

[0088] Wherein, the distance between samples calculated using signal data as a feature parameter is d. x (i,j), where d is the distance between samples calculated using the label value as a feature parameter. y (i,j).

[0089] The SPXY method was used to divide the 600 samples into three groups: 400 training samples, 100 validation samples, and 100 test samples, to ensure that the training samples were highly representative.

[0090] Step 7: Construct an SVR quantitative analysis model, using the training set as input, and employ the GAPSO algorithm to optimize the penalty parameter c and kernel function parameter g in the SVR, outputting quantitative detection results;

[0091] The values ​​of the penalty parameter c and kernel function parameter g in SVR are used as the particle positions. A genetic algorithm is employed to optimize the particle swarm optimization (PSO) algorithm. By crossbreeding and mutation of particles, the diversity of the population is increased, preventing premature convergence and avoiding getting trapped in local optima. Figure 2 As shown, the algorithm steps are as follows:

[0092] (1) Initialize the random positions and velocities of the particles in the particle swarm, so that the particles are dispersed throughout the space. The position of the i-th particle corresponds to the values ​​of parameters c and g, x. i =(c i ,g i The velocity of position change of the i-th particle is v. i =(v i1 ,v i2 Generally speaking, the number of particles ranges from 20 to 40.

[0093] (2) Use the root mean square error of the model training set as the fitness value, and calculate the fitness of each particle according to the fitness function.

[0094] (3) Compare the fitness value of each particle with its historical best position (denoted as gbest). If the current fitness value is better, then this fitness value is taken as the current best position gbest; otherwise, gbest remains unchanged. Compare the fitness value of each particle with the historical best positions of all particles (denoted as zbest). If the current fitness value is better, then this fitness value is taken as the current best position zbest; otherwise, zbest remains unchanged.

[0095] (4) Sort the particles according to their fitness, hybridize them, calculate the position and velocity of the offspring particles, and compare the fitness of the offspring particles with that of the parent particles. If the offspring's fitness is better, replace the parent's velocity and position with the offspring's velocity and position. Calculate the velocity and position of the offspring particles according to the following:

[0096] childv=(v1+v2)×|v1| / |v1+v2|

[0097] childx=rand()×x1+(1-rand())×x2

[0098] Where childv is the velocity of the offspring particle, v1 and v2 are the velocities of the selected particles to be hybridized, x1 and x2 are the positions of the selected particles to be hybridized, and rand() is a random number between (0 and 1).

[0099] (5) Select a particle to mutate, calculate the position and velocity of the mutated particle, and compare the fitness of the mutated particle with that of the original particle. If the fitness of the mutated particle is better, then replace the original particle's position with the mutated particle. Calculate the position of the mutated particle based on the following:

[0100] mutation = x3 × (1 + rand())

[0101] Where x3 is the position of the selected particle to be mutated.

[0102] (6) Calculate the fitness of the particles after hybridization and mutation. Compare the fitness value of each particle with the fitness value of the best position found by all particles (denoted as zbest, zbest is the best value among gpbest, which is the global best position of the entire population in one iteration). If the current fitness value is better, then this position is taken as the global best position zbest of all particles. Otherwise, zbest remains unchanged.

[0103] (7) Update the particle's velocity and position according to the following formula:

[0104] v id (new) = w × v id (old)+c1×rand()×(g id -x id )+c2×rand()×(z id -x id )

[0105] x id (new) = x id (old)+μ×v id (new)

[0106] Among them, v id Let x be the particle velocity. id For the particle position, g id For the individual's optimal position gbest, z id Let z be the global optimal position, d = 1, 2; w be the inertia weight; rand() be a random number between (0, 1); μ be the constraint factor, used to control the speed weights; c1 and c2 are learning factors, also known as acceleration constants, calculated as follows:

[0107] c1 = c 1s +(c 1e -c 1s )×iter / iter max

[0108] c2 = c 2s +(c 2e -c2s )×iter / iter max

[0109] Among them, c 1s and c 2s Let c1 and c2 be the initial values ​​of the learning factors, respectively. 1e and c 2e These are the final values ​​of iterations c1 and c2, respectively, where iter is the current iteration number. max To determine the maximum number of iterations, the range of c1 is (1, 2.5), and the range of c2 is (1.5, 2.75).

[0110] (8) Repeat steps (2) to (7) until the termination condition is met, i.e., the set number of iterations is reached. Obtain the optimized parameters c and g, and determine the SVR model.

[0111] Step 8: Calculate the model accuracy;

[0112] The model building and validation process involves some statistical parameters used to evaluate the model's accuracy, including root mean square error, correlation coefficient, and accuracy.

[0113] Root mean square error (RMSE)

[0114]

[0115] Among them, y i Let ypred be the actual value of the i-th sample. i This is the predicted value for the i-th sample. When n is the number of samples in the training set, the root mean square error of the training set is calculated. When n is the number of samples in the validation set, the root mean square error of the validation set is calculated. The smaller the root mean square error of the validation set, the stronger the model's predictive ability for external samples.

[0116] Coefficient of determination (R) 2 Or the correlation coefficient (R)

[0117]

[0118] Among them, y i Let ypred be the actual value of the i-th sample. i Let be the predicted value of the i-th sample, y be the average of the actual values ​​of all samples, and n be the number of samples in the calibration set or prediction set. The closer the R-value is to 1, the better the model's prediction results should be.

[0119] Accuracy

[0120]

[0121] Among them, TP (True Positive): the number of positive classes predicted as positive; FN (False Negative): the number of negative classes predicted as negative; FP (False Positive): the number of negative classes predicted as positive (Type I error); TN (True Negative): the number of positive classes predicted as negative (Type II error). The closer the ACC value is to 1, the higher the prediction accuracy of the model.

[0122] Step 9: For the SVR quantitative analysis model trained on the training set, take the validation set as input, output the quantitative detection results and calculate the model accuracy. Adjust the model parameters according to the results of the validation set, and finally determine the SVR quantitative analysis model.

[0123] Step 10: Input the test set into the SVR quantitative analysis model to obtain the quantitative detection results and calculate the detection accuracy;

[0124] Step 11: Real-time acquisition of wire rope leakage magnetic field signal data is performed using a wire rope detection device. The acquired signal data undergoes the abnormal signal point elimination, baseline drift elimination, and noise reduction processes described in steps 2-4. The processed data is then used as input for quantitative analysis using the SVR quantitative analysis model to output the wire rope damage status. The wire rope detection device used is the WT-GSNDT wire rope detection device.

[0125] Meanwhile, the wire rope is repeatedly tested during the actual testing process to improve the accuracy of the test.

[0126] Figure 3 , Figure 4 The waveforms before and after denoising a portion of the magnetic memory signal extracted from a data set are shown to compare the denoising effects.

[0127] This invention provides a quantitative detection method for wire rope damage. The method eliminates abnormal signal points by setting threshold values ​​for signals acquired by the WT-GSNDT wire rope testing device, eliminates baseline drift using a detrending method, performs noise reduction using an adaptive wavelet denoising algorithm, and optimizes the SVR parameters using a heuristic optimization algorithm to finally determine the quantitative SVR analysis model. This invention improves the detection accuracy by constructing a quantitative detection method for wire rope damage, enabling non-destructive testing of wire ropes.

Claims

1. A method for quantitatively detecting damage of a steel wire rope, characterized by, The method comprises the following steps: Step 1: Extract the magnetic memory signals of different damage conditions of the steel wire detected by the steel wire detection device; Step 2: Eliminate abnormal signal points by setting a threshold value; Step 3: Fit a curve for the signal data, and eliminate baseline drift by using a detrending method; Step 4: Perform noise reduction processing on the signal data after detrending processing; Step 5: Establish a steel wire damage sample data set, each sample containing signal data and a label; Step 6: Divide the sample set into a training set, a validation set and a test set; Step 7: Construct an SVR quantitative analysis model, take the training set as input, optimize the penalty parameter c and the kernel function parameter g in the SVR by using the GAPSO algorithm, and output the quantitative detection result; Step 8: Calculate the model accuracy; Step 9: Verify the model parameters and determine the SVR quantitative analysis model; Step 10: Input the test set into the SVR quantitative analysis model to obtain the quantitative detection result and calculate the detection accuracy; Step 11: Real-time acquisition of the magnetic leakage signal data of the steel wire by using the steel wire detection device, elimination of abnormal signal points, elimination of baseline drift and noise reduction processing on the acquired signal data according to steps 2-4, input of the processed data, quantitative analysis by using the SVR quantitative analysis model, and output of the damage condition of the steel wire; In step 7, the values of the penalty parameter c and the kernel function parameter g in the SVR are taken as the positions of particles, the particle swarm algorithm is optimized by using the idea of genetic algorithm, the diversity of the population is increased by hybridization and mutation of the particles, premature convergence of the fitness of the particle swarm algorithm is avoided, and local optimization is avoided, and the algorithm steps are as follows: (1) the random position and speed of the particle in the particle swarm are initialized, so that the particles are dispersed in the whole space; the position of the i-th particle corresponds to the value of the parameters c and g , the speed of the i-th particle position change Generally, the particle number is in the range of 20-40; (2) Take the root mean square error of the model training set as the fitness value, and calculate the fitness of each particle according to the fitness function; (3) Compare the fitness value of each particle with the historical optimal position of this particle, denoted as gbest, if the current fitness value is better, then take this fitness value as the current best position gbest, otherwise, gbest remains unchanged; Compare the fitness value of each particle with the historical optimal position of all particles, denoted as zbest, if the current fitness value is better, then take this fitness value as the current best position zbest, otherwise, zbest remains unchanged; (4) Sort according to the fitness value, hybridize the particles, calculate the position and speed of the offspring particles, compare the fitness of the offspring particles with the fitness of the parent particles, if the fitness of the offspring is better, then replace the position and speed of the parent with the position and speed of the offspring; the speed and position of the offspring particles are calculated as follows: wherein, is the velocity of the child particle, is the velocity of the selected particle to be hybridized, is the position of the selected particle to be hybridized, and rand() is a random number between (0, 1). (5) Select particles for mutation, calculate the position and speed of the mutated particles, compare the fitness of the mutated particles with the fitness of the original particles, if the fitness of the mutated particles is better, then replace the position of the original particles with the position of the mutated particles; the position of the mutated particles is calculated as follows: ； wherein, is the position of the selected particle to be mutated; (6) Calculate the fitness of the hybridized particle, compare the fitness value of each particle with the fitness value of the optimal position found by all particles, if the current fitness value is better, then take this position as the global optimal position zbest of all particles, otherwise, zbest remains unchanged; (7) Update the velocity and position of the particle according to the following formula: ； ； where, is the particle velocity, is the particle position, is the individual best position gbest, is the global best position zbest, ; w is the inertia weight; rand() is a random number between (0, 1); μ is the constraint factor for controlling the weight of the velocity; and is the learning factor, also known as the acceleration constant, and is calculated as follows: ； ； wherein, and are initial values of learning factors and respectively, and are final values of and respectively, is the current iteration number, is the maximum iteration number, varies in the range (1, 2.5), varies in the range (1.5, 2.75); (8) Repeat steps (2)-(7) until the end condition is met, i.e. the set number of iterations is reached; obtain the optimized parameters c and g, and determine the SVR model.

2. The method of claim 1, wherein the method further comprises: In step 2, for short interference pulse signals, outliers and stray signals, set a threshold value T, and perform moving window smoothing processing on abnormal points, using the following method: ； Wherein, the spatial domain signal sequence is Firstly, it is judged whether the signal point is singular point or not. For the signal point meeting , the moving window smoothing method is adopted, the window width is , the average value of k points before and after the center point m of the window is calculated, , T is preset by experience according to the amplitude of signal change, i.e. the derivative of signal data.

3. The method of claim 1, wherein the step of detecting the damage of the steel wire rope is characterized by, In step 3, a polynomial fitting is performed on a plurality of signal data to obtain a fitting curve of the data trend, and the specific steps of obtaining the fitting curve are as follows: In the current problem, a polynomial of degree m is fitted, letting ; wherein, is a pending coefficient , the fitting criterion is to minimize the sum of squares of distances between signal data and , is an extended extension of , is the actual value of the i-th sample, that is, to make reach a minimum, the selection of the fitting number m is determined according to the existing signal data, specifically, m is taken from 3 in turn, and the appropriate value of m is determined according to the fitting condition, so that the fitting curve can just reflect the data trend. The data trend curve is obtained After, the trend line is subtracted from the original data according to the detrending method Thus, the baseline drift is eliminated.

4. The method of claim 1, wherein the method further comprises: In step 4, an adaptive wavelet denoising algorithm is used, a suitable wavelet basis and decomposition layer number J are selected according to the data characteristics, the optimal threshold is found based on the Stein unbiased estimate optimization algorithm according to the result of each layer wavelet transform, and the adaptive change of the threshold is realized.

5. The method of claim 1, wherein the step of determining the damage of the steel wire rope is characterized by, The step 6 calculates the distance between samples according to the SPXY method as , ； wherein the distance between samples calculated with the signal data as characteristic parameter is the distance between samples calculated with the label value as characteristic parameter is .

6. The method of claim 1, wherein the step of determining the damage of the steel wire rope is characterized by, The accuracy calculation in step 8 includes root mean square error, correlation coefficient and accuracy, Root mean square error ； wherein, is the actual value of the i-th sample, is the predicted value of the i-th sample; when n is the number of samples in the training set, the root mean square error of the training set is calculated at this time, when n is the number of samples in the validation set, the root mean square error of the validation set is calculated at this time, the smaller the root mean square error of the validation set, the stronger the prediction ability of the model to external samples; coefficient of determination (R2) ) or correlation coefficient (r) ) ； wherein, is the actual value of the i-th sample, is the predicted value of the i-th sample, is the average of the actual values of all samples, n is the number of samples in the calibration set or in the prediction set; Accuracy ； Wherein, TP: the number of positive classes predicted as positive classes; FN: the number of negative classes predicted as negative classes; FP: the number of negative classes predicted as positive classes false alarm; TN: the number of positive classes predicted as negative classes false alarm.

7. The method of claim 1, wherein the step of determining the damage of the steel wire rope is characterized by, In step 1, 10ms is collected for one data, 4000 points are collected for each data, and a total of 600 data are collected.

Images