Method, device and equipment for ultrasonic guided wave detection of weld defects based on machine learning

Abstract

The present invention discloses a method, device and equipment for detecting weld defects by ultrasonic guided wave based on machine learning, which relates to the technical field of nondestructive testing. The method obtains characteristic reflection signals for a large number of weld defect samples with known defects through ultrasonic guided wave technology, extracts the main component and the prominent component features by using the P‑PCA algorithm, trains with the known defects, and builds a defect identification model library in combination with the K nearest neighbor model; during the detection, the defect features of the weld to be tested are identified and extracted, and the K nearest neighbor model is used to perform fuzzy matching with the model library to determine the defect classification and calculate the specific defect location; in addition, supervised learning and SVM regression are used to derive samples from the model library based on the difference feature points between the detection results and the model library. The present application can not only realize the accurate and rapid identification, extraction, classification and positioning of weld defect signals, but also reduce the dependence on operators and improve the reliability and accuracy of nondestructive detection of weld defects.

Description

Technical Field

[0001] The present application relates to the field of nondestructive testing technology, and specifically to a method, device and equipment for detecting weld defects using ultrasonic guided waves based on machine learning. Background Art

[0002] In engineering structures such as bridges, ships, and building construction, the quality of welds is directly related to the stability and safety of the entire engineering structure. Although traditional weld defect detection methods such as visual inspection, magnetic particle testing, and ultrasonic testing have played an important role in the past, they often face challenges such as insufficient detection accuracy, complex operation, and low efficiency when faced with complex weld structures. Especially in large and complex structures, these traditional methods have difficulty in accurately identifying the presence and type of internal defects in welds.

[0003] In recent years, ultrasonic guided wave nondestructive testing technology has shown great potential in the field of weld defect detection due to its advantages of long-distance and rapid detection. When sound waves are transmitted in the waveguide medium, the boundary of the contact medium is constantly reflected, refracted, and converted into longitudinal waves and transverse waves to generate special sound waves. The ultrasonic waves propagating in the waveguide medium are ultrasonic guided waves. The ultrasonic guided wave signals are transmitted through the solid surface or the internal medium of the main body to achieve long-distance monitoring of the weld. The emergence of this technology has brought new possibilities to the field of weld detection, especially in industrial fields such as metal pipelines, bridges and building structures, and its application value cannot be ignored.

[0004] However, ultrasonic guided wave nondestructive testing technology also faces some challenges. First, it involves knowledge of multiple disciplines such as acoustics and materials science, and requires high professional skills of operators. Second, the signals generated by ultrasonic guided wave testing are relatively complex, and professional knowledge and experience are required to accurately interpret the test results. These challenges limit the widespread application of ultrasonic guided wave nondestructive testing technology. Summary of the invention

[0005] The present application provides a method, device and equipment for detecting weld defects by ultrasonic guided waves based on machine learning, which can achieve accurate detection and classification of weld defects, while reducing personnel dependence and significantly improving the accuracy and reliability of weld defect detection.

[0006] In a first aspect, an embodiment of the present application provides a method for detecting weld defects using ultrasonic guided waves based on machine learning, and the method for detecting weld defects using ultrasonic guided waves based on machine learning includes:

[0007] Based on ultrasonic guided wave technology, the reflection signal of the weld defect characteristics of the weld defect sample with known defects is obtained, and an asymptotic approximation model is established. The P-PCA algorithm is used to extract the main component and prominent component features and perform pattern recognition on the reflection signal of the obtained weld defect characteristics.

[0008] According to the characteristics and patterns of the extracted weld defect samples, they are compared with known defects and trained in combination with the K nearest neighbor model to establish a defect identification model library for storing weld defect characteristics;

[0009] The ultrasonic guided wave signal is emitted to the weld to be inspected and the reflected signal data is captured. The P-PCA algorithm is used to extract the weld defect characteristic signal of the captured reflected signal, and the extracted data is input into the trained K nearest neighbor model.

[0010] The K nearest neighbor model is used to identify the characteristic signal data of the weld defect of the reflection signal, and the defect identification model library is combined for comparative analysis to determine the detection results. The specific information of the weld defect in the detection results is determined based on the distance method.

[0011] Based on the difference feature points between the test results and the defect identification model library, supervised learning is adopted to use the different test results as new sample data, and the defect identification model library is trained through supervised learning based on the feature information of the sample data, and the feature information of the newly derived samples is classified and identified;

[0012] Through support vector machine regression and training with known defects, nonlinear mapping is performed on the weld defect features of the weld defect samples obtained through supervised learning. The identification information of the weld defect samples is improved according to the format of the defect identification model library, the weld defect features are converted into a form that is easy to classify and identify, and samples are derived from the defect identification model library.

[0013] Among them, for the selection and quantity setting of the parameters of the main component and the salient component feature points, the maximum mean difference algorithm is used as the measurement indicator for selection.

[0014] In combination with the first aspect, in an embodiment,

[0015] The P-PCA algorithm is used to extract the global principal component features and local key salient component features of the guided wave detection signal, so as to extract the defect features from each local area or segment to realize the extraction of useful features of the detection signal;

[0016] The extraction function of the global principal component feature is specifically:

[0017]

[0018] in, represents the spectrum amplitude difference, represents the amplitude adjustment coefficient, Indicates the reference value or standard value of the spectrum amplitude difference when there is no defect. Indicates the detection value of the spectrum amplitude difference when there is a defect, Indicates the angular frequency at which the waveguide detection signal ends, Indicates the angular frequency at which the waveguide detection signal starts;

[0019] The extraction function of the local key salient component feature is specifically:

[0020]

[0021] in, Indicates the degree of spectrum amplitude deviation, represents the deviation adjustment coefficient, Indicates the reference value or standard value of the spectrum amplitude deviation when there is no defect. Indicates the detection value of spectrum amplitude deviation when there is a defect, Indicates the time when the waveguide detection signal ends. Indicates the time when the waveguide detection signal starts.

[0022] In combination with the first aspect, in an embodiment, the ultrasonic guided wave technology is used to obtain the reflection signal of the weld defect characteristics of the weld defect sample with known defects, an asymptotic approximation model is established, and the P-PCA algorithm is used to extract the main component and prominent component features of the obtained reflection signal of the weld defect characteristics and perform pattern recognition, specifically including:

[0023] For weld defect samples with known defects, the reflection signal of the weld defect characteristic signal of the weld defect sample is obtained based on ultrasonic guided wave technology, and an asymptotic approximation model is established;

[0024] Based on the principal component analysis model and the P-PCA algorithm, the main component and prominent component features of the acquired weld defect characteristic signals are extracted and the pattern is recognized.

[0025] In combination with the first aspect, in an embodiment, the training in combination with the K nearest neighbor model specifically includes:

[0026] Create a K-nearest neighbor model and use the extracted features to train the K-nearest neighbor model;

[0027] The cross entropy is used as the loss function and compared with known defects to optimize the training process of the K nearest neighbor model.

[0028] In combination with the first aspect, in an embodiment, the ultrasonic guided wave signal is emitted to the weld to be inspected and the reflected signal data is captured, the weld defect characteristic signal of the captured reflected signal is extracted using the P-PCA algorithm, and the extracted data is input into the trained K nearest neighbor model, specifically including:

[0029] The ultrasonic guided wave sensor is placed near the weld to be inspected, and the ultrasonic guided wave signal is emitted to the weld to be inspected through the ultrasonic guided wave generating device;

[0030] The reflected signal data is captured by an ultrasonic guided wave sensor, and the P-PCA algorithm is used to extract the weld defect characteristic signal of the captured reflected signal. The extracted data is input into the trained K nearest neighbor model.

[0031] In combination with the first aspect, in an embodiment, the specific information of the weld defect in the detection result is determined based on the distance method, specifically including:

[0032] The confusion matrix rule is used to sort the detection results;

[0033] The distances between different weld defect categories are calculated based on the distance method to determine the specific information of the weld defects in the detection results.

[0034] In combination with the first aspect, in an embodiment, the difference feature points between the detection results and the defect identification model library are supervised by learning, the detection results with differences are used as new sample data, and the defect identification model library is supervised by learning and training in combination with the feature information of the sample data, and the feature information of the newly derived samples is classified and identified, wherein the identification of the difference feature points specifically includes:

[0035] Calculate the similarity or difference between the detection results and the characteristics of various types of weld defects in the defect identification model library, and determine whether there is a difference by combining the similarity or difference calculation results and referring to the preset threshold.

[0036] In combination with the first aspect, in an embodiment, the support vector machine regression is used for comparison training with known defects, and nonlinear mapping is performed on the weld defect features of the weld defect samples obtained by supervised learning, and the identification information of the weld defect samples is improved according to the format of the defect identification model library, and the weld defect features are converted into a form that is easy to classify and identify, and the defect identification model library is sample derived, wherein the comparison training specifically includes:

[0037] The acquired weld defect features are matched with known weld defect samples for similarity to achieve optimal adjustment of vector machine regression model parameters.

[0038] In a second aspect, an embodiment of the present application provides a device for detecting weld defects by ultrasonic guided waves based on machine learning, and the device for detecting weld defects by ultrasonic guided waves based on machine learning comprises:

[0039] An ultrasonic guided wave generating device, which is used to emit ultrasonic guided wave signals to the weld defect sample and to emit ultrasonic guided wave signals to the weld to be inspected;

[0040] An ultrasonic guided wave sensor is used to capture the weld defect characteristic signal of the weld defect sample and the reflection signal data of the weld to be detected;

[0041] The first machine learning module is used to establish an asymptotic approximation model for the reflection signal of the weld defect characteristics of the weld defect sample with known defects, and use the P-PCA algorithm to extract the main component and prominent component features and perform pattern recognition. According to the characteristics and patterns of the extracted weld defect samples, they are compared with the known defects, and the K nearest neighbor model is trained. At the same time, a defect identification model library for storing weld defect characteristics is established, and the reflection signal data is identified by the trained K nearest neighbor model, and compared and analyzed in combination with the defect identification model library to determine the detection results, and the specific information of the weld defects in the detection results is determined based on the distance method.

[0042] The second machine learning module is used to use supervised learning based on the difference feature points between the test results and the defect identification model library, take the test results with differences as new sample data, and perform supervised learning training on the defect identification model library in combination with the feature information of the sample data, and classify and identify the feature information of the newly derived samples;

[0043] The third machine learning module is used to perform nonlinear mapping on the weld defect features of the weld defect samples obtained through supervised learning by using support vector machine regression and comparison training with known defects, improve the identification information of the weld defect samples according to the format of the defect identification model library, convert the weld defect features into a form that is easy to classify and identify, and derive samples from the defect identification model library;

[0044] A data collector, which is used to transmit the reflection signal of the weld defect characteristics of the weld defect sample captured by the ultrasonic guided wave sensor and the reflection signal data of the weld to be detected to the machine learning module;

[0045] Among them, the weld defect samples with known defects are used to construct the initial defect identification model library and used for weld defect feature mapping when using support vector machine regression;

[0046] The identification of the difference feature points is to calculate the similarity or difference between the detection results and the characteristics of various types of weld defects in the defect identification model library, and determine whether there is a difference by combining the similarity or difference calculation results and referring to the preset threshold;

[0047] The control training is to perform similarity matching between the acquired weld defect features and known weld defect samples to achieve optimal adjustment of the vector machine regression model parameters.

[0048] In a third aspect, an embodiment of the present application provides an ultrasonic guided wave device for detecting weld defects based on machine learning, wherein the ultrasonic guided wave device for detecting weld defects based on machine learning comprises a processor, a memory, and an ultrasonic guided wave program for detecting weld defects based on machine learning stored in the memory and executable by the processor, wherein when the ultrasonic guided wave program for detecting weld defects based on machine learning is executed by the processor, the steps of the above-mentioned method for detecting weld defects based on machine learning by ultrasonic guided waves are implemented.

[0049] The beneficial effects brought by the technical solution provided in the embodiments of the present application include:

[0050] The weld defect characteristic signals of a large number of weld defect samples with known defects are obtained based on ultrasonic guided wave technology, and an asymptotic approximation model containing quasi-modal displacement components is established. The P-PCA algorithm is used to extract the main component and prominent component features and recognize the patterns of the obtained weld defect characteristic signals. Then, according to the characteristics and patterns of the extracted weld defect samples, they are compared with the known defects, and the K nearest neighbor model is trained. At the same time, a defect identification model library for storing weld defect characteristics is established. Then, an ultrasonic guided wave signal is emitted to the weld to be inspected and the reflected signal data is captured. The P-PCA algorithm is used to extract the weld defect characteristic signal of the captured reflected signal, and the extracted data is input into the trained K nearest neighbor model. Then, the K nearest neighbor model identifies the reflected signal data, and compares and analyzes it in combination with the defect identification model library to determine the detection results, and the specific information of the weld defects in the detection results is determined based on the distance method. In addition, supervised learning and SVM regression are used to derive samples from the model library based on the difference feature points between the detection results and the model library. It can effectively improve detection accuracy and efficiency, combine ultrasonic guided wave technology and machine learning algorithms to achieve accurate detection and classification of weld defects, while reducing dependence on personnel, significantly improving the accuracy and reliability of weld defect detection, and providing strong protection for the safety of engineering structures. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A schematic diagram of the specific process of implementing the method for detecting weld defects using ultrasonic guided waves based on machine learning in this application;

[0052] Figure 2 This is a schematic diagram of the functional modules of the ultrasonic guided wave device for detecting weld defects based on machine learning in this application;

[0053] Figure 3 This is a schematic diagram of the hardware structure of the ultrasonic guided wave detection weld defect equipment based on machine learning in this application. DETAILED DESCRIPTION

[0054] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0055] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0056] On the first aspect, the embodiment of the present application provides a method for detecting weld defects by ultrasonic guided waves based on machine learning. By combining ultrasonic guided wave technology and machine learning algorithms, accurate detection and classification of weld defects are achieved, and a defect identification model library is established to improve detection accuracy and efficiency, reduce dependence on operators, and provide strong protection for the safety of engineering structures. The present application has broad application prospects in mechanical engineering and automation, artificial intelligence and machine learning, non-destructive testing, intelligent manufacturing or Industry 4.0 and other fields.

[0057] In an embodiment, a method for detecting weld defects using ultrasonic guided waves based on machine learning includes:

[0058] S1: Based on the ultrasonic guided wave technology, the reflection signal of the weld defect characteristics of the weld defect sample with known defects is obtained, an asymptotic approximation model is established, and the P-PCA algorithm is used to extract the main component and prominent component features and perform pattern recognition on the reflection signal of the obtained weld defect characteristics; among them, the number of weld defect samples needs to be larger.

[0059] It should be noted that when extracting the main components and prominent components of the acquired weld defect characteristic signals, the extracted characteristic points are greater than or equal to 5, and the Maximum Mean Discrepancy Algorithm (MMD algorithm) is used as a measurement indicator for selection and quantity requirements of the parameters of the main components and prominent components of the defect characteristic points.

[0060] In view of the drawback that feature extraction needs to rely on empirical knowledge, it has been proposed to use the principal component analysis (PCA) algorithm, that is, based on mathematical statistics, from the data of the waveguide detection signal, the law between the data is discovered to achieve feature extraction. However, PCA is a global feature extraction method. In the feature extraction of the waveguide detection signal, it causes the loss of local weak information in the signal; when facing the defects of the long welds of the structure (especially the structural welds connecting the U rib and the top plate of the orthotropic steel bridge deck of the bridge), it will cause the modal conversion and dispersion of the waveguide signal, and different defect information can be retained in the local area of ​​the waveguide detection signal. For this reason, the present application proposes a P-PCA (Principal & Prominence Component Analysis) algorithm for the extraction of global principal component features and local key salient component features of the waveguide detection signal, so as to extract the defect features from each local area or segment to achieve the extraction of useful features of the detection signal.

[0061] The global principal component feature focuses on the spectrum amplitude difference and is only affected by the change of signal energy, but not by the change of time domain waveform and time. At the same time, an amplitude adjustment coefficient is added. Therefore, the extraction function of the global principal component feature is defined as:

[0062]

[0063] in, represents the spectrum amplitude difference, represents the amplitude adjustment coefficient, Indicates the reference value or standard value of the spectrum amplitude difference when there is no defect. Indicates the detection value of the spectrum amplitude difference when there is a defect, Indicates the angular frequency at which the waveguide detection signal ends, Indicates the angular frequency at which the waveguide detection signal starts;

[0064] The local key salient component feature focuses on the spectrum amplitude deviation, which is affected by the time domain waveform and time changes, and reflects the degree of deviation of the defective signal relative to the non-defective signal. The ratio of the defective signal to the baseline signal is used for extraction; at the same time, a deviation adjustment coefficient is added. Therefore, the extraction function of the local key salient component feature is defined as:

[0065]

[0066] in, Indicates the degree of spectrum amplitude deviation, represents the deviation adjustment coefficient, Indicates the reference value or standard value of the spectrum amplitude deviation when there is no defect. Indicates the detection value of spectrum amplitude deviation when there is a defect, Indicates the time when the waveguide detection signal ends. Indicates the time when the waveguide detection signal starts.

[0067] It should be noted that for the amplitude adjustment coefficient and the deviation adjustment coefficient, the parametric adjustment technology of the software platform can quickly and efficiently extract the main defect components and prominent component characteristics in the weld defect characteristic signal, and realize accurate pattern recognition, overcoming the tediousness and limitations of the traditional method of relying on adjusting the signal gain or amplifier to re-collect the signal, and realizing a more comprehensive and detailed signal feature analysis in the global and local areas, thereby improving the efficiency and accuracy of weld defect detection.

[0068] Furthermore, in the embodiment, the reflection signal of the weld defect characteristics of the weld defect sample with known defects is obtained based on the ultrasonic guided wave technology, an asymptotic approximation model is established, and the P-PCA algorithm is used to extract the main component and prominent component features of the obtained reflection signal of the weld defect characteristics and perform pattern recognition, specifically including:

[0069] S101: For a weld defect sample with known defects, a reflection signal of a weld defect characteristic of the weld defect sample is obtained based on ultrasonic guided wave technology, and an asymptotic approximation model is established;

[0070] S102: Based on the principal component analysis model and using the P-PCA algorithm, the main component and prominent component features of the acquired weld defect characteristic signal are extracted and the pattern is recognized.

[0071] Specifically, the weld defect characteristic signal of the weld defect sample is obtained by ultrasonic guided wave technology, and then the principal component analysis model and the P-PCA algorithm are used to perform feature extraction and pattern recognition on the obtained weld defect characteristic signal.

[0072] S2: Based on the characteristics and patterns of the extracted weld defect samples, they are compared with known defects and trained in combination with the K-nearest neighbor model to establish a defect identification model library for storing weld defect characteristics;

[0073] It should be noted that to solve the problem of test accuracy affected by insufficient training samples, the method of random neuron inactivation can be used to suppress overfitting.

[0074] Furthermore, in the embodiment, the training is performed in combination with the K nearest neighbor model, specifically including:

[0075] S201: creating a K nearest neighbor model, and using the extracted features to train the K nearest neighbor model;

[0076] S202: Using cross entropy as the loss function, comparing with known defects, and optimizing the training process of the K nearest neighbor model.

[0077] That is, in the training process of the K-nearest neighbor model, cross entropy is used as the loss function to optimize the training process of the machine learning model, so as to achieve accurate detection and classification of weld defects. This application not only improves the detection accuracy, but also optimizes the model training process, significantly improving the reliability and efficiency of weld defect detection.

[0078] It should be noted that the ultrasonic guided wave detection method for weld defects based on machine learning in this application is suitable for weld defect detection in engineering structures with narrow spaces such as bridges, ships, building construction, pipelines, etc. By optimizing the model training process, it not only improves the detection accuracy and efficiency, but also reduces the dependence on the professional knowledge of operators, providing a strong guarantee for the safe operation of engineering structures.

[0079] S3: transmitting ultrasonic guided wave signals to the weld to be inspected and capturing reflected signal data, extracting weld defect feature signals of the captured reflected signal using the P-PCA algorithm, and inputting the extracted data into the trained K nearest neighbor model;

[0080] Furthermore, in the embodiment, an ultrasonic guided wave signal is emitted to the weld to be inspected and reflected signal data is captured, a P-PCA algorithm is used to extract the weld defect characteristic signal of the captured reflected signal, and the extracted data is input into the trained K nearest neighbor model, specifically including:

[0081] S301: placing an ultrasonic guided wave sensor near the weld to be inspected, and transmitting an ultrasonic guided wave signal to the weld to be inspected through an ultrasonic guided wave generating device;

[0082] S302: The reflected signal data is captured by an ultrasonic guided wave sensor, the P-PCA algorithm is used to extract the weld defect characteristic signal of the captured reflected signal, and the extracted data is input into the trained K nearest neighbor model.

[0083] S4: Use the K nearest neighbor model to identify the weld defect characteristic signal data of the reflection signal, and compare and analyze it with the defect identification model library to determine the detection result, and calculate and determine the specific information of the weld defect in the detection result based on the distance method;

[0084] The reflected signal data is identified through the trained K nearest neighbor model. At the same time, the defect identification model library stores the weld defect characteristics. The results obtained by the K nearest neighbor model can be compared and analyzed with the weld defect characteristics stored in the defect identification model library, thereby improving the accuracy of the identification results, thereby achieving accurate identification of whether there are defects in the weld to be detected and the defect characteristics. At the same time, in the identification process, the recognition efficiency and eigenvalue sensitivity are improved based on the recall rate. By analyzing the recall rate, the recognition efficiency and eigenvalue sensitivity can be improved, thereby more accurately determining the type and degree of weld defects, providing an important basis for the safety assessment of engineering structures.

[0085] The defect identification model library stores characteristic signals and classification information of various weld defect samples. By comparing and analyzing the actual captured reflection signal data, the defect identification model library can provide strong data support for the accurate detection of weld defects. In addition, the defect identification model library also supports continuous updating and optimization of data to meet the evolving safety needs of engineering structures.

[0086] Furthermore, in the process of model training and the identification of reflected signal data, redundant data can be eliminated to ensure the coverage of key abnormal data, thereby reducing the complexity of the model. By optimizing the training process, the generalization ability of the model can be improved, the risk of overfitting can be reduced, and the detection accuracy and efficiency can be improved, providing strong support for the practical application of weld defect detection.

[0087] Further, in the embodiment, the specific information of the weld defect in the detection result is determined based on the distance method, specifically including:

[0088] S401: sorting the detection results using the confusion matrix rule;

[0089] S402: Calculate the distances between different weld defect categories based on the distance method, and determine the specific information of the weld defect in the detection result. The specific information of the weld defect includes the type of weld defect, the severity of the weld defect, and the location of the weld defect.

[0090] The detection results are sorted by using the confusion matrix rule. By calculating the distance between different categories, the information of weld defects, including the type, degree and location of the defects, can be determined more accurately, which not only improves the detection accuracy and reliability, but also provides an important basis for subsequent maintenance and treatment.

[0091] S5: Based on the difference feature points between the test results and the defect identification model library, supervised learning is adopted to use the test results with differences as new sample data, and supervised learning training is performed on the defect identification model library in combination with the feature information of the sample data, and the feature information of the newly derived samples is classified and identified;

[0092] S6: Through support vector machine regression and training with known defects, nonlinear mapping is performed on the weld defect features of the weld defect samples obtained through supervised learning. The identification information of the weld defect samples is improved according to the format of the defect identification model library, the weld defect features are converted into a form that is easy to classify and identify, and samples are derived from the defect identification model library.

[0093] That is, supervised learning and support vector machine regression (SVM regression) are used to derive samples of the defect identification model library based on the difference feature points between the detection results and the defect identification model library. The defect identification model library is trained through supervised learning by taking the different detection results as new sample data and combining them with their corresponding labels (i.e., defect type, degree and other information). This allows the defect identification model library to gradually adapt to and identify more types of weld defects. Through SVM regression, the weld defect feature data can be nonlinearly mapped and converted into a form that is easier to classify and identify, thereby further improving the accuracy and practicality of the defect identification model library.

[0094] Through supervised learning training of the defect identification model library, defects are identified based on the difference feature points between the detection results and the defect identification model library, thereby improving the accuracy and generalization ability of the defect feature points. At the same time, support vector machine regression technology is used to compare with known defects, perform nonlinear mapping on the weld defect features, and convert the original features into a form that is easier to classify and identify, thereby improving the sample management capabilities of the defect identification model library.

[0095] Furthermore, based on the difference feature points between the test results and the defect identification model library, supervised learning is adopted, the test results with differences are used as new sample data, and the defect identification model library is trained by supervised learning in combination with the feature information of the sample data, and the feature information of the newly derived samples is classified and identified, wherein the identification of the difference feature points specifically includes:

[0096] Calculate the similarity or difference between the detection results and the characteristics of various types of weld defects in the defect identification model library, and determine whether there is a difference by combining the similarity or difference calculation results and referring to the preset threshold.

[0097] Furthermore, through support vector machine regression and comparison training with known defects, nonlinear mapping is performed on the weld defect features of the weld defect samples obtained through supervised learning, the identification information of the weld defect samples is improved according to the format of the defect identification model library, the weld defect features are converted into a form that is easy to classify and identify, and the defect identification model library is sample derived, wherein the comparison training specifically includes:

[0098] The acquired weld defect features are matched with known weld defect samples for similarity to achieve optimal adjustment of vector machine regression model parameters.

[0099] See also Figure 1 As shown, for the ultrasonic guided wave detection method for weld defects based on machine learning in this application, the specific process during implementation includes:

[0100] S11: Defect sample feature extraction: For a large number of weld defect samples with known defects, the reflected signal of the weld defect characteristics is obtained through ultrasonic guided wave technology, and the main component and prominent component characteristics of the defect are extracted using the P-PCA algorithm;

[0101] S12: Building a defect identification model library: training with known defects and building a defect identification model library in combination with the K nearest neighbor model;

[0102] S13: Extraction of weld defect features to be detected: using the P-PCA algorithm to identify and extract weld defect features of the weld to be detected;

[0103] S14: Defect assessment of the weld to be inspected: fuzzy matching is performed between the K nearest neighbor model and the defect identification model library to quickly determine the weld defect classification and calculate the specific weld defect location;

[0104] S15: Confirmation of defect assessment results of the weld to be inspected: Based on the difference feature points between the inspection results and the defect identification model library, supervised learning is adopted to use the inspection results with differences as new sample data, and supervised learning training is performed on the defect identification model library in combination with the feature information of the sample data, and the feature information of the newly derived samples is classified and identified;

[0105] S16: Sample derivation: Through support vector machine regression and training with known defects, nonlinear mapping is performed on the weld defect features of the weld defect samples obtained through supervised learning. The identification information of the weld defect samples is improved according to the format of the defect identification model library, the weld defect features are converted into a form that is easy to classify and identify, and sample derivation is performed on the defect identification model library.

[0106] This application can reduce the reliance on the professional knowledge of operators through an automatic detection method based on machine learning, and can reduce the impact of human factors on the detection results through automated and intelligent processing, thereby improving detection efficiency and accuracy, and providing a broader prospect for the practical application of weld defect detection. This application can timely discover and handle weld defects, and through precise detection and analysis, it can provide strong support for the safety assessment and maintenance of engineering structures. In addition, this application can also reduce the reliance on operators, improve detection efficiency and accuracy, and provide strong guarantees for the safe operation of engineering structures.

[0107] By continuously optimizing and updating the machine learning model, the accuracy and efficiency of weld defect detection are continuously improved. With the continuous development and advancement of technology, the ultrasonic guided wave method for detecting weld defects in this application can adapt to the ever-changing safety requirements of engineering structures and provide more advanced and reliable technical support for weld defect detection. This application not only uses ultrasonic guided wave data, but also combines data from other modalities (such as visual images, thermal imaging, etc.) for comprehensive analysis. Through multimodal data fusion, it can more comprehensively understand the state and defects of the weld, thereby improving the accuracy and reliability of detection. Regularly calibrate and maintain the weld defect detection equipment to ensure the accuracy and stability of the equipment. Regular calibration and maintenance can reduce equipment errors and failure rates, and improve the accuracy and reliability of weld defect detection.

[0108] Machine learning technology can automatically extract the features of weld defects and classify them, thereby significantly improving the accuracy and efficiency of detection. Through machine learning algorithms, the complex signals generated by ultrasonic guided wave detection can be deeply analyzed, and key features related to defects can be extracted, thereby accurately identifying weld defects. In addition, ultrasonic guided wave detection technology based on machine learning can also optimize the model training process and improve detection accuracy and efficiency. Through continuous learning and optimization, machine learning models can gradually adapt to different weld structures and material properties, thereby achieving more accurate and efficient detection of weld defects. It not only helps to improve the accuracy and efficiency of weld detection, but also helps to promote the further development of ultrasonic guided wave nondestructive testing technology.

[0109] The method for detecting weld defects with ultrasonic guided waves based on machine learning in the embodiment of the present application obtains weld defect characteristic signals of a large number of weld defect samples with known defects based on ultrasonic guided wave technology, establishes an asymptotic approximation model including a quasi-modal displacement component, and uses the P-PCA algorithm to extract the main component and prominent component features and perform pattern recognition on the acquired weld defect characteristic signals, then compares the features and patterns of the extracted weld defect samples with the known defects, trains the K nearest neighbor model, and establishes a defect identification model library for storing weld defect features; then, an ultrasonic guided wave signal is emitted to the weld to be detected and the reflected signal data is captured, and the P-PCA algorithm is used to extract the weld defect characteristic signal of the captured reflected signal, and the extracted data is input into the trained K nearest neighbor model; then, the K nearest neighbor model identifies the reflected signal data, and performs comparative analysis in combination with the defect identification model library to determine the detection result, and determines the specific information of the weld defect in the detection result based on the distance method calculation; in addition, supervised learning and SVM regression are used to derive samples from the model library based on the difference feature points between the detection result and the model library. It can effectively improve detection accuracy and efficiency, combine ultrasonic guided wave technology and machine learning algorithms to achieve accurate detection and classification of weld defects, while reducing dependence on personnel, significantly improving the accuracy and reliability of weld defect detection, and providing strong protection for the safety of engineering structures.

[0110] In a second aspect, an embodiment of the present application also provides an ultrasonic guided wave device for detecting weld defects based on machine learning.

[0111] In the embodiment, refer to Figure 2 , Figure 2 This is a functional module diagram of the ultrasonic guided wave detection device for weld defects based on machine learning in this application. Figure 2 As shown, the ultrasonic guided wave detection device for weld defects based on machine learning includes: an ultrasonic guided wave generating device, an ultrasonic guided wave sensor, a first machine learning module, a second machine learning module, a third machine learning module, and a data collector.

[0112] The ultrasonic guided wave generating device is used to transmit ultrasonic guided wave signals to the weld defect sample, and to transmit ultrasonic guided wave signals to the weld to be inspected; the ultrasonic guided wave sensor is used to capture the reflection signal of the weld defect characteristics of the weld defect sample, and the reflection signal data of the weld to be inspected.

[0113] The first machine learning module is used to establish an asymptotic approximation model for the reflection signal of the weld defect characteristics of the weld defect samples with known defects, and use the P-PCA algorithm to extract the main components and prominent components of the defects and perform pattern recognition. According to the characteristics and patterns of the extracted weld defect samples, they are compared with the known defects, and the K nearest neighbor model is trained. At the same time, a defect identification model library for storing weld defect characteristics is established, and the reflected signal data is identified through the trained K nearest neighbor model, and compared and analyzed in combination with the defect identification model library to determine the detection results, and the specific information of the weld defects in the detection results is determined based on the distance method. The second machine learning module is used Based on the difference feature points between the detection results and the defect identification model library, supervised learning is adopted, and the detection results with differences are used as new sample data. The defect identification model library is trained through supervised learning in combination with the feature information of the sample data, and the feature information of the newly derived samples is classified and identified; the third machine learning module is used to perform nonlinear mapping of the weld defect features of the weld defect samples obtained through supervised learning through support vector machine regression and training against known defects, and improve the identification information of the weld defect samples according to the format of the defect identification model library, convert the weld defect features into a form that is easy to classify and identify, and derive samples from the defect identification model library.

[0114] The data collector is used to transmit the reflection signal of the weld defect characteristics of the weld defect sample captured by the ultrasonic guided wave sensor and the reflection signal data of the weld to be inspected to the machine learning module.

[0115] Among them, the identification of the difference feature points is to calculate the similarity or difference between the detection results and the various types of weld defect features in the defect identification model library, and determine whether there is a difference by combining the similarity or difference calculation results and referring to the preset threshold; the control training is to perform similarity matching between the acquired weld defect features and the known weld defect samples to achieve the optimization adjustment of the vector machine regression model parameters.

[0116] In the present application, the ultrasonic guided wave sensor is located near the weld defect sample to capture the weld defect characteristic signal reflected by the weld defect sample, and is located near the weld to be inspected to capture the reflection signal data reflected by the weld to be inspected.

[0117] Among them, the functional implementation of each module in the above-mentioned ultrasonic guided wave detection device for weld defects based on machine learning corresponds to the various steps in the above-mentioned embodiment of the ultrasonic guided wave detection method for weld defects based on machine learning, and their functions and implementation processes will not be repeated here one by one.

[0118] In a third aspect, an embodiment of the present application provides an ultrasonic guided wave device for detecting weld defects based on machine learning. The ultrasonic guided wave device for detecting weld defects based on machine learning can be a personal computer (PC), a laptop computer, a server, or other device with data processing capabilities.

[0119] Reference Figure 3 , Figure 3 The hardware structure diagram of the ultrasonic guided wave detection device for weld defects based on machine learning involved in the embodiment of the present application is shown in FIG. In the embodiment of the present application, the ultrasonic guided wave detection device for weld defects based on machine learning may include a processor, a memory, a communication interface, and a communication bus.

[0120] The communication bus may be of any type and is used to interconnect the processor, the memory, and the communication interface.

[0121] The communication interface includes input / output (I / O) interface, physical interface and logical interface, etc., which are used to realize the interconnection of devices inside the ultrasonic guided wave detection device for weld defects based on machine learning, and the interface used to realize the interconnection between the ultrasonic guided wave detection device for weld defects based on machine learning and other devices (such as other computing devices or user devices). The physical interface can be an Ethernet interface, a fiber optic interface, an ATM interface, etc.; the user device can be a display, a keyboard, etc.

[0122] The memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0123] The processor may be a general-purpose processor, which may call the ultrasonic guided wave detection of weld defects program based on machine learning stored in the memory, and execute the ultrasonic guided wave detection of weld defects method based on machine learning provided in the embodiment of the present application. For example, the general-purpose processor may be a central processing unit (CPU). The method executed when the ultrasonic guided wave detection of weld defects program based on machine learning is called may refer to the various embodiments of the ultrasonic guided wave detection of weld defects method based on machine learning of the present application, which will not be repeated here.

[0124] Those skilled in the art will understand that Figure 3 The hardware structure shown in the figure does not constitute a limitation on the present application, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0125] It should be noted that the serial numbers of the above-mentioned embodiments of the present application are only for description and do not represent the advantages or disadvantages of the embodiments.

[0126] The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices. The terms "first", "second" and "third" are used to distinguish different objects, etc., and do not represent a sequence, nor do they limit "first", "second" and "third" to different types.

[0127] In the description of the embodiments of the present application, "exemplary", "for example" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary", "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary", "for example" or "for example" is intended to present related concepts in a specific way.

[0128] In the description of the embodiments of the present application, unless otherwise specified, “ / ” means or, for example, A / B can mean A or B; the “and / or” in the text is merely a description of the association relationship of associated objects, indicating that three relationships may exist, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, “multiple” refers to two or more than two.

[0129] In some processes described in the embodiments of the present application, multiple operations or steps that appear in a specific order are included, but it should be understood that these operations or steps may not be executed in the order in which they appear in the embodiments of the present application or in parallel, and the sequence number of the operation is only used to distinguish the different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed in sequence or in parallel, and these operations or steps may be combined.

[0130] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD) as described above, and includes a number of instructions for a terminal device to execute the methods described in each embodiment of the present application.

[0131] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A method for detecting weld defects using ultrasonic guided waves based on machine learning, characterized in that: The method for detecting weld defects by ultrasonic guided waves based on machine learning includes: Based on ultrasonic guided wave technology, the reflection signal of the weld defect characteristics of the weld defect sample with known defects is obtained, and an asymptotic approximation model is established. The P-PCA algorithm is used to extract the main component and prominent component features and perform pattern recognition on the reflection signal of the obtained weld defect characteristics. According to the characteristics and patterns of the extracted weld defect samples, they are compared with known defects and trained in combination with the K nearest neighbor model to establish a defect identification model library for storing weld defect characteristics; The ultrasonic guided wave signal is emitted to the weld to be inspected and the reflected signal data is captured. The P-PCA algorithm is used to extract the weld defect characteristic signal of the captured reflected signal, and the extracted data is input into the trained K nearest neighbor model. The K nearest neighbor model is used to identify the characteristic signal data of the weld defect of the reflection signal, and the defect identification model library is combined for comparative analysis to determine the detection results. The specific information of the weld defect in the detection results is determined based on the distance method. Based on the difference feature points between the test results and the defect identification model library, supervised learning is adopted to use the different test results as new sample data, and the defect identification model library is trained through supervised learning based on the feature information of the sample data, and the feature information of the newly derived samples is classified and identified; Through support vector machine regression and training with known defects, nonlinear mapping is performed on the weld defect features of the weld defect samples obtained through supervised learning. The identification information of the weld defect samples is improved according to the format of the defect identification model library, the weld defect features are converted into a form that is easy to classify and identify, and samples are derived from the defect identification model library. Among them, for the selection and quantity setting of the characteristic point parameters of the main component and the salient component, the maximum mean difference algorithm is used as the measurement index for selection; The P-PCA algorithm is used to extract the global principal component features and local key salient component features of the guided wave detection signal, so as to extract the defect features from each local area or segment to realize the extraction of useful features of the detection signal; The extraction function of the global principal component feature is specifically: in, represents the spectrum amplitude difference, represents the amplitude adjustment coefficient, Indicates the reference value or standard value of the spectrum amplitude difference when there is no defect. Indicates the detection value of the spectrum amplitude difference when there is a defect, Indicates the angular frequency at which the waveguide detection signal ends, Indicates the angular frequency at which the waveguide detection signal starts; The extraction function of the local key salient component feature is specifically: in, Indicates the degree of spectrum amplitude deviation, represents the deviation adjustment coefficient, Indicates the reference value or standard value of the spectrum amplitude deviation when there is no defect. Indicates the detection value of spectrum amplitude deviation when there is a defect, Indicates the time when the waveguide detection signal ends. Indicates the time when the waveguide detection signal starts.

2. The method for detecting weld defects using ultrasonic guided waves based on machine learning as claimed in claim 1, characterized in that: The method of obtaining the reflection signal of the weld defect characteristics of the weld defect sample with known defects based on ultrasonic guided wave technology, establishing an asymptotic approximation model, and using the P-PCA algorithm to extract the main component and prominent component features of the obtained reflection signal of the weld defect characteristics and perform pattern recognition, specifically includes: For weld defect samples with known defects, the reflection signal of the weld defect characteristics of the weld defect samples is obtained based on ultrasonic guided wave technology, and an asymptotic approximation model is established; Based on the principal component analysis model and the P-PCA algorithm, the main component and prominent component features of the acquired weld defect characteristics are extracted and the pattern is recognized.

3. The method for detecting weld defects using ultrasonic guided waves based on machine learning as claimed in claim 1, characterized in that: The training in combination with the K nearest neighbor model specifically includes: Create a K-nearest neighbor model and use the extracted features to train the K-nearest neighbor model; The cross entropy is used as the loss function and compared with known defects to optimize the training process of the K nearest neighbor model.

4. The method for detecting weld defects using ultrasonic guided waves based on machine learning as claimed in claim 1, characterized in that: The method of transmitting an ultrasonic guided wave signal to the weld to be inspected and capturing reflected signal data, extracting the weld defect characteristic signal of the captured reflected signal using the P-PCA algorithm, and inputting the extracted data into the trained K nearest neighbor model specifically includes: The ultrasonic guided wave sensor is placed near the weld to be inspected, and the ultrasonic guided wave signal is emitted to the weld to be inspected through the ultrasonic guided wave generating device; The reflected signal data is captured by an ultrasonic guided wave sensor, and the P-PCA algorithm is used to extract the weld defect characteristic signal of the captured reflected signal. The extracted data is input into the trained K nearest neighbor model.

5. The method for detecting weld defects using ultrasonic guided waves based on machine learning as claimed in claim 1, characterized in that: The specific information of the weld defects in the detection results determined by the distance method specifically includes: The confusion matrix rule is used to sort the detection results; The distances between different weld defect categories are calculated based on the distance method to determine the specific information of the weld defects in the detection results.

6. The method for detecting weld defects using ultrasonic guided waves based on machine learning as claimed in claim 1, characterized in that: Based on the difference feature points between the detection results and the defect identification model library, supervised learning is adopted, the detection results with differences are used as new sample data, and the defect identification model library is trained by supervised learning in combination with the feature information of the sample data, and the feature information of the newly derived samples is classified and identified, wherein the identification of the difference feature points specifically includes: Calculate the similarity or difference between the detection results and the characteristics of various types of weld defects in the defect identification model library, and determine whether there is a difference by combining the similarity or difference calculation results and referring to the preset threshold.

7. The method for detecting weld defects using ultrasonic guided waves based on machine learning as claimed in claim 1, characterized in that: The support vector machine regression is used for comparison training with known defects, and nonlinear mapping is performed on the weld defect features of the weld defect samples obtained through supervised learning. The identification information of the weld defect samples is improved according to the format of the defect identification model library, and the weld defect features are converted into a form that is easy to classify and identify. Samples are derived from the defect identification model library, wherein the comparison training specifically includes: The acquired weld defect features are matched with known weld defect samples for similarity to achieve optimal adjustment of vector machine regression model parameters.

8. An ultrasonic guided wave device for detecting weld defects based on machine learning, characterized in that: The ultrasonic guided wave weld defect detection device based on machine learning includes: An ultrasonic guided wave generating device, which is used to emit ultrasonic guided wave signals to the weld defect sample and to emit ultrasonic guided wave signals to the weld to be inspected; An ultrasonic guided wave sensor, which is used to capture the reflection signal of the weld defect characteristics of the weld defect sample and the reflection signal data of the weld to be inspected; The first machine learning module is used to establish an asymptotic approximation model for the reflection signal of the weld defect characteristics of the weld defect sample with known defects, and use the P-PCA algorithm to extract the main component and prominent component features and perform pattern recognition. According to the characteristics and patterns of the extracted weld defect samples, they are compared with the known defects, and the K nearest neighbor model is trained. At the same time, a defect identification model library for storing weld defect characteristics is established, and the reflection signal data is identified by the trained K nearest neighbor model, and compared and analyzed in combination with the defect identification model library to determine the detection results, and the specific information of the weld defects in the detection results is determined based on the distance method. The second machine learning module is used to use supervised learning based on the difference feature points between the test results and the defect identification model library, take the test results with differences as new sample data, and perform supervised learning training on the defect identification model library in combination with the feature information of the sample data, and classify and identify the feature information of the newly derived samples; The third machine learning module is used to perform nonlinear mapping on the weld defect features of the weld defect samples obtained through supervised learning by using support vector machine regression and comparison training with known defects, improve the identification information of the weld defect samples according to the format of the defect identification model library, convert the weld defect features into a form that is easy to classify and identify, and derive samples from the defect identification model library; A data collector, which is used to transmit the reflection signal of the weld defect characteristics of the weld defect sample captured by the ultrasonic guided wave sensor and the reflection signal data of the weld to be detected to the machine learning module; Among them, the weld defect samples with known defects are used to construct the initial defect identification model library and used for weld defect feature mapping when using support vector machine regression; The identification of the difference feature points is to calculate the similarity or difference between the detection results and the characteristics of various types of weld defects in the defect identification model library, and determine whether there is a difference by combining the similarity or difference calculation results and referring to the preset threshold; The control training is to perform similarity matching between the acquired weld defect features and known weld defect samples to achieve optimal adjustment of the vector machine regression model parameters; The P-PCA algorithm is used to extract the global principal component features and local key salient component features of the guided wave detection signal, so as to extract the defect features from each local area or segment to realize the extraction of useful features of the detection signal; The extraction function of the global principal component feature is specifically: in, represents the spectrum amplitude difference, represents the amplitude adjustment coefficient, Indicates the reference value or standard value of the spectrum amplitude difference when there is no defect. Indicates the detection value of the spectrum amplitude difference when there is a defect, Indicates the angular frequency at which the waveguide detection signal ends, Indicates the angular frequency at which the waveguide detection signal starts; The extraction function of the local key salient component feature is specifically: in, Indicates the degree of spectrum amplitude deviation, represents the deviation adjustment coefficient, Indicates the reference value or standard value of the spectrum amplitude deviation when there is no defect. Indicates the detection value of spectrum amplitude deviation when there is a defect, Indicates the time when the waveguide detection signal ends. Indicates the time when the waveguide detection signal starts.

9. An ultrasonic guided wave device for detecting weld defects based on machine learning, characterized in that: The device for detecting weld defects by ultrasonic guided waves based on machine learning includes a processor, a memory, and a program for detecting weld defects by ultrasonic guided waves based on machine learning stored in the memory and executable by the processor. When the program for detecting weld defects by ultrasonic guided waves based on machine learning is executed by the processor, the steps of the method for detecting weld defects by ultrasonic guided waves based on machine learning as described in any one of claims 1 to 7 are implemented.

Images