Bolt corrosion classification method based on acoustic emission sensor system

Through the acoustic emission sensor system and GOOSE-ELM model, the problem of insufficient accuracy of bolt corrosion evaluation in the existing technology is solved, and high-precision bolt corrosion classification is achieved, which improves equipment maintenance and safety.

CN119064467AInactive Publication Date: 2024-12-03NANJING TECH UNIV

Patent Information

Application Number
CN202411563476.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-11-05
Publication Date
2024-12-03
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

The prior art has limitations in evaluating the degree of bolt corrosion, which is difficult to accurately classify, affecting the maintenance and safety of equipment.

Method used

Using the bolt rust classification method based on the acoustic emission sensor system, the acoustic emission signals of bolts are collected through the acoustic emission sensor network, combined with the ReliefF algorithm to screen features, and the improved goose flock algorithm optimized extreme learning machine model is used to classify, and the GOOSE-ELM classification model is constructed. , improve classification accuracy.

Benefits of technology

It realizes high-precision classification of the degree of bolt corrosion, can more accurately reflect the actual rust status of bolts, avoid equipment failures and safety accidents caused by rust, and has strong generalization ability and adaptability.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119064467A_ABST
    Figure CN119064467A_ABST
Patent Text Reader

Abstract

The invention discloses a bolt corrosion classification method based on an acoustic emission sensor system, and relates to the technical field of acoustic emission sensing, and the bolt corrosion classification method comprises the following steps: selecting a bolt test piece, controlling the electrifying duration through an electrochemical corrosion method to prepare bolt samples with different corrosion degrees, and arranging an acoustic emission sensor to carry out bolt corrosion detection; an acoustic emission sensor network is used for detecting the bolt, and acoustic emission signals of the bolt under different corrosion degrees are collected. According to the method, the acoustic emission sensor network is used for collecting the acoustic emission signals of the bolts, the ReliefF algorithm is combined for feature selection, and the improved goose swarm algorithm optimized extreme learning machine model is used for classification, so that the high-precision classification of the bolt corrosion degree is realized, the actual corrosion state of the bolts can be reflected more accurately, and the reliability of the bolts is improved. And a favorable basis is provided for subsequent maintenance and replacement, so that equipment faults and safety accidents caused by bolt corrosion are avoided.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of acoustic emission sensing, and in particular to a bolt corrosion classification method based on an acoustic emission sensor system. Background Art

[0002] High-strength bolts play a vital role in ultra-high-voltage equipment such as bridges and rails. These equipment are often subjected to enormous pressure and stress. Therefore, the state of high-strength bolt connections is directly related to the service life and safety of the entire equipment. With the increase in service life and the influence of the external environment, bolts will inevitably rust. Rust is the main factor affecting thread accuracy, which will lead to difficulty in disassembly and insufficient preload between structural parts, which may cause loosening, loosening or fatigue fracture of the connection structure. This not only brings new difficulties to the normal service and daily maintenance of structural parts, but may even cause major safety accidents. Therefore, timely and accurate detection of the degree of bolt corrosion is crucial to ensure equipment safety.

[0003] For example, the image classification and recognition method for the degree of bolt corrosion on transmission line towers disclosed in Chinese patent publication number CN110334750A detects bolt corrosion faults in images through image processing methods, which can be applied to drone inspections and online monitoring systems. This helps avoid the shortcomings of manual inspections, such as low efficiency, high workload, and slow detection speed. While reducing the input of manpower and material resources, it can eliminate the interference of human factors and quickly and accurately evaluate the bolt corrosion condition. It can extract the rusted area based on the accurate extraction of bolts and the color characteristics of the rust. The severity of the bolt corrosion can be reflected by the rust rate, which is used to guide the maintenance work of operation and maintenance personnel.

[0004] In the prior art, although the rust area is extracted by the color characteristics of the rust and the severity of the bolt corrosion can be judged by the rust rate, the rust degree of the bolt is affected by the coupling of multiple factors. Therefore, there are certain limitations when evaluating the health status of the bolt, and it is difficult to effectively classify the rust degree of the bolt. Therefore, a bolt corrosion classification method based on an acoustic emission sensor system is proposed. Summary of the invention

[0005] The present invention aims to provide a bolt corrosion classification method based on an acoustic emission sensor system to solve the problems raised in the above background technology.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: A bolt corrosion classification method based on an acoustic emission sensor system comprises the following steps: Step 1, selecting bolt specimens and controlling the power-on time by an electrochemical corrosion method to prepare bolt samples with different corrosion degrees, and deploying acoustic emission sensors to perform bolt corrosion detection; Step 2: Use the acoustic emission sensor network (WASN) to detect the bolts, collect the acoustic emission signals of the bolts at different corrosion levels, and pre-process the collected acoustic emission signals, including filtering, denoising, and signal amplification, to improve the signal quality and prepare for subsequent analysis. The acoustic emission signal is the AE signal; Step 3: Use the ReliefF data selection algorithm to select the optimal feature combination from the preprocessed acoustic emission signal to reflect the key information of the bolt corrosion degree and assist in the classification of the corrosion degree; Step 4: Use the extreme learning machine (ELM) model and the selected acoustic emission signal features to classify the bolt corrosion degree. The ELM model is selected for its fast training and good generalization ability. The improved goose flock algorithm (GOOSE) is used to optimize the parameters in the ELM model, build a GOOSE-ELM classification model, find the optimal parameter combination, ensure that the model uses the most appropriate parameter combination, and improve the classification accuracy. Step 5: Apply the optimized GOOSE-ELM classification model to the actual classification of bolt corrosion degree, evaluate the classification results, and compare and evaluate the classification performance of the GOOSE-ELM classification model.

[0007] A further improvement of the technical solution of the present invention is that in step 1, the process of bolt corrosion detection is: Step 101, selecting bolt specimens that are consistent in material, size and structure, the bolt specimens are 10.9 grade high-strength bolts with five holes of 54 mm in diameter; Step 102, conducting an accelerated corrosion test on the bolt by applying electricity, setting an electrochemical corrosion environment, constructing an electrochemical corrosion system including an electrolyte solution, placing the bolt specimen as an anode and another metal as a cathode in the electrolyte solution; Step 103, according to Faraday's law, the relationship between the theoretical mass loss of the bolt and the power-on time is analyzed. The corrosion current density and the power-on time jointly determine the degree of rust of the bolt. The corrosion process of the bolt at different times is simulated by controlling the power-on current and the power-on time. Different power-on times of 0 hours, 6 hours, 12 hours, 18 hours, and 24 hours are set, corresponding to 0%, 25%, 50%, 75%, and 100% rust degrees of the bolt, respectively. The rust degree of the bolt is divided and prepared according to the different power-on times. The rust degree of the bolts after 30 hours of power-on is 100%, and the rust degree of bolts samples with 0%, 25%, 50%, and 75% respectively correspond to the rust degree of 0%, 25%, 50%, and 75% after 0, 6, 12, 18, and 24 hours of power-on. The classification results obtained based on this are consistent with the definition of rust grade in the national standard, that is, the coverage area and adhesion state of the oxide scale on the surface of the specimen are observed by visual method, and the ratio of the rust area to the total surface area of ​​the bolts also conforms to the percentage defined by the rust degree. During the power-on process, the rust condition of the bolts is regularly monitored and recorded to ensure that the rust degree meets expectations; Step 104, sequentially screwing the prepared bolt specimens into a flange with an inner diameter of 5.2 cm, an outer diameter of 24 cm, and a thickness of 4 cm to conduct an experiment. Since the frequency of the acoustic emission signal of a bolt in normal operation is between 0 kHz and 125 kHz, and the frequency of the signal caused by rusting and deformation is within the range of 125 kHz to 150 kHz, a UT1000 wideband acoustic emission sensor is selected and installed on the bolt specimen in close contact to ensure that the acoustic emission signal is effectively captured; Step 105, adding coupling agent to the joint between the broadband acoustic emission sensor and the surface of the bolt specimen, installing the broadband acoustic emission sensor on the upper and lower nuts of the bolt by means of magnetic suction jacket adsorption, with the two probes separated by the flange plate, on the same straight line, with obvious far and near distribution from the flange plate surface, to increase the tightness of the connection, based on the sound source simulation generation method of the wood stress wave measuring instrument, using a handheld rubber hammer to strike the upper nut, using the interaction between the rubber hammer and the upper nut as the sound source, using the striking action to simulate the sound source so that the stress wave generated by the vibration propagates in the specimen structure, propagates to the surface of the bolt specimen through reflection and refraction, and is collected by the broadband acoustic emission sensor.

[0008] A further improvement of the technical solution of the present invention is that the relationship between the theoretical mass loss of the analyzed bolt and the power-on time is expressed as follows: ; in, is the theoretical mass loss of the bolt (g), is the power-on time (s), is the Faraday constant (96485C / mol), is the corrosion current intensity (A), for The relative atomic mass of the alloy is (55.847), the corrosion current density is generally 0.045-3 mA / cm2, the experimental current density is 2 mA / cm2, and the corresponding current should be maintained at about 0.22A after calculation.

[0009] A further improvement of the technical solution of the present invention is that in step 2, the process of collecting the acoustic emission signals of the bolts at different corrosion degrees is as follows: Step 201, start the acoustic emission sensor network, collect acoustic emission signals of bolts at different corrosion levels, ensure that the data acquisition system can record the waveform, frequency, amplitude and other parameters of the acoustic emission signal in real time and accurately, and perform a preliminary signal check before signal preprocessing to confirm the integrity of the signal and the quality of the acquisition; Step 202, preprocessing the collected acoustic emission signal, including filtering, denoising, and signal amplification, etc., applying a filter to remove high-frequency noise and low-frequency interference in the signal, retaining the frequency components related to bolt corrosion, and using denoising technology to reduce random noise in the signal to improve the signal-to-noise ratio of the signal; Step 203, further analyzing the pre-processed acoustic emission signal to extract acoustic emission characteristic parameters, which are amplitude, rise time, duration, number of rings and power. The amplitude is the maximum voltage threshold, in decibels (dB), which is used for wave source type identification; the rise time is the time interval between the acoustic emission signal first exceeding the threshold voltage and reaching the maximum voltage amplitude, which is used for noise identification; the duration is the time difference between the acoustic emission signal first exceeding the threshold voltage and the last exceeding the threshold voltage; the number of rings is the number of times the acoustic emission signal exceeds the threshold voltage; the power is the area under the energy envelope spectrum or the sum of the squares of the sampling values, which is used for wave source type identification; Step 204, the characteristic parameters are summarized to generate a table, and a data sample set for bolt state measurement is constructed. Comparative analysis shows that bolt samples with a high degree of corrosion have denser data and can present more acoustic emission characteristic points.

[0010] A further improvement of the technical solution of the present invention is that in step 3, the process of screening out the optimal feature combination is: Step 301, taking the data sample set of bolt state measurement as input, including bolt acoustic emission signals of different corrosion degrees, and normalizing the bolt acoustic emission signals and converting them into the processing format of the ReliefF algorithm, and performing subsequent analysis on AE signals including: proximal energy / amplitude / duration / ring count, distal energy / amplitude / duration / ring count, proximal / distal energy / amplitude / duration / ring count difference, a total of 12 features, wherein the expression for normalizing the bolt acoustic emission signal is: ; is the original data, and are the maximum and minimum values ​​of the input characteristic data, respectively, and are calculated based on the expression for normalizing the bolt acoustic emission signal. is the normalized result; Step 302, initialize the weights of all features to zero, randomly select a sample point from the data sample set of bolt state measurement as a query point, find its k nearest neighbor samples, including the nearest neighbors of positive examples and negative examples, and calculate the weight update value according to the difference between the query point and the nearest neighbor sample on each feature. For the positive example neighbor, the larger the difference, the greater the weight of the feature; for the negative example neighbor, the larger the difference, the smaller the weight of the feature; Step 303, update the weight of each feature, adjusting it according to the difference with the positive and negative neighbors of the query point; Step 304, repeatedly randomly selecting sample points from the data sample set of bolt state measurement, updating the weights, and performing multiple iterations, randomly selecting new query points in each iteration, until a predetermined number of iterations is reached or all samples are selected as query points; Step 305, after the iteration is completed, each feature obtains a cumulative weight, which reflects the contribution of the feature to the classification task, and then obtains the final feature weight; Step 306, sort the features according to the feature weights, select the top 7 features with the highest weights as the optimal feature combination, and the AE feature weight sorting results are: near / far energy difference, near / far amplitude difference, near-end amplitude, near-end energy, near-end ringing count, near / far duration difference, near-end duration, near / far ringing count difference, far-end ringing count, far amplitude, far-end duration, and far energy.

[0011] A further improvement of the technical solution of the present invention is that in step 4, the process of constructing the GOOSE-ELM classification model is: Step 401, extracting the acoustic emission signal features with the best classification ability through the ReliefF algorithm, and normalizing each feature value in the data set; Step 402, initializing the ELM model, randomly assigning weights and bias values ​​from the input layer to the hidden layer, setting the activation function of the hidden layer, and inputting the screened features into the ELM model; Step 403, calculate the output of the hidden layer according to the input features and the random weights, and use the least squares method to calculate the weight from the hidden layer to the output layer, and use the training data set to train the ELM model to obtain preliminary model parameters; Step 404, using the GOOSE algorithm to optimize the weights and bias items of the ELM model to improve classification performance, and initialize the parameters of the GOOSE algorithm, including the number of geese, the number of iterations, and the search space; Step 405, in each iteration, the weights and bias items of the ELM are updated according to the behaviors of the geese (such as resting, alerting, exploring, etc.), and the optimal combination of weights and bias items is found through continuous iteration, the optimal parameters are locked for output, and the ELM model based on the GOOSE algorithm is established.

[0012] A further improvement of the technical solution of the present invention is that in step 5, the process of comparing and evaluating the classification performance of the GOOSE-ELM classification model is: Step 501, using the trained GOOSE-ELM model to classify and predict the corrosion degree of the test set, and recording the classification result of the GOOSE-ELM model for each test sample; Step 502, using confusion matrix, precision, recall, and F1 score indicators to comprehensively and quantitatively evaluate the classification results, and statistically analyze the classification results, including the number of correctly classified and incorrectly classified samples; Step 503, comparing the performance of the GOOSE-ELM model with the traditional ELM classification algorithm, the HPO-ELM predator optimization classification algorithm, and the SSA-ELM sparrow optimization classification algorithm, and calculating the accuracy, precision, recall rate, and F1 coefficient evaluation indicators of each classification model; Step 504, verify the effectiveness of the ReliefF feature selection algorithm and the superiority of the GOOSE-ELM classification model, sort the feature importance from the two levels of correlation and weight by using the Pearson correlation coefficient method and the ReliefF feature selection method, add the substituted feature dimensions to the original ELM classification model one by one, and the one with the highest accuracy is the respective optimal feature dimension, and the advantages and disadvantages of the two methods are proved by comparing the maximum accuracy and kappa coefficient under the optimal feature dimension; Step 505, analyzing the performance of the GOOSE-ELM model based on the comprehensive quantitative evaluation and comparison results of the classification results, and deploying the evaluated GOOSE-ELM model to the actual bolt corrosion detection system for real-time monitoring and classification.

[0013] A further improvement of the technical solution of the present invention is that the expression of the Pearson correlation coefficient is: ; in, Pearson correlation coefficient is an important tool to measure the strength of the correlation between two variables. Its value is obtained by dividing the covariance between the variables by the product of the standard deviation. The closer it is to 1 or -1, the stronger the linear relationship between the two variables. On the contrary, the closer its value is to 0, the weaker the linear relationship between the two variables. and are two separate characteristic parameters, is the standard deviation, is the corresponding covariance.

[0014] A further improvement of the technical solution of the present invention is that the expression for the comprehensive quantitative evaluation of the classification results is: ; ; ; in, is the total number of categories, For the The number of true positive examples of the class rust, indicating that the model correctly predicts The number of samples of class rust, For the The number of false positives of class rust, indicating that the model incorrectly predicts The number of samples of class rust, For the The recall rate of class rust, also known as the true positive rate or sensitivity, indicates the probability that the model correctly identifies the The proportion of rust-like For the The accuracy of the rust class indicates that the model is The correct proportion of predictions on the rust class, For the The number of false negatives for the class rust, indicating that the model failed to correctly identify Number of rust-like samples, is the total number of samples, that is, the sum of samples of all categories.

[0015] Due to the adoption of the above technical solution, the present invention has the following technical advances compared with the prior art: 1. The present invention provides a bolt corrosion classification method based on an acoustic emission sensor system. The acoustic emission signals of bolts are collected by an acoustic emission sensor network, and feature selection is performed in combination with the ReliefF algorithm. Then, the extreme learning machine model optimized by the improved gush algorithm is applied for classification. The high-precision classification of the degree of bolt corrosion is achieved, and the actual corrosion state of the bolts can be more accurately reflected, providing a strong basis for subsequent maintenance and replacement, thereby avoiding equipment failures and safety accidents caused by bolt corrosion.

[0016] 2. The present invention provides a bolt corrosion classification method based on an acoustic emission sensor system. The improved GOOSE algorithm can improve the classification performance of the ELM model. The combined use makes the bolt corrosion classification method have a strong generalization ability and can adapt to the needs of bolt corrosion detection of different types and specifications. The GOOSE-ELM model optimizes parameters through the improved goose flock algorithm, which improves the adaptability of the model to different working conditions and environmental changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0018] Figure 1 is a flow chart of the method of the present invention; Figure 2 A flow chart of constructing a GOOSE-ELM classification model for the present invention; Figure 3 The flowchart of the present invention for comparing and evaluating the classification performance of the GOOSE-ELM classification model. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention. Embodiment 1, as Figure 1 , Figure 2 As shown, the present invention provides a bolt corrosion classification method based on an acoustic emission sensor system, comprising the following steps: Step 1: Select bolt specimens and prepare bolt samples with different degrees of corrosion by controlling the power-on time through an electrochemical corrosion method, arrange acoustic emission sensors for bolt corrosion detection, select bolt specimens that are consistent in material, size and structure, and the bolt specimens are 10.9 grade high-strength bolts with 5 apertures of 54 mm that meet the national standard NB / T31082-2016. Perform an electrified accelerated corrosion experiment on the bolts, set up an electrochemical corrosion environment, and construct an electrochemical corrosion system containing an electrolyte solution. The bolt specimen is used as an anode and another metal is used as a cathode, and is placed in the electrolyte solution. The specific process is to immerse all the specimens. In NaCl solution, the positive electrode of the external DC power supply is connected to one end of the bolt. At the same time, a conductive cathode material is placed in the solution and connected to the negative electrode of the DC power supply. A positive constant DC is applied between the bolt and the cathode material to accelerate corrosion. Under the action of the external electric field, a half-cell electrochemical oxidation reaction occurs on the surface of the bolt, and a half-cell electrochemical reduction reaction occurs on the surface of the anode material. According to Faraday's law, the relationship between the theoretical mass loss of the bolt and the power-on time is analyzed. The corrosion current density and the power-on time jointly determine the degree of corrosion of the bolt. The relationship between the theoretical mass loss of the bolt and the power-on time is analyzed, and its expression is: ; in, is the theoretical mass loss of the bolt (g), is the power-on time (s), is the Faraday constant (96485C / mol), is the corrosion current intensity (A), for The relative atomic mass is (55.847), the corrosion current density is generally 0.045~3mA / cm2, the experimental current density is 2mA / cm2, and the corresponding power-on current should be maintained at about 0.22A after calculation. The corrosion process of the bolts at different times is simulated by controlling the power-on current and the power-on time. Different power-on times of 0 hours, 6 hours, 12 hours, 18 hours, and 24 hours are set, corresponding to 0%, 25%, 50%, 75%, and 100% rust degrees of the bolts, respectively. The rust degree of the bolts is divided and prepared according to the different power-on times. The rust degree of the bolts is 100% when the power-on time is 30 hours, and the power-on time of 0, 6, 12, 18, and 24 hours corresponds to 0%, 25%, 50%, 75%, and 100% rust degrees, respectively. The classification results of bolt samples with a percentage of 25%, 50%, and 75% are consistent with the definition of rust grade in the national standard, that is, the coverage area and adhesion state of the oxide scale on the surface of the specimen are observed by visual method, and the ratio of the rust area to the total surface area of ​​the bolt also meets the percentage defined by the rust degree. During the power-on process, the rust condition of the bolt is regularly monitored and recorded to ensure that the rust degree meets expectations. The prepared bolt specimens are screwed into the flange with an inner diameter of 5.2 cm, an outer diameter of 24 cm, and a thickness of 4 cm for experiments. Since the frequency of the acoustic emission signal of a normally operating bolt is between 0KHZ and 125KHZ, and the frequency of the signal caused by deformation due to rust is between 125KHZ and 150KHZ, UT1000 is selected. A wide-band acoustic emission sensor is installed on the bolt specimen in close contact to ensure that the acoustic emission signal is effectively captured. Coupling agent is added to the joint between the wide-band acoustic emission sensor and the surface of the bolt specimen. The wide-band acoustic emission sensor is installed on the upper and lower nuts of the bolt by magnetic suction jacket adsorption. The two probes are separated by the flange and are on the same straight line. They are obviously distributed at the far and near ends from the flange surface to increase the tightness of the connection. Based on the sound source simulation generation method of the wood stress wave measuring instrument, a rubber hammer is used to strike the upper nut. The interaction between the rubber hammer and the upper nut is used as the sound source. The knocking action is used to simulate the sound source to make the stress wave generated by the vibration propagate in the specimen structure and propagate to the bolt through reflection and refraction. The surface of the specimen is collected by the wide-band acoustic emission sensor; when the bolt and flange specimens are subjected to dynamic excitation or collision friction, the specimen will generate elastic waves that propagate in the structure, and after reflection and refraction, reach the surface of the connected specimen and are detected and received by the acoustic emission sensor. In the experiment, the sensor needs to be in close contact with the surface of the object. In order to avoid signal loss caused by sensor selection and layout, the center of the proximal probe is 1.5 cm away from the lower flange surface, and the distal probe is 5.5 cm away from the upper flange surface. The AE signals received by the two probes can be differentially calculated to avoid interference from other factors as much as possible, so as to give full play to the advantages of the acoustic emission sensor, reduce system measurement errors, and collect purer signals. Step 2: Use the acoustic emission sensor network (WASN) to detect the bolts, collect the acoustic emission signals of the bolts at different corrosion levels, and pre-process the collected acoustic emission signals, including filtering, denoising, and signal amplification, to improve the signal quality and prepare for subsequent analysis. The acoustic emission signal is the AE signal. The acoustic emission sensor network is powered by a battery consisting of 6 nickel-hydrogen batteries with a rated voltage of 1.2V connected in series. The UT 1000 acoustic emission sensor is used to collect the bolt AE data from two independent channels. The OPA627 The operational amplifier is used for amplification, and then the ultra-low power analog-to-digital converter AD7356 is used to convert the analog quantity of the AE signal into a digital quantity and output it to the STM32F405RG chip to read the acoustic emission signal. After data preprocessing and data storage, the LoRa module based on SX1278 is used to build a low-power, long-distance, and low-cost wireless communication network. In the 5V to 3.3V and 2.5V modules of the power supply system, the ADM7170 voltage regulator chip is selected to enable the regulator to maintain a low noise level in the output voltage; the OPA627 chip is used to amplify the original acoustic emission signal, enhance the noise suppression capability of the circuit, and improve the signal-to-noise ratio of the acoustic emission signal; Frequency-shift is used in the wireless communication module. The keying modulation method also has good anti-noise performance. A threshold voltage is selected to filter noise. When the AE signal strength exceeds the threshold voltage, it is identified as an acoustic emission signal, otherwise it is noise. The acoustic emission sensor network is started to collect acoustic emission signals of bolts at different degrees of rust, ensuring that the data acquisition system can record the waveform, frequency, amplitude and other parameters of the acoustic emission signal in real time and accurately. Before signal preprocessing, a preliminary signal check is performed to confirm the integrity of the signal and the quality of the acquisition. The collected acoustic emission signals are preprocessed, including filtering, denoising and signal amplification. Filters are applied to remove high-frequency noise and low-frequency interference in the signal, retaining the frequency components related to bolt corrosion, and using denoising technology to reduce random noise in the signal to improve the signal-to-noise ratio of the signal. The preprocessed acoustic emission signal The signal was further analyzed to extract the acoustic emission characteristic parameters, namely amplitude, rise time, duration, number of rings and power. The amplitude is the maximum voltage threshold, in decibels (dB), which is used to identify the wave source type; the rise time is the time interval between the acoustic emission signal exceeding the threshold voltage for the first time and reaching the maximum voltage amplitude, which is used for noise identification; the duration is the time difference between the acoustic emission signal exceeding the threshold voltage for the first time and the last time it exceeds the threshold voltage; the number of rings is the number of times the acoustic emission signal exceeds the threshold voltage; the power is the area under the energy envelope spectrum or the sum of the squares of the sampling values, which is used to identify the wave source type. The characteristic parameters are summarized to generate a table to construct a data sample set for bolt state measurement. Comparative analysis shows that the bolt samples with a high degree of corrosion have denser data and can present more acoustic emission characteristic points. Step 3, use the ReliefF data selection algorithm to select the optimal feature combination from the preprocessed acoustic emission signal, reflect the key information of the bolt corrosion degree, assist in the classification of the corrosion degree, take the data sample set of the bolt state measurement as input, including the bolt acoustic emission signals of different corrosion degrees, and normalize the bolt acoustic emission signals and convert them into the processing format of the ReliefF algorithm. The AE signal includes: proximal energy / amplitude / duration / ringing count, distal energy / amplitude / duration / ringing count, proximal / distal energy / amplitude / duration / ringing count difference, a total of 12 features are analyzed later, among which the expression for normalizing the bolt acoustic emission signal is: ; is the original data, and are the maximum and minimum values ​​of the input characteristic data, respectively, and are calculated based on the expression for normalizing the bolt acoustic emission signal. To normalize the results, all features are initialized to zero weights. A sample point is randomly selected from the data sample set of the bolt state measurement as the query point. For the query point, its k nearest neighbor samples are found, including the nearest neighbors of the positive and negative examples. The weight update value is calculated based on the difference between the query point and the nearest neighbor sample on each feature. For the positive example neighbors, the larger the difference, the greater the weight of the feature; for the negative example neighbors, the larger the difference, the smaller the weight of the feature. The weight of each feature is updated and adjusted according to the difference with the positive and negative example neighbors of the query point. Sample points are randomly selected from the data sample set of the bolt state measurement repeatedly, and the weight is updated. Multiple iterations are performed, and new query points are randomly selected in each iteration until the predetermined number of iterations is reached or all samples are selected as query points. After the iteration is completed, each feature obtains a cumulative weight, which reflects the contribution of the feature to the classification task, and then the final feature weight is obtained. The features are sorted according to the feature weights, and the top 7 features with the highest weights are selected as the optimal feature combination. The AE feature quantity weight sorting results are: near / far energy difference, near / far amplitude difference, near amplitude, near energy The 12 features are as follows: near-end energy / amplitude / duration / ringing count, far-end energy / amplitude / duration / ringing count, near-end energy / amplitude / duration / ringing count, near-end energy / amplitude / duration / ringing count difference, near-end energy / amplitude / duration / ringing count difference. The results are summarized and the average value is obtained for analysis. The near-end energy difference is the main factor that best reflects the characteristics of the acoustic emission signal under the state of bolt corrosion, which will directly reflect the signal changes of different degrees. The second is the near / far amplitude difference. The weights of the third and fourth features are relatively close, indicating that the near amplitude and near energy are both important criteria for judging the degree of corrosion. In actual operation, the number of optimal feature sets is gradually increased from 1, and the resulting combinations are successively brought into the original ELM extreme learning machine classification model with default parameters to obtain the approximate classification result accuracy. The final analysis shows that when the top 7 feature combinations are selected, the classification accuracy reaches a maximum value of 93.4%, and then the accuracy begins to decline. Therefore, all operations will be classified based on the selected 7 features. Step 4: Use the extreme learning machine (ELM) model and the selected acoustic emission signal features to classify the degree of bolt corrosion. The ELM model is selected for its fast training and good generalization ability. The improved goose flock algorithm (GOOSE) is used to optimize the parameters in the ELM model, build the GOOSE-ELM classification model, find the optimal parameter combination, ensure that the model uses the most appropriate parameter combination, improve the classification accuracy, extract the acoustic emission signal features with the most classification ability selected by the ReliefF algorithm, and normalize each feature value in the data set. Initialize the ELM model, randomly assign weights and bias values ​​from the input layer to the hidden layer, and set the activation function of the hidden layer. Input the selected features into the ELM model, calculate the output of the hidden layer according to the input features and random weights, and use the least squares method to calculate the weights from the hidden layer to the output layer. Use the training data set to train the ELM model to obtain preliminary model parameters. Use the GOOSE algorithm to optimize the weights and bias items of the ELM model to improve the classification performance, and initialize the parameters of the GOOSE algorithm, including the number of geese, the number of iterations, and the search space. In each iteration, update the weights and bias items of the ELM according to the behavior of the geese (such as resting, alerting, exploring, etc.). Through continuous iteration, find the optimal combination of weights and bias items, lock the optimal parameters for output, and establish an ELM model based on the GOOSE algorithm. Furthermore, the GOOSE algorithm is initialized overall, the X matrix is ​​established, and the position of the goose is X. After the population is initialized, the search agents that are out of the search space are returned. In each iteration, the fitness of each search agent is determined by a standardized benchmark function and compared with the fitness of the remaining agents to obtain the best fitness and the best position. The fitness of each current row is compared, and during the iteration, the fitness of the previous row is returned; By using the conditions and random variables, a variable rnd is determined, whose value is evenly distributed according to the number of iterations, which is used to balance the exploration and development stages. In the case of the specified random variables, assuming rnd=0.5, it means that the probability of development or exploration is 50%. In order, the iteration uses the conditions to be evenly distributed between exploration and development. In addition, the pro variable and the coe variable need to be introduced to find out the effective equation and whether the discrimination value is less than or equal to 0.17 respectively; For the development stage, assume that the weight of the stones stored on the goose ranges from 5 grams to 25 grams. The weight of the stones in any iteration is randomly generated by the following formula: ; Suppose the time required for the stone to reach the ground is , with random changes between dimensions at each iteration in the loop: ; The time it takes for this sound to be transmitted to any individual goose in the flock is: ; Next, find the average time it takes for the sound to propagate and reach the individual geese during the entire iteration: ; Considering the parameter restrictions, the value of pro must be greater than 0.2 and Should be greater than or equal to 12, is the speed of the falling object, and 9.81 is the acceleration of gravity. Therefore, in order to ensure that the geese can be protected and awakened, the following formula should be used: ; The distance that sound travels is the speed of sound in air With propagation time Next, find the distance between the guard goose and the other goose that is resting or eating. Note that you need to multiply by 1 / 2 because you only need the time for the sound to propagate outward, not the time it takes for the sound to return: ; Set an optimal , which is equal to the speed of the falling object Plus the distance between the two geese Multiplying the square of time: ; If the parameters pro and If the value of does not meet the above conditions, then find a new ,at this time will be redefined as: ; For the exploration phase, if one of the geese immediately makes a sound to protect the other individuals upon waking up, then it enters the exploration phase, that is, when rnd is less than 0.5, the following equation applies: ; in is the number of problem dimensions, is the best position found in the search area .variable It takes values ​​between 0 and 2, and decreases significantly with the number of loop iterations; ; In the algorithm, the formula is used to continuously find the new position of the goose group in the search space. Depending on the values ​​of different parameters, the optimal parameters are finally locked for output through continuous updating and iteration, and the ELM model based on the GOOSE algorithm is established; Step 5: Apply the optimized GOOSE-ELM classification model to the actual classification of bolt corrosion degree, evaluate the classification results, and compare and evaluate the classification performance of the GOOSE-ELM classification model.

[0020] Embodiment 2, as Figure 3 As shown, based on Example 1, the present invention provides a technical solution: Preferably, in step 5, the process of comparing and evaluating the classification performance of the GOOSE-ELM classification model is: The trained GOOSE-ELM model was used to classify and predict the degree of rust on the test set, and the classification results of the GOOSE-ELM model for each test sample were recorded. The confusion matrix, accuracy, recall, and F1 score indicators were used to conduct a comprehensive quantitative evaluation of the classification results, and the classification results were statistically analyzed, including the number of correctly classified and incorrectly classified samples. The performance of the GOOSE-ELM model was compared with the traditional ELM classification algorithm, the HPO-ELM predator optimization classification algorithm, and the SSA-ELM sparrow optimization classification algorithm. The accuracy, precision, recall, and F1 coefficient evaluation indicators of each classification model were calculated, and the ReliefF feature was verified. The effectiveness of the selection algorithm and the superiority of the GOOSE-ELM classification model were evaluated. The importance of features was ranked from the perspectives of correlation and weight using the Pearson correlation coefficient method and the ReliefF feature selection method. The feature dimensions were added one by one to the original ELM classification model. The best feature dimension was the one with the highest accuracy. The advantages and disadvantages of the two methods were demonstrated by comparing the maximum accuracy and kappa coefficient under the best feature dimension. The performance of the GOOSE-ELM model was analyzed based on the comprehensive quantitative evaluation of the classification results and the comparison results. The evaluated GOOSE-ELM model was deployed in the actual bolt corrosion detection system for real-time monitoring and classification. Furthermore, the expression of Pearson correlation coefficient is: ; in, Pearson correlation coefficient is an important tool to measure the strength of the correlation between two variables. Its value is obtained by dividing the covariance between the variables by the product of the standard deviation. The closer it is to 1 or -1, the stronger the linear relationship between the two variables. On the contrary, the closer its value is to 0, the weaker the linear relationship between the two variables. and are two separate characteristic parameters, is the standard deviation, is the corresponding covariance. Analysis shows that there is a strong correlation between the near / far energy difference and the near / far amplitude difference and the near / far duration difference. Therefore, the features are sorted according to the degree of correlation with other features. The features with the highest correlation are added one by one and substituted into the ELM classification model. The accuracy and kappa coefficient of this method are calculated and compared with the ReliefF algorithm. The Kappa coefficient is an indicator used for consistency testing and measuring classification accuracy. Its value range is between -1 and 1. It indicates the proportion of classification error reduced by the algorithm's classification results compared with completely random classification results. The larger the value, the better the sorting effect. It not only considers the accuracy of the model prediction, but also compares it with random guessing, so it has good interpretability. By comparison, it can be found that the accuracy and kappa coefficient of the ReliefF algorithm are higher than the Pearson correlation coefficient method, which proves its superiority. Furthermore, the expression for the comprehensive quantitative evaluation of the classification results is: ; ; ; in, is the total number of categories, For the The number of true positive examples of the class rust, indicating that the model correctly predicts The number of samples of class rust, For the The number of false positives of class rust, indicating that the model incorrectly predicts The number of samples of class rust, For the The recall rate of class rust, also known as the true positive rate or sensitivity, indicates the probability that the model correctly identifies the The proportion of rust-like For the The accuracy of the rust class indicates that the model is The correct proportion of predictions on the rust class, For the The number of false negatives for the class rust, indicating that the model failed to correctly identify Number of rust-like samples, is the total number of samples, that is, the sum of samples of all categories. It should be noted that by calculating the precision and recall rate of each category, for each category , multiply the precision and recall by the ratio of true positives to false positives to calculate the weighted performance index, then divide the total number of samples by the total number of categories, calculate the negative power of the natural exponent e, and get the exponential decay term. For all categories, multiply the weighted performance index by the exponential decay term and then sum them. When the total number of categories n tends to infinity, calculate The limit value of .

[0021] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A bolt corrosion classification method based on an acoustic emission sensor system, characterized in that: The following steps are involved: Step 1, selecting bolt specimens and controlling the power-on time by an electrochemical corrosion method to prepare bolt samples with different corrosion degrees, and deploying acoustic emission sensors to perform bolt corrosion detection; Step 2: Use an acoustic emission sensor network to detect the bolts, collect acoustic emission signals of the bolts at different corrosion degrees, and pre-process the collected acoustic emission signals, where the acoustic emission signals are AE signals; Step 3: Use the ReliefF data selection algorithm to select the optimal feature combination from the preprocessed acoustic emission signal to reflect the key information of the bolt corrosion degree and assist in the classification of the corrosion degree; Step 4: Use the ELM model and the selected acoustic emission signal features to classify the bolt corrosion degree, and apply the improved goose flock algorithm to optimize the parameters in the ELM model, build the GOOSE-ELM classification model, and find the optimal parameter combination; Step 5: Apply the optimized GOOSE-ELM classification model to the actual classification of bolt corrosion degree, evaluate the classification results, and compare and evaluate the classification performance of the GOOSE-ELM classification model.

2. The bolt corrosion classification method based on the acoustic emission sensor system according to claim 1 is characterized in that: In step 1, the process of bolt corrosion detection is as follows: Step 101, selecting a bolt specimen having the same material, size and structure; Step 102, conducting an accelerated corrosion test on the bolt by applying electricity, setting an electrochemical corrosion environment, constructing an electrochemical corrosion system including an electrolyte solution, placing the bolt specimen as an anode and another metal as a cathode in the electrolyte solution; Step 103, according to Faraday's law, the relationship between the theoretical mass loss of the bolt and the power-on time is analyzed, and the corrosion process of the bolt at different times is simulated by controlling the power-on current and the power-on time. Different power-on times of 0 hours, 6 hours, 12 hours, 18 hours, and 24 hours are set to correspond to 0%, 25%, 50%, 75%, and 100% corrosion degrees of the bolt, respectively. During the power-on process, the corrosion condition of the bolt is regularly monitored and recorded; Step 104, sequentially screw the prepared bolt specimens into a flange with an inner diameter of 5.2 cm, an outer diameter of 24 cm, and a thickness of 4 cm to conduct an experiment, select a UT1000 wide-band acoustic emission sensor, and install it on the bolt specimen in close contact; Step 105, adding coupling agent to the joint between the broadband acoustic emission sensor and the surface of the bolt specimen, installing the broadband acoustic emission sensor on the upper and lower nuts of the bolt by means of magnetic suction jacket adsorption, with the two probes separated by the flange and on the same straight line, with obvious far and near distribution from the flange surface, using the knocking action to simulate the sound source so that the stress wave generated by the vibration propagates in the specimen structure, propagates to the surface of the bolt specimen through reflection and refraction, and is collected by the broadband acoustic emission sensor.

3. The bolt corrosion classification method based on the acoustic emission sensor system according to claim 2 is characterized in that: The relationship between the theoretical mass loss of the analyzed bolt and the energization time is expressed as follows: ; in, is the theoretical mass loss of the bolt, is the power-on time, is the Faraday constant, is the corrosion current intensity, for The relative atomic mass.

4. The bolt corrosion classification method based on the acoustic emission sensor system according to claim 3 is characterized in that: In step 2, the process of collecting acoustic emission signals of bolts at different corrosion degrees is as follows: Step 201, start the acoustic emission sensor network, collect acoustic emission signals of bolts at different corrosion levels, and perform a preliminary signal check before signal preprocessing; Step 202, preprocessing the collected acoustic emission signals; Step 203, further analyzing the pre-processed acoustic emission signal to extract acoustic emission characteristic parameters, which are amplitude, rise time, duration, number of ringings and power; Step 204, various characteristic parameters are summarized to generate a table, and a data sample set for bolt state measurement is constructed.

5. The bolt corrosion classification method based on the acoustic emission sensor system according to claim 4 is characterized in that: In step 3, the process of selecting the optimal feature combination is as follows: Step 301, taking the data sample set of bolt state measurement as input, including bolt acoustic emission signals of different corrosion degrees, and normalizing the bolt acoustic emission signals and converting them into the processing format of the ReliefF algorithm, and performing subsequent analysis on AE signals including: proximal energy / amplitude / duration / ring count, distal energy / amplitude / duration / ring count, proximal / distal energy / amplitude / duration / ring count difference, a total of 12 features, wherein the expression for normalizing the bolt acoustic emission signal is: ; is the original data, and are the maximum and minimum values ​​of the input characteristic data, respectively, and are calculated based on the expression for normalizing the bolt acoustic emission signal. is the normalized result; Step 302, initialize the weights of all features to zero, randomly select a sample point from the data sample set of bolt state measurement as a query point, find its k nearest neighbor samples, including the nearest neighbors of positive and negative examples, and calculate the weight update value according to the difference between the query point and the nearest neighbor sample on each feature; Step 303, update the weight of each feature, adjusting it according to the difference with the positive and negative neighbors of the query point; Step 304, repeatedly randomly selecting sample points from the data sample set of bolt state measurement, updating the weights, and performing multiple iterations, randomly selecting new query points in each iteration, until a predetermined number of iterations is reached or all samples are selected as query points; Step 305, after the iteration is completed, each feature obtains a cumulative weight to obtain the final feature weight; Step 306, sort the features according to the feature weights, select the top 7 features with the highest weights as the optimal feature combination, and the AE feature weight sorting results are: near / far energy difference, near / far amplitude difference, near-end amplitude, near-end energy, near-end ringing count, near / far duration difference, near-end duration, near / far ringing count difference, far-end ringing count, far amplitude, far-end duration, and far energy.

6. The bolt corrosion classification method based on the acoustic emission sensor system according to claim 5 is characterized in that: In step 4, the process of constructing the GOOSE-ELM classification model is: Step 401, extracting the acoustic emission signal features with the best classification ability through the ReliefF algorithm, and normalizing each feature value in the data set; Step 402, initializing the ELM model, randomly assigning weights and bias values ​​from the input layer to the hidden layer, setting the activation function of the hidden layer, and inputting the screened features into the ELM model; Step 403, calculate the output of the hidden layer according to the input features and the random weights, and use the least squares method to calculate the weight from the hidden layer to the output layer, and use the training data set to train the ELM model to obtain preliminary model parameters; Step 404, using the GOOSE algorithm to optimize the weights and bias items of the ELM model to improve classification performance, and initialize the parameters of the GOOSE algorithm, including the number of geese, the number of iterations, and the search space; Step 405, in each iteration, the weights and bias items of the ELM are updated according to the behavior of the goose group, and the optimal combination of weights and bias items is found through continuous iteration, the optimal parameters are locked for output, and the ELM model based on the GOOSE algorithm is established.

7. The bolt corrosion classification method based on the acoustic emission sensor system according to claim 6 is characterized in that: In step 5, the process of comparing and evaluating the classification performance of the GOOSE-ELM classification model is as follows: Step 501, using the trained GOOSE-ELM model to classify and predict the corrosion degree of the test set, and recording the classification result of the GOOSE-ELM model for each test sample; Step 502, using confusion matrix, precision, recall, and F1 score indicators to comprehensively and quantitatively evaluate the classification results, and statistically analyze the classification results, including the number of correctly classified and incorrectly classified samples; Step 503, comparing the performance of the GOOSE-ELM model with the traditional ELM classification algorithm, the HPO-ELM predator optimization classification algorithm, and the SSA-ELM sparrow optimization classification algorithm, and calculating the accuracy, precision, recall rate, and F1 coefficient evaluation indicators of each classification model; Step 504, verifying the effectiveness of the ReliefF feature selection algorithm and the superiority of the GOOSE-ELM classification model, and ranking the importance of features from the perspectives of relevance and weight respectively by using the Pearson correlation coefficient method and the ReliefF feature selection method; Step 505, analyzing the performance of the GOOSE-ELM model based on the comprehensive quantitative evaluation and comparison results of the classification results, and deploying the evaluated GOOSE-ELM model to the actual bolt corrosion detection system for real-time monitoring and classification.

8. The bolt corrosion classification method based on the acoustic emission sensor system according to claim 7 is characterized in that: The expression of the Pearson correlation coefficient is: ; in, Pearson correlation coefficient is an important tool to measure the strength of the correlation between two variables. Its value is obtained by dividing the covariance between variables by the product of the standard deviation. and are two separate characteristic parameters, is the standard deviation, is the corresponding covariance.

9. The bolt corrosion classification method based on the acoustic emission sensor system according to claim 8 is characterized in that: The expression for the comprehensive quantitative evaluation of the classification results is: ; ; ; in, is the total number of categories, For the The number of true positive examples of the class rust, indicating that the model correctly predicts The number of samples of class rust, For the The number of false positives of class rust, indicating that the model incorrectly predicts The number of samples of class rust, For the The recall rate of the class rust indicates that the model correctly identifies the The proportion of rust-like For the The accuracy of the rust class indicates that the model is The correct proportion of predictions on the rust class, For the The number of false negatives for the class rust, indicating that the model failed to correctly identify Number of rust-like samples, is the total number of samples, that is, the sum of samples of all categories.

Citation Information

Patent Citations

  • Rusted beam model correction and damage identification method

    CN113283144A

  • Coronary artery stenosis lesion degree identification method based on multi-classifier fusion

    CN113408603A

  • Damage identification method of flexible material based on acoustic emission technology

    CN114487129A

  • Neighborhood mutual information and random forest fusion fault detection method, system and application

    CN115828140A

  • Valve inner leakage grade prediction method based on acoustic emission signals

    CN116842324A

Cited By

  • Intelligent hole quality detection method and system based on acoustic emission signal and medium

    CN119848642A