X80 pipeline steel defect quantitative characterization method based on BP neural network optimized by improved crayfish optimization algorithm

Abstract

The invention discloses an X80 pipeline steel defect quantitative characterization method based on a BP neural network optimized by an improved crayfish optimization algorithm. The method comprises the following steps: S1, acquiring experiment and simulation data; s2, extracting a time-frequency characteristic quantity, and constructing a time-domain characteristic vector T and a frequency-domain characteristic vector F; s3, performing feature dimension reduction based on Spearman correlation analysis; s4, inputting the feature vector I after dimension reduction into an improved crayfish optimization algorithm to optimize a BP neural network (CHCOA-BP) algorithm; and S5, outputting a defect length value and a defect depth value of the X80 pipeline steel pipeline volume defect. According to the CHCOA-BP algorithm constructed by the invention, three strategies of mirror reflection learning, a Tianwan algorithm extended search stage and a vertical crossover operation are fused, a mapping relation between a pulse eddy current response-based time-frequency feature fusion feature quantity and a volume defect size is obtained through model training, and quantitative characterization of pipeline defects is realized. The method has the characteristic of high prediction accuracy, effectively solves the problems that the traditional quantitative algorithm is low in later search efficiency and is easy to fall into local optimum, and effectively improves the quantitative detection and characterization capability of the volume defect size.

Description

Technical Field

[0001] The present invention relates to the technical field of non-destructive testing, and is a method for quantitatively characterizing X80 pipeline steel defects by optimizing a BP neural network based on an improved crayfish optimization algorithm, which is applicable to the intelligent detection and quantitative characterization of pipeline defects. Background Art

[0002] With the development of oil and gas transmission pipelines towards high pressure, large diameter, and high steel grade, X80 high-strength steel has become the preferred material for long-distance pipelines due to its excellent mechanical properties and economic benefits. However, in a complex service environment, pipeline steel is prone to defects such as wall thickness reduction, cracks, and holes due to factors such as corrosion, stress concentration, corrosion, mechanical damage, and stress fatigue. In severe cases, it may lead to leakage or even explosion accidents, threatening the safe operation of pipelines. Therefore, the development of high-precision and high-efficiency pipeline defect detection technology is crucial for pipeline structural integrity assessment.

[0003] Pulsed Eddy Current (PEC) testing has unique advantages in defect quantitative assessment due to its non-contact and broadband characteristics, and has the advantage of multi-parameter analysis of response results, and is widely used in pipeline steel defect detection. The rise of artificial intelligence has provided new ideas for the application of pulsed eddy current testing technology in pipeline intelligent detection. The conventional BP neural network is one of the commonly used artificial intelligence algorithms for defect inversion modeling, but it has problems such as slow convergence speed, easy to fall into local extrema, and sensitivity to initial weight thresholds, resulting in large defect quantification errors. Especially for materials such as X80 steel with complex microstructures, it is more difficult to accurately establish the non-linear mapping relationship between the response signal and the defect size. Existing optimization algorithms (such as genetic algorithms and particle swarm algorithms) still have problems such as premature convergence when optimizing BP network parameters, and are difficult to adapt to the high-dimensional and strong-noise characteristics of X80 steel defect signals. In recent years, the Crayfish Optimization Algorithm (COA) has received attention due to its global search ability in complex problems, but it still has problems such as poor population diversity and insufficient local search accuracy, and it is difficult to balance model accuracy and efficiency when directly used for BP network optimization. Summary of the Invention

[0004] The present invention provides a method for quantitatively characterizing X80 pipeline steel defects by optimizing a BP neural network based on an improved crayfish optimization algorithm, which overcomes the above-mentioned deficiencies of the prior art. The purpose is to enhance the convergence ability and global search ability of COA by introducing mirror reflection learning, the extended search stage of the Aquila algorithm, and the vertical crossover operation strategy, and construct a BP network model with time-frequency feature fusion to achieve high-precision inversion of X80 steel defect sizes and provide reliable technical support for pipeline safety assessment.

[0005] The technical solution of the present invention is as follows:

[0006] A method for quantitatively characterizing X80 pipeline steel defects by optimizing the BP neural network based on an improved crayfish optimization algorithm, comprising the following steps:

[0007] Step S1, experimental and simulation data acquisition;

[0008] Step S2, time-frequency feature extraction, constructing a time-domain feature vector T and a frequency-domain feature vector F;

[0009] Step S3, feature dimensionality reduction based on Spearman correlation analysis;

[0010] Step S4, input the reduced feature vector I into the improved crayfish optimization algorithm to optimize the BP neural network (CHCOA-BP) algorithm;

[0011] Step S5, output the defect length value and defect depth value of the X80 pipeline steel pipe volume defect.

[0012] Preferably, the operation of step S1 is specifically as follows:

[0013] Collect experimental data of prefabricated artificial volume defects, construct a finite element model to collect simulation data, jointly form a sample data set, and the collected data is pulsed eddy current detection signals.

[0014] Preferably, the operation of step S2 is specifically as follows:

[0015] In order to fully characterize the defect information, statistical analysis methods are used to extract time-domain characteristic parameters such as the mean value, standard deviation, skewness factor, kurtosis factor, maximum value, minimum value, waveform factor, pulse factor, margin factor, and energy of the sample data set, so as to construct the time-domain feature vector T. The frequency-domain characteristic parameters such as AM, CF, MSF, RMSF, and FVAR of the sample data set are extracted using FFT, so as to construct the frequency-domain feature vector F.

[0016] Preferably, the operation of step S3 is specifically as follows:

[0017] Adopt the Spearman correlation analysis method, calculate the Spearman correlation coefficients between the characteristic parameters and the defect length and depth, and select the feature dimension with the highest correlation with the predicted value as the model input feature vector I.

[0018] Preferably, the quantitative characterization mathematical model established in step S4 uses the improved crayfish optimization algorithm to optimize the BP neural network (CHCOA-BP) algorithm to realize the prediction of the defect length and depth dimension parameters. Specifically, it is embodied as:

[0019] Firstly, the mirror reflection learning strategy is used to initialize the crayfish population to increase the diversity of the population and improve the quality of the population, which helps the algorithm to find the global optimal solution faster in the subsequent search process; secondly, the extended search strategy of the Eagle optimization algorithm is introduced in the summer vacation stage of the crayfish optimization algorithm to expand the cave search range, in order to improve the global search ability and convergence speed of the algorithm and avoid falling into the local optimum; finally, the vertical crossover operation is used to exchange information and competitively screen the individuals of the crayfish optimization algorithm in different dimensions, continuously generate new and better solutions, help escape the dilemma of falling into the local optimum, and enhance the global search ability of the crayfish algorithm, so as to intelligently optimize the weights and thresholds of the BP neural network, help the BP neural network find the global optimal point, achieve global convergence and avoid local optimum.

[0020] The present invention optimizes the BP neural network mathematical algorithm by improving the crayfish optimization algorithm, and establishes a mapping relationship between the pulse eddy current response time-frequency feature fusion feature quantity and the volume defect size, thereby achieving the purpose of quantitative characterization of pipeline defects; the present invention has the characteristics of high regression accuracy and effectively improves the prediction accuracy of the defect size. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flow chart of the method for quantitative characterization of X80 pipeline steel defects based on the improved crayfish optimization algorithm to optimize the BP neural network.

[0022] Figure 2 Optimize the BP neural network model flow chart to improve the crayfish optimization algorithm. DETAILED DESCRIPTION

[0023] The present invention is further described in detail below in conjunction with specific implementations. The present invention is not limited by the following embodiments, and the specific implementations can be adjusted and determined according to the technical solutions of the present invention.

[0024] like Figure 1 As shown, the method specifically includes the following steps:

[0025] Step S1: Experimental and simulation data collection. The specific operations of step S1 are:

[0026] Artificial volume defects of different depths and lengths are prefabricated, pulsed eddy current testing experiments are carried out, and experimental data are collected. At the same time, a pulsed eddy current testing finite element model is constructed to collect simulation data, which together constitute a sample data set. The collected data is a pulsed eddy current testing signal.

[0027] Step S2: extracting time-frequency features, constructing time domain feature vector T and frequency domain feature vector F. The specific operation is as follows:

[0028] In order to fully characterize the defect information, the statistical analysis method is used to extract the time domain characteristic parameters such as the mean value, standard deviation, skewness factor, kurtosis factor, maximum value, minimum value, waveform factor, pulse factor, margin factor, energy, etc. of the sample data set, so as to construct the time domain characteristic vector T. The FFT is used to extract the frequency domain characteristic parameters such as AM, CF, MSF, RMSF, FVAR, etc. of the sample data set, so as to construct the frequency domain characteristic vector F.

[0029] Step S3: feature dimensionality reduction based on Spearman correlation analysis, the specific operations are as follows:

[0030] The Spearman correlation analysis method is used to calculate the Spearman correlation coefficient between the characteristic parameters and the defect length and depth, and the characteristic dimension with the highest correlation with the predicted value is selected as the model input feature vector I. The Spearman correlation analysis expression is:

[0031]

[0032] Where r is the Spearman correlation coefficient, n is the total sample size, and is the rank difference between the two variables.

[0033] Step S4, input the reduced eigenvector I into the CHCOA-BP model, complete the model pre-training, and solidify the pre-trained model parameters. The model achieves efficient global optimization of neural network parameters by deeply integrating the improved crayfish optimization algorithm (COA) and the BP neural network.

[0034] The model first uses the mirror reflection learning strategy to initialize the crayfish population, expressed as:

[0035]

[0036] Where: X is the randomly generated initial solution, [a,b] is the search space boundary, Mirror image solution.

[0037] Secondly, the extended search strategy of the Eagle optimization algorithm is introduced in the summer vacation stage of the crayfish optimization algorithm, and the expression is:

[0038]

[0039]

[0040] Where: X 1 (t+1) is the solution of the t+1th iteration, X best (t) is the optimal solution after the tth iteration, t is the current iteration number, T is the maximum iteration number, X M (t) is the average value of the solution at the tth iteration, rand is a random number between 0 and 1, Dim is the problem dimension, and N is the population size.

[0041] Finally, the vertical crossover operation is used to exchange information and perform competitive screening among individuals of the crayfish optimization algorithm in different dimensions, continuously generating new and better solutions. The expression is as follows:

[0042] MS vc (i, d 1 ) = r × X(i, d 1 ) + (1 - r) × X(i, d 2 ), i ∈ N(1, M), d 1 , d 2 ∈ N(1, D)

[0043] In the formula: r is a random number between 0 and 1, M is the population size, D is the dimension, and MS vc (i, d 1 ) is the solution generated by the vertical crossover operation of two different dimensions of individual X(i) in the d 1 th dimension.

[0044] Step S5: Input the detected time-frequency feature quantities into the model. After running, the defect length value and defect depth value are output, thereby realizing the quantitative characterization of the defect.

[0045] The following is the defect sample data set of the above embodiment. Table 1 shows the partial time-domain characteristic parameters of the experimental samples with different length defects, Table 2 shows the frequency-domain characteristic parameters of the experimental samples with different length defects, Table 3 shows the correlation coefficient table between the defect length and the characteristic parameters, and Table 4 shows the comparison of the defect length prediction results of the CHCOA-BP model.

[0046] It can be seen from Table 3 that the five characteristic parameters with the strongest correlation with the defect length are the margin factor, average value, root mean square, energy, and amplitude average value. Among them, the margin factor shows a positive correlation with the defect length, and the rest show negative correlations; the correlations between the standard deviation, peak-to-peak value, maximum value, pulse factor, waveform factor, amplitude factor, skewness factor, and kurtosis factor and the defect length are relatively weak. Among them, the pulse factor, waveform factor, amplitude factor, skewness factor, and kurtosis factor show positive correlations with the defect length, and the standard deviation, peak-to-peak value, and maximum value show negative correlations with the defect length; the correlations between the center frequency, mean square frequency, root mean square frequency, frequency variance, and minimum value and the defect length are the weakest, and except for the minimum value, the rest of the characteristic parameters show negative correlations with the defect length.

[0046] It can be seen from Table 4 that the maximum error of the CHCOA-BP model is 0.31 mm, the minimum error is 0.01 mm, and the average error is 0.14 mm, with a small error range. The quantitative characterization error of the 1.2-mm length defect is only 0.83%, and it has a high regression accuracy in the defect length range above 1 mm, effectively improving the quantitative accuracy of the defect. Table 1 Partial Time-domain Characteristic Parameters of Experimental Samples with Defects of Different Lengths Table 2 Frequency-domain Characteristic Parameters of Experimental Samples with Defects of Different Lengths Table 3 Correlation Coefficient Table of Defect Length and Characteristic Parameters Table 4 Comparison of Defect Length Prediction Results of CHCOA-BP Model

Claims

1. A quantitative characterization method for X80 pipeline steel defects based on BP neural network optimized by improved crayfish optimization algorithm, characterized by: Step S1, experimental and simulation data collection; Step S2, extracting time-frequency features, constructing a time-domain feature vector T and a frequency-domain feature vector F; Step S3: feature dimension reduction based on Spearman correlation analysis; Step S4, the feature vector I after dimension reduction is input into the improved crayfish optimization algorithm optimized BP neural network (CHCOA-BP) algorithm; Step S5: output the defect length value and defect depth value of the X80 pipeline steel volume defect.

2. The X80 pipeline steel defect quantitative characterization method based on the improved crayfish optimization algorithm to optimize the BP neural network according to claim 1, characterized in that The operation of step S1 is specifically as follows: prefabricate artificial volume defects to collect experimental data, build a finite element model to collect simulation data, and together constitute a sample data set, where the collected data is a pulsed eddy current detection signal.

3. The X80 pipeline steel defect quantitative characterization method based on the improved crayfish optimization algorithm to optimize the BP neural network according to claim 1, characterized in that The operation of step S2 is specifically as follows: using statistical analysis methods to extract the mean, standard deviation, skewness factor, kurtosis factor, maximum value, minimum value, waveform factor, pulse factor, margin factor, energy and other time domain characteristic parameters of the sample data set, thereby constructing a time domain characteristic vector T; using FFT to extract the frequency domain characteristic parameters such as AM, CF, MSF, RMSF, FVAR of the sample data set, thereby constructing a frequency domain characteristic vector F.

4. The X80 pipeline steel defect quantitative characterization method based on the improved crayfish optimization algorithm to optimize the BP neural network according to claim 1, characterized in that The operation of step S3 is specifically as follows: using the Spearman correlation analysis method, calculating the Spearman correlation coefficient between the characteristic parameters and the defect length and depth, and screening out the characteristic dimension with the highest correlation with the predicted value as the model input feature vector I.

5. The X80 pipeline steel defect quantitative characterization method based on the improved crayfish optimization algorithm to optimize the BP neural network according to claim 1, characterized in that The quantitative characterization mathematical model established in step S4 uses the improved crayfish optimization algorithm to optimize the BP neural network (CHCOA-BP) algorithm to achieve the prediction of defect length and depth size parameters.

6. The method for quantitative characterization of X80 pipeline steel defects based on the improved crayfish optimization algorithm and BP neural network according to claim 4, characterized in that: Firstly, the mirror reflection learning strategy is used to initialize the crayfish population to increase the diversity of the population and improve the quality of the population, which helps the algorithm to find the global optimal solution faster in the subsequent search process; secondly, the extended search strategy of the Eagle optimization algorithm is introduced in the summer vacation stage of the crayfish optimization algorithm to expand the cave search range, in order to improve the global search ability and convergence speed of the algorithm and avoid falling into the local optimum; finally, the vertical crossover operation is used to exchange information and competitively screen the individuals of the crayfish optimization algorithm in different dimensions, continuously generate new and better solutions, help escape the dilemma of falling into the local optimum, and enhance the global search ability of the crayfish algorithm, so as to intelligently optimize the weights and thresholds of the BP neural network, help the BP neural network find the global optimal point, achieve global convergence and avoid local optimum.

7. The method for quantitative characterization of X80 pipeline steel defects based on the improved crayfish optimization algorithm and BP neural network according to claim 1, characterized in that The shape of the volume defect is groove-shaped.

Images