Polyethylene pipe sweat soldering defect detection system based on microwaves

Abstract

The invention discloses a microwave-based polyethylene pipe hot-melt welding defect detection system, and relates to the technical field of polyethylene pipe hot-melt welding joint defect detection, and the method comprises the following steps: firstly, collecting microwave reflection signals of normal and defect-containing hot-melt welding joints, and carrying out normalization; the processed signals are sliced, sliced fragments are labeled, time domain and frequency domain features are extracted, and a defect sample data set is constructed; carrying out feature selection on the extracted features, and selecting 10 features with the highest contribution to the recognition accuracy; constructing a support vector classifier model, performing training by using the data set after feature selection, and performing hyper-parameter optimization on the model to obtain an optimal classification model; designing and realizing a polyethylene pipe hot-melt welding defect detection system, and detecting whether a welding position contains defects and defect positions or not; according to the invention, accurate detection of polyethylene pipe hot melting welding defects can be realized, defects and positions can be rapidly identified, and technical support is provided for quality safety detection of polyethylene pipes.

Description

Technical Field

[0001] The invention relates to the technical field of polyethylene pipe hot-melt welding joint defect detection, in particular to a polyethylene pipe hot-melt welding defect detection method based on microwave reflection signals. Background Art

[0002] Polyethylene pipes are widely used in water and gas transportation due to their excellent corrosion resistance, toughness and long service life. To ensure the continuity and sealing of the pipeline, PE pipes are usually connected by hot melt welding. Hot melt welding forms a welded joint by heating the ends of the pipe to a molten state and then connecting them under high pressure. During the welding process, holes, slag inclusions and unfused defects may appear in the welded joint due to improper temperature control and uneven pressure. These defects not only affect the strength of the welded joint, but may also cause leakage or rupture during the use of the pipeline. These safety hazards directly affect the safety and reliability of the pipeline, so the detection of welded joints is particularly important. Traditional non-destructive testing methods, such as ultrasonic testing and X-ray testing, can be used to detect internal defects in welds, but due to the non-metallic characteristics of PE materials, these methods are affected by the sound velocity or electromagnetic properties of the material in terms of detection effect, and there are certain limitations. The detection results are not accurate enough, and even some internal minor defects cannot be effectively identified. Microwave detection technology is based on the interaction between high-frequency electromagnetic waves and the material interface, and measures the reflected signal to judge the structural changes and defects inside the material. Compared with other non-destructive testing methods, microwave testing has the advantages of non-contact, strong penetration, and fast detection speed, and can quickly locate potential defects in welds. In recent years, with the development of microwave technology, this technology has gradually been applied to defect detection of polyethylene pipe welded joints, and has achieved remarkable results.

[0003] Reference 1 (Zhu XW, Pan JP, Tan L J. Microwave scan inspection of HDPE piping thermal fusion welds for lack of fusion defect [J]. Applied Mechanics and Materials, 2013, 333: 1523-1528.) conducted microwave scanning inspection for the lack of fusion defect of polyethylene pipe hot melt welds, and identified the lack of fusion defect based on scanning image analysis, but it required imaging processing of signal data. Reference 2 (Carrigan TD, Forrest BE, Andem HN, et al. Nondestructive testing of nonmetallic pipelines using microwave reflectometry on an in-line inspection robot [J]. IEEE Transactions on Instrumentation and Measurement, 2018, 68 (2): 586-594.) measured the reflection signal of polyethylene pipelines using a K-band microwave reflectometry instrument installed on a pipeline crawling robot, and detected cracks and drilling defects, explaining the influence of frequency and lift-off distance on the quality of scanning images, but did not achieve intelligent and automated detection of defects.

[0004] In the detection of hot-melt welding defects in PE pipes, the analysis of microwave reflection signals has its own unique complexity. First, the material of the hot-melt joint is prone to local changes under the action of heat and pressure, such as uneven material thickness, internal bubbles or slag inclusions. These local microscopic defects will cause complex nonlinear changes in the amplitude, phase and other parameters of the microwave reflection signal, increasing the difficulty of signal feature extraction. Microwave signals are continuous high-frequency fluctuations, and directly using the original signal for defect identification is easily interfered by background noise and normal welding fluctuations. In addition, different defect types are often distributed nonlinearly and unbalanced in the feature space, further exacerbating the difficulty of feature extraction and classification. Summary of the invention

[0005] In order to solve the above-mentioned problems in the detection of PE pipe hot-melt welding defects, the present invention proposes a polyethylene pipe hot-melt welding defect detection system based on microwaves: first, the microwave reflection signals of normal and defective hot-melt welding joints are scanned and collected and normalized; the processed signals are sliced, the slice segments are labeled and the time domain and frequency domain features are extracted to construct a defect sample data set; feature selection is performed on the extracted features to select the 10 features that contribute the most to the recognition accuracy; a support vector classifier model is constructed, the data set after feature selection is used for training, and the model is hyper-parameter optimized to obtain the best classification model; a polyethylene pipe hot-melt welding defect detection system is designed and implemented to detect whether the weld contains defects and the defect location; the present invention can realize the accurate detection of polyethylene pipe hot-melt welding defects, quickly identify defects and locations, and provide technical support for the quality and safety detection of polyethylene pipes actually produced.

[0006] A polyethylene pipe hot-melt welding defect detection system based on microwave reflection signals specifically comprises the following steps: S1. Use a microwave probe to scan the detection area of ​​the polyethylene pipe hot-melt welding joint and collect the corresponding microwave reflection signal; The microwave waveguide probe is used to accurately scan the detection area of ​​the polyethylene pipe hot-melt joint to obtain the corresponding microwave reflection signal; the microwave waveguide probe performs high-precision scanning on the polyethylene pipe weld area under the specified detection path and scanning mode; collects S 11 Parameter signal, which describes the reflection coefficient and energy reflection of microwaves at the material interface. By capturing the reflection characteristics of microwaves at different defects, defects generated during the hot-melt welding of polyethylene pipes can be identified; The microwave operating frequency used is 12.4 GHz to 18 GHz. This frequency range can effectively penetrate polyethylene materials and provide sufficient resolution, which is suitable for the detection of defects of different types and sizes. 11 Parameter signals can accurately distinguish between holes, slag inclusions and lack of fusion defects, thus ensuring a comprehensive assessment of the quality of polyethylene pipe hot-melt joints; S2, normalizing the microwave reflection signal obtained in step S1 to ensure consistency of subsequent processing; Normalization processing limits the signal amplitude to the range of [0,1], with the purpose of standardizing the microwave reflection signals obtained under different detection conditions to eliminate non-target fluctuations caused by external factors such as the sensitivity of the signal acquisition equipment, the detection environment, or the change of the position of the microwave waveguide probe; S3, slicing the signal normalized in step S2 according to a preset slice width and step length to generate a plurality of signal segments of the same length; and labeling each slice according to whether the slice signal contains a defective part; The preprocessed microwave reflection signal is sliced ​​according to a preset slice width and step length so as to divide a continuous one-dimensional signal sequence into a plurality of signal segments of the same length; each slice signal is labeled according to whether it contains holes, slag inclusions and unfused defects; specifically, each signal segment is labeled as "0" or "1" by referring to actual detection conditions and known defect locations; and supervised learning is performed using the labeled slice data, so that the detection accuracy of the microwave detection algorithm for three common defects can be effectively improved, ensuring that the model has high robustness and accuracy under different working conditions; S4, extracting time domain and frequency domain features for each slice signal and complete signal, establishing a weld defect feature signal sample data set, and dividing all labeled samples into a training set and a test set; When analyzing the slice signal, a multidimensional feature extraction method in the time domain and frequency domain is used to fully capture the potential defect information in the signal; these features include the maximum value, minimum value, amplitude, energy, peak value, rise time, root mean square, variance, standard deviation, kurtosis, skewness, peak factor, waveform factor, pulse factor, margin factor features in the signal time domain, and the average power spectrum density, peak frequency, harmonics, correlation, phase response, autocorrelation, and cross-correlation features in the frequency domain; the purpose of feature extraction is to convert the one-dimensional microwave reflection signal into a high-dimensional feature vector through mathematical calculation, so as to construct a signal sample data set containing rich defect feature information { , },in Representative i The feature vector of the signal segment, =0 means the signal segment corresponds to the normal area, =1 indicates that the signal segment contains a defect area; then the data set is divided into a training set and a test set to further improve the learning and prediction accuracy of the model; S5. Based on the signal sample data set prepared in step S4, a defect detection classification model is constructed using a support vector classifier (SVC); the classifier performs supervised learning through the labeled training data, and the samples in the training set are classified according to their corresponding label information, and the labels include normal areas and areas containing defects; The SVC is trained using the signal sample data set generated in step S4 to construct a classification model for welding defect detection; the support vector classifier is a supervised learning algorithm commonly used for binary classification tasks, which can effectively detect holes, slag inclusions and unfused defects in polyethylene pipe hot-melt joints; in this process, the classifier performs supervised learning by using pre-labeled training data, so that the model can classify the slice signal according to the input signal characteristics; during the training process, the support vector classifier adjusts the model parameters by minimizing the classification loss function, and according to the given labeled training data, the model is optimized by step-by-step learning, so that it can be based on the input feature vector To predict the category of the signal segment, i.e., normal region or defective region; S6. The features extracted in step S4 are selected using a feature importance method. The feature selection method uses a Mann-Whitney U test for preliminary screening, and then the unimportant features are removed based on the SVC constructed in step S5 and a recursive feature elimination (RFE) method is used. The features are ranked according to their contribution to the model classification, and the top 10 important features are retained; Feature selection is divided into two stages: first, the Mann-Whitney U test is used to preliminarily screen the features; then, RFE is used to further select and finally retain the top ten features that contribute most to the classifier. The retained features are: maximum value, minimum value, amplitude, energy, peak value, root mean square, standard deviation, pulse factor, margin factor, and average power spectral density; S7. Use the Bayesian optimization algorithm to optimize the hyperparameters of the constructed SVC to improve the generalization ability of the classifier and the accuracy of defect detection; The optimization algorithm uses the Bayesian optimization algorithm to optimize the hyperparameters of the SVC; the scope of the hyperparameter optimization includes the regularization parameter, the kernel function type and the kernel parameter; the Bayesian optimization defines the accuracy of the test set as the objective function, continuously updates the proxy model, and performs SVC training and verification on each selected hyperparameter combination, and finally finds the optimal hyperparameter combination that minimizes the objective function as the final defect classifier; the classifier will be used to process new microwave signals to be detected and predict whether they contain defects; S8. Design a polyethylene pipe hot-melt welded joint defect detection system, use the SVC model optimized in step S7 to analyze the microwave scanning reflection signal in real time, detect and locate defects, and support fast and accurate defect detection.

[0007] A microwave-based polyethylene pipe hot-melt welding defect detection system includes the following modules: A microwave signal import module is used to import the collected microwave reflection signals of the microwave waveguide scanning polyethylene pipe hot-melt welding joint with an operating frequency range of 12.4 GHz to 18 GHz, and perform normalization; The signal slicing module is used to slice the normalized preprocessed signal according to the preset slice width and step length to generate signal segments. The step length is the same as the slice width by default. A feature extraction module is used to extract the aforementioned selected time domain features and frequency domain features from the signal fragments to provide a rich feature basis for classification; The defect recognition module is used to input the features extracted from the microwave reflection signal to be detected into the trained and optimized SVC model, automatically detect the weld defects of the polyethylene pipe hot-melt joint, and obtain the detection results of the signal to be detected; The information display module is used to visualize the detection results obtained by the detection module, display the defects and their locations in the form of pictures and output them on the display interface.

[0008] Due to the adoption of the above technical solution, the present invention has the following advantages: 1. The present invention proposes a microwave-based polyethylene pipe hot-melt welding defect detection method, which can use the simple structure of microwave reflection signal for processing, avoiding the calculation overhead brought by the imaging method; by slicing and local processing of the signal, the method can focus on a specific area of ​​the weld, thereby effectively capturing tiny signal changes and accurately identifying defects and locations; 2. The multi-dimensional feature extraction method based on time domain and frequency domain proposed in the present invention fully mines the time domain and frequency domain information of the signal, constructs a rich high-dimensional feature vector, and enhances the comprehensiveness and accuracy of detection through the extraction of multi-dimensional features; 3. The Bayesian optimization algorithm proposed in this invention optimizes the hyperparameters of the support vector classifier (SVC), efficiently searches for the best hyperparameter configuration while reducing the number of training times, thereby ensuring the optimal performance of the classifier. This method significantly improves the accuracy of defect identification and enhances its robustness under different operating environments; 4. The present invention proposes a microwave-based polyethylene pipe hot-melt welding defect detection system. The system achieves fast and accurate defect detection through precise microwave reflection signal analysis, comprehensive feature extraction and efficient classification network, significantly improving the automation and intelligence level of polyethylene pipe welding quality inspection, while effectively reducing the inspection cost of enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 This is a schematic diagram of a microwave-based polyethylene pipe hot-melt welding defect detection system according to an embodiment of the present invention; Figure 2This is a flow chart of a microwave-based polyethylene pipe hot-melt welding defect detection algorithm according to an embodiment of the present invention; Figure 3 This is a structural diagram of a microwave-based polyethylene pipe hot-melt welding defect detection system according to an embodiment of the present invention; Figure 4 A comparison diagram of defect detection accuracy between the comparison method and the embodiment of the present invention; DETAILED DESCRIPTION

[0010] In order to further describe the microwave-based polyethylene pipe hot-melt welding defect detection system proposed by the present invention, an embodiment of the present invention is described in conjunction with the accompanying drawings to enhance the understanding of the purpose, method structure and application effect of the present invention. The embodiment is only a further explanation of the present invention, and researchers in the field can make further non-fundamental technical improvements based on the present invention, which are also within the scope of protection of the present invention.

[0011] A microwave-based polyethylene pipe hot-melt welding defect detection system specifically comprises the following steps: S1. Use a microwave probe to scan the detection area of ​​the polyethylene pipe hot-melt welding joint and collect the corresponding microwave reflection signal; The microwave waveguide probe is used to accurately scan the detection area of ​​the polyethylene pipe hot-melt joint to obtain the corresponding microwave reflection signal; the microwave waveguide probe performs high-precision scanning on the polyethylene pipe weld area under the specified detection path and scanning mode; collects S 11 Parameter signal, which describes the reflection coefficient and energy reflection of microwaves at the material interface. By capturing the reflection characteristics of microwaves at different defects, defects generated during the hot-melt welding of polyethylene pipes can be identified; In the non-destructive testing of hot-melt welded joints, common defects include holes, slag inclusions and lack of fusion. Holes are caused by internal voids due to incomplete filling of materials. Slag inclusions are caused by foreign matter or impurities mixed in the weld surface, which affects the structural integrity of the joint. Lack of fusion refers to the incomplete fusion of materials at the weld, resulting in insufficient bonding strength of the weld surface. All three types of defects will cause abnormal changes in microwave reflection signals, which are specifically manifested as S 11 Sudden changes in parameters or significant differences in reflective characteristics; The microwave operating frequency used is 12.4 GHz to 18 GHz. This frequency range can effectively penetrate polyethylene materials and provide sufficient resolution, which is suitable for the detection of defects of different types and sizes. 11 Parameter signals can accurately distinguish between holes, slag inclusions and lack of fusion defects, thus ensuring a comprehensive assessment of the quality of polyethylene pipe hot-melt joints; S2, normalizing the microwave reflection signal obtained in step S1 to ensure consistency of subsequent processing; Normalization processing limits the signal amplitude to the range of [0,1], with the purpose of standardizing the microwave reflection signals obtained under different detection conditions to eliminate non-target fluctuations caused by external factors such as the sensitivity of the signal acquisition equipment, the detection environment, or the change in the position of the microwave waveguide probe; This normalization step plays an important role in signal processing, especially in the analysis of common defects in hot-melt welding of polyethylene pipes. Standardized signals can make feature extraction and subsequent defect detection processing steps more stable and accurate. In addition, normalization can also improve the comparability of microwave reflection signals between different detection batches, making microwave S-based 11 The parameter detection algorithm can maintain robustness under various environmental conditions, thereby improving the performance and reliability of the overall detection system; S3, slicing the signal normalized in step S2 according to a preset slice width and step length to generate multiple signal segments of the same length; and labeling each slice according to whether the slice signal contains a defective part; The pre-processed microwave reflection signal is sliced ​​according to the preset slice width and step length, and the continuous one-dimensional signal sequence is divided into multiple fixed-length signal segments. The step length is the same as the slice width to ensure that the generated slice lengths are consistent. Through the slice processing, the original signal can be refined into local segments, which is convenient for subsequent feature analysis and defect identification; Each slice signal is labeled according to whether it contains holes, slag inclusions, and unfused defects. Specifically, each signal segment is labeled as "0" or "1" by referring to the actual detection conditions and known defect locations. Supervised learning can effectively improve the detection accuracy of the microwave detection algorithm for three common defects by using labeled slice data, ensuring that the model has high robustness and accuracy under different working conditions. S4, extracting time domain and frequency domain features for each slice signal and complete signal, establishing a weld defect feature signal sample data set, and dividing all labeled samples into a training set and a test set; When analyzing the slice signal, a multidimensional feature extraction method in the time domain and frequency domain is used to fully capture the potential defect information in the signal; these features include the maximum value, minimum value, amplitude, energy, peak value, rise time, root mean square, variance, standard deviation, kurtosis, skewness, peak factor, waveform factor, pulse factor, margin factor features in the signal time domain, and the average power spectrum density, peak frequency, harmonics, correlation, phase response, autocorrelation, and cross-correlation features in the frequency domain; the purpose of feature extraction is to convert the one-dimensional microwave reflection signal into a high-dimensional feature vector through mathematical calculation, so as to construct a signal sample data set containing rich defect feature information { , },in Representative i The feature vector of the signal segment, =0 means the signal segment corresponds to the normal area, =1 indicates that the signal segment contains a defect area; then the data set is divided into a training set and a test set to further improve the learning and prediction accuracy of the model; By extracting these time domain and frequency domain features, a signal sample dataset containing rich defect information can be constructed; this dataset not only contains the physical properties of each signal slice, but also reflects the possible types of holes, slag inclusions, and unfused defects in the signal; based on this dataset, it can effectively support subsequent machine learning algorithms for high-precision defect detection; the establishment of the signal sample dataset ensures the efficiency of data processing and the accuracy of model training, making this method widely applicable and robust in actual industrial applications; S5. Based on the signal sample data set prepared in step S4, a defect detection classification model is constructed using a support vector classifier (SVC); the classifier performs supervised learning through the labeled training data, and the samples in the training set are classified according to their corresponding label information, and the labels include normal areas and areas containing defects; The SVC is trained using the signal sample data set generated in step S4 to construct a classification model for welding defect detection; the support vector classifier is a supervised learning algorithm commonly used for binary classification tasks, and can effectively detect holes, slag inclusions and unfused defects in polyethylene pipe hot-melt joints; in this process, the classifier performs supervised learning by using pre-labeled training data, so that the model can classify the slice signal according to the input signal characteristics; SVC will construct an optimal hyperplane in, is the weight of the feature vector, b is a bias term; for the binary classification problem of defect judgment, find a hyperplane that meets the following conditions: By optimizing the problem, we find the hyperplane that minimizes the classification error while maximizing the classification interval. We also use the kernel function to map the feature space to a higher dimension, so that the originally inseparable signal fragments become separable in the high-dimensional space. During the training process, the support vector classifier adjusts the model parameters by minimizing the classification loss function. Based on the given labeled training data, the model is optimized through gradual learning, so that it can be based on the input feature vector To predict the category of the signal segment, i.e., normal region or defective region; The model is evaluated through classification evaluation indicators, and its generalization ability is verified on the test set; the trained SVC is finally used to automatically detect new weld signals. The input signal will be predicted by the model to determine whether it contains defects, thereby achieving automated and intelligent defect detection; S6. The features extracted in step S4 are selected using a feature importance method. The feature selection method uses a Mann-Whitney U test for preliminary screening, and then the unimportant features are removed based on the SVC constructed in step S5 and a recursive feature elimination (RFE) method is used. The features are ranked according to their contribution to the model classification, and the top 10 important features are retained; Feature selection is divided into two stages: first, the Mann-Whitney U test is used to preliminarily screen the features; then, RFE is used to further select and finally retain the top ten features that contribute most to the classifier. The retained features are: maximum value, minimum value, amplitude, energy, peak value, root mean square, standard deviation, pulse factor, margin factor, and average power spectral density; The Mann-Whitney U test is a nonparametric statistical method used to compare the difference in characteristics between defect signals and normal signals; given a characteristic Divide them into two groups: defective sample group and normal sample group, and calculate the U value: in, is the number of defective samples, is the number of normal samples, is the rank sum of the defect sample group; if the feature The distribution difference between the two groups is significant, which means that the feature has a greater ability to distinguish the classification task, and vice versa. By performing the Mann-Whitney U test on each feature, we select p Features with values ​​less than the threshold of 0.01 are considered important candidate features and are retained for the next step of recursive feature elimination. p The value is an important statistic used to determine whether there is a significant difference in the distribution of features between normal signals and defective signals; The recursive feature elimination (RFE) algorithm is an algorithm that recursively trains the model and removes the least important features; based on the support vector classifier training and calculates the weight of each feature , the importance of each feature is measured according to the absolute value of the weight. Features with smaller absolute weights are considered to contribute less to the classification results. The feature importance function is defined as: Remove features with smaller weights and repeat the training process until the ten most important features are retained; and recursively perform feature selection through the following optimization process: in C is a regularization parameter that controls the influence of weights in the feature selection process; The recursive elimination method can gradually filter out the features that contribute most to the classification task and remove redundant or noisy features. Finally, the top ten most important features will be retained in this step for subsequent classification model training and defect detection tasks. S7. Use the Bayesian optimization algorithm to optimize the hyperparameters of the constructed SVC to improve the generalization ability of the classifier and the accuracy of defect detection; The optimization algorithm uses the Bayesian optimization algorithm to optimize the hyperparameters of the support vector classifier (SVC), and the optimization range includes regularization parameters, kernel function types and kernel parameters. Bayesian optimization uses the test set accuracy as the objective function, updates the proxy model, continuously adjusts the hyperparameter combination, conducts SVC training and verification, and finally determines the hyperparameter configuration that optimizes the objective function for building a defect classifier; S8. Design a polyethylene pipe hot-melt welded joint defect detection system, use the SVC model optimized in step S7 to analyze the microwave scanning reflection signal in real time, detect and locate defects, and support fast and accurate defect detection.

[0012] A microwave-based polyethylene pipe hot-melt welding defect detection system includes the following modules: A microwave signal import module is used to import the collected microwave reflection signals of the microwave waveguide scanning polyethylene pipe hot-melt welding joint with an operating frequency range of 12.4 GHz to 18 GHz, and perform normalization; The signal slicing module is used to slice the normalized preprocessed signal according to the preset slice width and step length to generate signal segments. The step length is the same as the slice width by default. The feature extraction module is used to extract the aforementioned selected time domain features and frequency domain features from the signal segment, providing a rich feature basis for subsequent feature selection and defect classification; The defect recognition module is used to input the features extracted from the microwave reflection signal to be detected into the trained and optimized SVC model, automatically detect the weld defects of the polyethylene pipe hot-melt joint, and obtain the detection results of the signal to be detected; The information display module is used to visualize the detection results obtained by the detection module, display the defects and their locations in the form of pictures and output them on the display interface.

[0013] In order to verify the effectiveness of the Bayesian hyperparameter optimization algorithm in the present invention in the detection of polyethylene pipe hot-melt welding defects, a variety of hyperparameter optimization methods were selected for comparison, including grid search algorithm, random search algorithm and Bayesian optimization algorithm. In the experiment, all algorithms used the same training set and test set, and the experimental parameters remained consistent. In order to evaluate the detection performance of the present invention, the training set accuracy, test set accuracy, F1 score, precision and recall were selected as evaluation indicators for the comparative experiment.

[0014] Figure 4 The defect recognition effect diagram of the method of the present invention and the comparative method in actual weld inspection is shown. From the actual defect position and algorithm detection results marked in the figure, it can be seen that the model after Bayesian hyperparameter optimization is significantly better than the other two models in terms of detection rate and false detection rate. Table 1 gives a comparison of the performance indicators of each method. The method of the present invention is superior to the comparative method in terms of training set accuracy, F1 score and precision index. Compared with grid parameter optimization, the method of the present invention is 5.349% higher in test accuracy, 2.590% higher in F1 score, and 6.194% higher in precision score; compared with random search optimization, the training accuracy of the method of the present invention is 5.127% higher, the F1 score is 6.604% higher, and the precision score is 1.524% higher, but the test accuracy is slightly lower by 0.233%; although random search is better than grid search, the overall performance is still not as good as the method of the present invention; therefore, in terms of comprehensive model performance, the method of the present invention performs best.

[0015] Table 1 Comparison of performance of different hyperparameter optimization model algorithms Evaluation index method Training set accuracy Test set accuracy F1 score Accuracy Recall Mesh parameter optimization 82.320% 81.627% 89.780% 81.743% 99.569% Random search optimization 92.643% 87.209% 85.766% 86.413% 87.209% Bayesian Optimization 97.770% 86.976% 92.370% 87.937% 97.370% The above experimental results fully demonstrate that the method of the present invention is more accurate in detecting hot-melt welding defects of polyethylene pipes, provides technical support for accurately obtaining the quality grade and safe maintenance of polyethylene pipes, and has good application prospects.

[0016] The above is an explanation of the method and system of the present invention, and researchers in the field can reproduce the present invention through these explanations. Based on the content of the present invention, other embodiments and results obtained by researchers without creative work should fall within the scope of protection of the present invention.

Claims

1. A microwave-based polyethylene pipe hot-melt welding defect detection system, characterized in that: The following steps are involved: S1. Use a microwave probe to scan the detection area of ​​the polyethylene pipe hot-melt welding joint and collect the corresponding microwave reflection signal; S2, normalizing the microwave reflection signal obtained in step S1 to ensure consistency of subsequent processing; S3, slicing the signal normalized in step S2 according to a preset slice width and step length to generate a plurality of signal segments of the same length; Each slice is labeled according to whether the slice signal contains a defective part; S4, extracting time domain and frequency domain features for each slice signal and complete signal, establishing a weld defect feature signal sample data set, and dividing all labeled samples into a training set and a test set; S5. Based on the signal sample data set prepared in step S4, a defect detection classification model is constructed using a support vector classifier (SVC); the classifier performs supervised learning through the labeled training data, and the samples in the training set are classified according to their corresponding label information, and the labels include normal areas and areas containing defects; S6. The features extracted in step S4 are selected using a feature importance method. The feature selection method uses a Mann-Whitney U test for preliminary screening, and then the unimportant features are removed based on the SVC constructed in step S5 and a recursive feature elimination (RFE) method is used. The features are ranked according to their contribution to the model classification, and the top 10 important features are retained; S7. Use the Bayesian optimization algorithm to optimize the hyperparameters of the constructed SVC to improve the generalization ability of the classifier and the accuracy of defect detection; S8. Design a polyethylene pipe hot-melt welded joint defect detection system, use the SVC model optimized in step S7 to analyze the microwave scanning reflection signal in real time, detect and locate defects, and support fast and accurate defect detection.

2. A microwave-based polyethylene pipe hot-melt welding defect detection system according to claim 1, characterized in that: Step S3 performs slicing processing according to a preset slice width and step length, wherein the slice width is a fixed value and the step length is the same as the slice width by default, so as to divide the continuous one-dimensional signal sequence into a plurality of signal segments of the same length; Each slice signal is annotated according to whether it contains holes, slag inclusions, and unfused defects, providing labels for subsequent supervised learning tasks.

3. The microwave-based polyethylene pipe hot-melt welding defect detection system according to claim 1 is characterized in that: Step S4 extracts time domain and frequency domain features from each slice signal and the complete signal; the features include maximum value, minimum value, amplitude, energy, peak value, rise time, root mean square, variance, standard deviation, kurtosis, skewness, peak factor, waveform factor, pulse factor, margin factor in the time domain, and average power spectrum density, peak frequency, harmonics, average correlation, phase response, autocorrelation, cross-correlation in the frequency domain, a total of 22 features, to construct a sample data set containing rich defect feature information { , },in Representative i The feature vector of the signal segment, =0 means the signal segment corresponds to the normal area, =1 indicates that the signal segment contains a defect area; then the data set is divided into a training set and a test set.

4. A microwave-based polyethylene pipe hot-melt welding defect detection system according to claim 1, characterized in that: Based on the feature vector extracted in step S4, feature selection is performed, and the feature selection is divided into two stages: first, the features are preliminarily screened using the Mann-Whitney U test; then, RFE is used for further screening, and finally the features are ranked according to their contribution to the model classification, and the top ten important features are retained. The retained features are: maximum value, minimum value, amplitude, energy, peak value, root mean square, standard deviation, pulse factor, margin factor, and average power spectral density.

5. The microwave-based polyethylene pipe hot-melt welding defect detection system according to claim 1 is characterized in that The optimization algorithm described in step S7 uses the Bayesian optimization algorithm to optimize the hyperparameters of SVC; the scope of hyperparameter optimization includes regularization parameters, kernel function types and kernel parameters; Bayesian optimization defines the accuracy of the test set as the objective function, continuously updates the proxy model, and performs SVC training and verification on each selected hyperparameter combination, and finally finds the optimal hyperparameter combination that minimizes the objective function as the final defect classifier.

6. The microwave-based polyethylene pipe hot-melt welding defect detection system according to claim 1 is characterized in that: The detection system in step S8 includes the following modules: A microwave signal import module is used to import the collected microwave reflection signals of the microwave waveguide scanning polyethylene pipe hot-melt welding joint with an operating frequency range of 12.4 GHz to 18 GHz, and perform normalization; The signal slicing module is used to slice the normalized preprocessed signal according to the preset slice width and step length to generate signal segments. The step length is the same as the slice width by default. A feature extraction module is used to extract the aforementioned selected time domain features and frequency domain features from the signal segment to provide a feature vector for subsequent defect classification; The defect recognition module is used to input the features extracted from the microwave reflection signal to be detected into the trained and optimized SVC model, automatically detect the weld defects of the polyethylene pipe hot-melt joint, and obtain the detection results of the signal to be detected; The information display module is used to visualize the detection results obtained by the detection module, display the defects and their locations in the form of pictures and output them on the display interface.

Images