Ultra-long small-diameter thin-wall heat absorber pipeline damage type identification method based on XGBoost algorithm
Through the XGBoost algorithm method, a strong robust classification recognition model is established, which solves the complexity of the damage type recognition of ultra-long small-diameter thin-wall heat-absorbing pipes, and improves detection accuracy and efficiency.
Patent Information
- Application Number
- CN202411523176.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-29
- Publication Date
- 2025-06-13
AI Technical Summary
The acoustic impedance difference and complex structure of ultra-long small diameter thin-walled heat absorber tubes lead to complex ultrasonic waveguide detection signals, low echo signal-to-noise ratio, making it difficult to accurately identify and locate damage types.
Using the XGBoost algorithm method, a strongly robust XGBoost classification recognition model is established by collecting and storing ultrasonic guided echo waveform data, which is used to identify the damage type of ultra-long small-diameter thin-walled heat absorbing pipe.
It improves the accuracy and efficiency of the judgment of damage types of thin-walled heat pipes in ultra-long diameters, ensures the reliability of detection results, and supports preventive maintenance and fault management.
Smart Images
Figure HDA0005108494200000011 
Figure HDA0005108494200000012
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of nondestructive testing, and particularly relates to a method for identifying damage types of ultra-long small-diameter thin-walled heat absorber pipes based on the XGBoost algorithm. Background Technique
[0002] With the transformation of the national energy strategy and the promotion of the construction of a new power system, the solar thermal energy storage tower solar system has developed rapidly. However, the ultra-long small-diameter thin-walled structure of the tower solar heat absorber pipe and its multi-layer heterogeneous characteristics result in differences in acoustic impedance, causing reflection, refraction, and mode conversion of ultrasonic guided waves during propagation, increasing the complexity of defect detection. The slender pipe structure leads to a low signal-to-noise ratio of echoes, and factors such as environmental noise, equipment vibration, and thermal expansion and contraction further interfere with the signals, making it difficult to identify effective information. Ultrasonic guided wave detection based on machine learning algorithms can effectively process complex data and automatically distinguish different types of defects, such as external ablation and internal corrosion, thus realizing accurate damage identification and positioning. This kind of identification is crucial for ensuring the safe operation of the heat absorber pipe. Different damage types have different impacts on the stability and thermal efficiency of the system. By accurately identifying typical defects and damage forms, it can provide key data support for preventive maintenance and fault management, which is of great significance for further improving the safety and operation efficiency of the tower solar heat absorber and ensuring the stable operation of the new heat mass energy storage system. Summary of the Invention
[0003] The purpose of the present invention is to provide a method for identifying damage types of ultra-long small-diameter thin-walled heat absorber pipes based on the XGBoost algorithm to improve the accuracy of judging damage types of ultra-long small-diameter thin-walled heat absorber pipes.
[0004] The technical solution adopted by the present invention is a method for identifying damage types of ultra-long small-diameter thin-walled heat absorber pipes based on the XGBoost algorithm, which is specifically implemented according to the following steps:
[0005] S1: Collect ultrasonic guided wave echo waveform data corresponding to various typical damage forms of heat absorber pipes, and store these waveforms in the form of data to form a typical damage waveform database of ultra-long small-diameter thin-walled heat absorber pipes;
[0006] S2: Establish an initial XGBoost algorithm, and use the waveform data in the typical damage waveform database of ultra-long small-diameter thin-walled heat absorber pipes to train the parameters of the XGBoost algorithm to obtain a strongly robust XGBoost classification and recognition model;
[0007] S3: Import the typical damage waveform database of ultra-long small-diameter thin-walled heat absorber pipes and the strongly robust XGBoost classification and recognition model into the ultrasonic guided wave detection system, and install the ultrasonic guided wave detection system on the heat absorber pipe to be tested;
[0008] S4: Use the ultrasonic guided wave detection system to detect the to-be-detected tower solar heat absorption tube, generate an ultrasonic guided wave signal, and after processing this signal, match it with waveform data to extract the matched waveform data;
[0009] S5: Discretize the matched waveform data to obtain discrete waveform data;
[0010] S6: Feed the discrete waveform data into the XGBoost classification and recognition model with strong robustness to obtain the output result, and judge the damage of the ultra-long small-diameter thin-walled heat absorption tube according to the output result.
[0011] The feature of the present invention also lies in that S1 is specifically implemented according to the following steps:
[0012] S1.1: Use a flaw detector to collect waveforms corresponding to various heat absorption tube damages, and the various heat absorption tube damages include damage defect types such as no damage, external ablation, internal corrosion, pipe welds, pipe expansion, pipe blockage, etc.;
[0013] S1.2: Convert the collected waveforms into binary waveform data, store the converted waveform data, and form a damage waveform database for ultra-long small-diameter thin-walled heat absorption tubes.
[0014] S2 is specifically implemented according to the following steps:
[0015] S2.1: Establish an initial XGBoost algorithm based on Python software. The initial XGBoost model framework includes input feature data, weak classifiers (tree models), and output label results;
[0016] S2.2: Load and process the waveform data in the heat absorption tube damage waveform database based on Python software, extract various time-domain feature parameters, frequency-domain feature parameters, and information entropy feature parameters respectively, divide the eigenvalue and damage category label value, and map the waveform feature data to the [0,1] interval for normalization processing as the initial model input feature data;
[0017] S2.3: Set the initial hyperparameters of the XGBoost classification model, such as loss function, evaluation index, and number of classifications, and conduct the initial training of the model based on the training set data. The training cycle is dynamically adjusted according to the convergence of the model, and usually at least 8 rounds of iterations are performed;
[0018] S2.4: According to the model output result, conduct an interpretability analysis based on SHAP, calculate the SHAP value and perform quantitative sorting on the features, and use the high-contribution feature subset as the input of the XGBoost model for the final model training;
[0019] S2.5. Perform hyperparameter optimization of the XGBoost model, such as the number of iterations, learning rate, maximum tree depth, and base estimators. Use the learning curve method to initially locate important hyperparameters; establish a parameter space and adopt the enumeration grid search method to return the optimal hyperparameter values;
[0020] S2.6. During model training, use four metrics, namely accuracy, precision, recall, and F 1-score to evaluate the classification performance of the model and verify that the model has strong classification and recognition performance for different defect types. Obtain a highly robust XGBoost classification and recognition model, and save the model tree structure and hyperparameters for subsequent use.
[0021] In S3, the ultrasonic guided wave detection system includes a flexible magnetostrictive probe (1) for both transmitting and receiving, which are connected in sequence, an ultrasonic guided wave excitation module (2), a D / A converter (3), an FPGA main control system (4), an ARM controller (5), a USB communication circuit module (6), a host computer software (7), an A / D converter (8), and an ultrasonic guided wave echo receiving module (9). The ultra-long small-diameter thin-walled heat absorption tube damage waveform database and the constructed XGBoost algorithm model are uploaded to the ARM controller (5), and the flexible magnetostrictive probe (1) is installed at a convenient position on one side of the heat absorption tube tube screen to be detected;
[0022] The ultrasonic guided wave excitation module (2) includes a power amplifier circuit and an ultrasonic guided wave excitation circuit, and is connected to the FPGA main control system (4) through the D / A converter (3);
[0023] The ARM controller (5) has an FSMC interface, and the FPGA main control system (4) is connected to the ARM controller (5) through the FSMC interface; inside the FPGA main control system (4), there are an ultrasonic guided wave excitation control module, a data acquisition control module, a digital filtering and enhancement module, a channel switching module, etc.
[0024] S4 is specifically implemented according to the following steps:
[0025] S4.1. Set detection parameters through the host computer software (7), and send instructions to the ARM controller (5) through the connected USB communication circuit (6). The ARM controller (5) sends an excitation signal to the FPGA main control system (4). After being processed by the excitation control module of the FPGA main control system (4), the signal is transmitted to the flexible magnetostrictive probe (1) through the action of the ultrasonic guided wave excitation module (2) and the power amplification module, prompting the flexible magnetostrictive probe (1) to generate an ultrasonic guided wave signal;
[0026] S4.2. The ultrasonic guided wave signal propagates in the heat absorption tube and generates an echo that returns to the flexible magnetostrictive probe (1), and the echo signal is converted into a voltage signal. Subsequently, the voltage signal is subjected to signal conditioning through the limiting circuit, preamplifier circuit, and bandpass filter circuit of the ultrasonic guided wave echo receiving module (9). Finally, the voltage signal is subjected to analog-to-digital conversion by the A / D converter (8), and the data is stored in the data acquisition control module in the FPGA main control system (4).
[0027] S4.3. The data cached in the FPGA main control system (4) is transmitted to the ARM controller (5) using the SPI communication method. The ARM processor (5) analyzes and processes the incoming data, matches the incoming data with the waveform data, and extracts the matching waveform data.
[0028] Specifically, S5 is to perform digital filtering processing on the matching waveform data through the FIR digital filtering module in the FPGA main control system (4) to convert it into discrete waveform data.
[0029] Specifically, S6 is implemented according to the following steps:
[0030] S6.1. After buffering the discrete waveform data through the recognition algorithm module in the FPGA main control system (4), it is transmitted through cross-clock domain processing to the strongly robust XGBoost classification and recognition model in the ARM controller (5).
[0031] S6.2. The strongly robust XGBoost classification and recognition model calculates the discrete waveform data and outputs the result, and judges the output result. The output result will be interpreted according to the following labels: 0 indicates no damage, 1 indicates external ablation, 2 indicates internal corrosion, 3 indicates weld defect, 4 indicates pipe expansion, 5 indicates pipe blockage. Finally, the damage detection result of the ultra-long small-diameter thin-walled heat absorption tube is obtained.
[0032] The beneficial effect of the present invention is that a method for identifying the damage type of an ultra-long small-diameter thin-walled heat absorber pipe based on the XGBoost algorithm of the present invention sends the collected discrete ultrasonic guided wave signals into the trained XGBoost algorithm to obtain the damage judgment result of the ultra-long small-diameter thin-walled heat absorption tube, so as to ensure that the damage condition of the heat absorption tube in the detection interval is detected in time, greatly improving the damage detection efficiency, detection accuracy, and detection reliability of the ultra-long small-diameter thin-walled heat absorption tube. Description of the Drawings
[0033] Figure 1 It is a schematic structural diagram of an ultrasonic guided wave detection system in a method for identifying the damage type of an ultra-long small-diameter thin-walled heat absorber pipe based on the XGBoost algorithm of the present invention.
[0034] In the figure, 1. Flexible magnetostrictive probe, 2. Ultrasonic guided wave excitation module, 3. D / A converter, 4. FPGA main control system, 5. ARM controller, 6. USB communication circuit, 7. Host computer software, 8. A / D converter, 9. Ultrasonic guided wave echo receiving module.
[0035] Figure 2 It is the flow chart for constructing the classification model in a method for identifying damage types of ultra-long small-diameter thin-walled heat absorber pipes based on the XGBoost algorithm of the present invention. Specific implementation manner
[0036] The present invention will be described in detail below in conjunction with the accompanying drawings and specific implementation manners.
[0037] A method for identifying damage types of ultra-long small-diameter thin-walled heat absorber pipes based on the XGBoost algorithm of the present invention is specifically implemented according to the following steps:
[0038] S1: Collect ultrasonic guided wave echo waveform data corresponding to various typical damage forms of heat absorption pipes, and store these waveforms in the form of data to form a typical damage waveform database of ultra-long small-diameter thin-walled heat absorption pipes;
[0039] S1.1: Use a flaw detector to collect waveforms corresponding to various damages of heat absorption pipes. The various damages of heat absorption pipes include damage defect types such as no damage, external ablation, internal corrosion, weld defects, pipe expansion, and pipe blockage;
[0040] S1.2: Convert the collected waveforms into binary waveform data, and store the converted waveform data to form a damage waveform database of ultra-long small-diameter thin-walled heat absorption pipes;
[0041] S2: Establish an initial XGBoost algorithm, and use the waveform data in the damage waveform database of ultra-long small-diameter thin-walled heat absorption pipes to train the parameters of the XGBoost algorithm to obtain a strongly robust XGBoost classification and recognition model;
[0042] S2.1: Establish an initial XGBoost algorithm based on Python software. The initial XGBoost model framework includes input feature data, weak classifiers (tree models), and output label results;
[0043] S2.2: Load and process the waveform data in the damage waveform database of heat absorption pipes based on Python software, extract various time-domain feature parameters, frequency-domain feature parameters, and information entropy feature parameters respectively, divide the eigenvalue and damage category label value, and map the waveform feature data to the [0,1] interval for normalization processing as the initial model input feature data;
[0044] S2.3. Set the initial hyperparameters of the XGBoost classification model, such as loss function, evaluation index, and number of classifications, and perform initial training on the model based on the training set data. The training cycle is dynamically adjusted according to the convergence of the model, and usually at least 8 rounds of iterations are performed.
[0045] S2.4. Perform SHAP-based interpretability analysis based on the model output results, calculate the SHAP value and quantify the features, and use the high-contribution feature subset as the input of the XGBoost model for the final model training;
[0046] S2.5. Optimize the hyperparameters of the XGBoost model, such as the number of iterations, learning rate, maximum tree depth, and base estimator. Use the learning curve method to preliminarily locate important hyperparameters; establish a parameter space and use the enumeration grid search method to return the best hyperparameter value results;
[0047] S2.6, model training uses accuracy, precision, recall and F 1-score The classification performance of the model is evaluated by four indicators, and the model is verified to have strong classification and recognition performance for all defect types. A strong and robust XGBoost classification and recognition model is obtained, and the model tree structure and hyperparameters are saved for subsequent use;
[0048] S3: Import the damage waveform database of ultra-long small-diameter thin-walled heat absorber tubes and the highly robust XGBoost classification and recognition model into the ultrasonic guided wave detection system, and install the ultrasonic guided wave detection system on the heat absorber tube to be tested;
[0049] The ultrasonic guided wave detection system in S3 includes a flexible magnetostrictive probe (1) with integrated transmitter and receiver, an ultrasonic guided wave excitation module (2), a D / A converter (3), an FPGA main control system (4) and an ARM controller (5), a USB communication circuit module (6), a host computer software (7), an A / D converter (8) and an ultrasonic guided wave echo receiving module (9) connected in sequence. The ultra-long small-diameter thin-walled heat-absorbing tube damage waveform database and the constructed XGBoost algorithm model will be uploaded to the ARM controller (5), and the flexible magnetostrictive probe (1) is installed at a convenient location on one side of the heat-absorbing tube panel to be detected;
[0050] The ultrasonic guided wave excitation module (2) comprises a power amplifier circuit and an ultrasonic guided wave excitation circuit, and is connected to an FPGA main control system (4) via a D / A converter (3);
[0051] The ARM controller (5) has an FSMC interface, and the FPGA main control system (4) is connected to the ARM controller (5) via the FSMC interface; the FPGA main control system (4) is internally provided with an ultrasonic guided wave excitation control module, a data acquisition control module, a digital filter enhancement module, a channel switching module, etc.;
[0052] S4: Use the ultrasonic guided wave detection system to detect the tower solar heat absorption tube to be measured, generate ultrasonic guided wave signals, and after processing these signals, match them with waveform data to extract the matching waveform data;
[0053] S4.1: Set detection parameters through the host computer software (7), and send instructions to the ARM controller (5) through the connected USB communication circuit (6). The ARM controller (5) sends an excitation signal to the FPGA main control system (4). After being processed by the excitation control module of the FPGA main control system (4), the signal is transmitted to the flexible magnetostrictive probe (1) through the action of the ultrasonic guided wave excitation module (2) and the power amplification module, prompting the flexible magnetostrictive probe (1) to generate ultrasonic guided wave signals;
[0054] S4.2: The ultrasonic guided wave signals propagate in the heat absorption tube and generate echoes that return to the flexible magnetostrictive probe (1), and the echo signals are converted into voltage signals. Subsequently, the voltage signals are sequentially conditioned through the limiting circuit, pre-amplification circuit, and band-pass filter circuit of the ultrasonic guided wave echo receiving module (9). Finally, the voltage signals are subjected to analog-to-digital conversion through the A / D converter (8), and the data is stored in the data acquisition control module in the FPGA main control system (4);
[0055] S4.3: Use the SPI communication method to transfer the data cached in the FPGA main control system (4) to the ARM controller (5). The ARM processor (5) analyzes and processes the incoming data, matches the incoming data with the waveform data, and extracts the matching waveform data;
[0056] S5: Discretize the matching waveform data to obtain discrete waveform data;
[0057] Specifically, S5 is to perform digital filtering processing on the matching waveform data through the FIR digital filtering module in the FPGA main control system (4) to convert it into discrete waveform data;
[0058] S6: Send the discrete waveform data into the strongly robust XGBoost classification and recognition model to obtain the output result, and judge the damage of the ultra-long small-diameter thin-walled heat absorption tube according to the output result;
[0059] S6.1: After buffering the discrete waveform data through the recognition algorithm module in the FPGA main control system (4), pass it into the strongly robust XGBoost classification and recognition model in the ARM controller (5) through cross-clock domain processing;
[0060] S6.2. The XGBoost classification and recognition model with strong robustness calculates the output results for the discrete waveform data and judges the output results. The output results will be interpreted according to the following labels: 0 indicates no damage, 1 indicates external ablation, 2 indicates internal corrosion, 3 indicates weld defects, 4 indicates pipe expansion, and 5 indicates pipe blockage. Finally, the damage detection results of the ultra-long small-diameter thin-walled heat-absorbing pipe are obtained.
[0061] In the method for identifying the damage types of ultra-long small-diameter thin-walled heat-absorbing pipe based on the XGBoost algorithm of the present invention, the functions of the main steps are as follows:
[0062] In S2.2, mapping the data to the interval [0,1] for normalization is to accelerate the convergence of the training network and avoid the loss falling into local minima.
[0063] In S2.3, training the data in the damage waveform database of the ultra-long small-diameter thin-walled heat-absorbing pipe established in S1 for no less than eight cycles. In the case of limited samples, an appropriate training cycle can improve the accuracy of the XGBoost algorithm calculation.
[0064] In S2.4, the interpretable analysis method based on SHAP is used to select the high-contribution feature subset, and high machine learning classification and recognition performance can be obtained with only fewer features, greatly saving the calculation cost.
[0065] In S2.5, the main function of hyperparameter tuning is to find out the hyperparameters that have an important impact on the model, so as to further improve the model performance, control the model complexity and enhance the robustness of the model.
[0066] In S4.2, the function of the amplitude limiting circuit is to control the amplitude of the excitation pulse, thereby affecting the energy of the ultrasonic wave. The function of the preamplifier circuit is to amplify the received ultrasonic guided wave signal without distortion to avoid the ultrasonic guided wave signal being likely to be submerged in noise and interference due to the intervention of noise. The band-pass filter circuit can improve the signal-to-noise ratio of the ultrasonic guided wave signal, and its function is to filter the ultrasonic guided wave signal after preamplification to reduce noise and interference.
[0067] Through the above method, the method for identifying the damage types of ultra-long small-diameter thin-walled heat-absorbing pipe based on the XGBoost algorithm of the present invention sends the collected discrete ultrasonic guided wave signals into the trained XGBoost algorithm to obtain the damage judgment results of the heat-absorbing pipe, so as to ensure that the damage conditions of the heat-absorbing pipe in the detection interval are detected in time, greatly improving the damage detection efficiency, detection accuracy and detection reliability of the ultra-long small-diameter thin-walled heat-absorbing pipe.
[0068] Embodiment
[0069] This embodiment provides a method for identifying damage types of ultra-long small-diameter thin-walled heat absorber pipes based on the XGBoost algorithm, which is specifically implemented according to the following steps:
[0070] S1: Collect ultrasonic guided wave echo waveform data corresponding to various typical damage forms of heat absorption pipes, and store these waveforms in the form of data to form a typical damage waveform database for ultra-long small-diameter thin-walled heat absorption pipes;
[0071] S1.1: Use a flaw detector to collect waveforms corresponding to various damages of heat absorption pipes. The various damages of heat absorption pipes include damage defect types such as no damage, external ablation, internal corrosion, pipe welds, pipe expansion, and pipe blockage;
[0072] S1.2: Convert the collected waveforms into binary waveform data, and store the converted waveform data to form a damage waveform database for ultra-long small-diameter thin-walled heat absorption pipes;
[0073] S2: Establish an initial XGBoost algorithm, and use the waveform data in the damage waveform database of ultra-long small-diameter thin-walled heat absorption pipes to train the parameters of the XGBoost algorithm to obtain a strongly robust XGBoost classification and recognition model;
[0074] S2.1: Establish an initial XGBoost algorithm based on Python software. The initial XGBoost model framework includes input feature data, weak classifiers (tree models), and output label results;
[0075] S2.2: Load and process the waveform data in the damage waveform database of heat absorption pipes based on Python software, extract 16 time-domain feature parameters (parameters describing waveform features: peak value, duration, rise time, fall time, skewness, and kurtosis. Parameters describing signal amplitude features: mean value, absolute mean value, standard deviation, root mean square value, root mean square amplitude, peak factor, pulse factor, waveform factor, margin factor, and energy), 16 frequency-domain feature parameters (parameters characterizing frequency distribution and concentration: spectral peak value, center frequency, frequency variance, mean square frequency. Parameters characterizing the energy distribution of the amplitude spectrum: skewness, kurtosis, mean value, absolute mean value, standard deviation, root mean square value, root mean square amplitude, peak factor, pulse factor, waveform factor, margin factor, and energy), and 3 information entropy feature parameters (singular spectrum entropy, power spectrum entropy, wavelet packet energy entropy), divide the eigenvalue and defect label value, and map the waveform data to the [0,1] interval for normalization processing as the model input feature data;
[0076] S2.3. Set the initial hyperparameters of the XGBoost classification model, such as the loss function, evaluation metrics, and number of classes. The feature data is divided into a training set, a validation set, and a test set in the ratio of 8:1:1. 16 time-domain features, 16 frequency-domain features, and 3 information entropy features are extracted from each feature data, totaling 35 features, and a 35×1000 feature matrix is constructed. And based on the training set data, the model is initially trained, and the training cycle is dynamically adjusted according to the model convergence situation, and 8 rounds of iteration are performed;
[0077] S2.4. According to the model output results, perform SHAP-based interpretability analysis, calculate SHAP values, and quantitatively sort the features. Select the top 8 high-contribution features (waveform peak value, amplitude mean value, spectral peak value, peak factor, impulse factor, waveform factor, root mean square amplitude, singular spectrum entropy), form a feature subset, and use the high-contribution feature subset as the input of the final XGBoost model for subsequent classification model training;
[0078] S2.5. Perform hyperparameter optimization of the XGBoost model, such as the number of iterations, learning rate, maximum tree depth, and base estimator; use the method of learning curves to initially locate important hyperparameters; establish a parameter space, and use the enumeration grid search method to return the optimal hyperparameter values; the hyperparameter tuning results are: the number of iterations (n_estimators = 340), the learning rate (learning_rate = 0.17), the maximum tree depth (max_depth = 6.4), and the base estimator (Booster = "gbtree");
[0079] S2.6. The model training uses 4 metrics, such as accuracy, precision, recall, and F 1-score to evaluate the classification performance of the model, and verify that the model has strong classification and recognition performance for different damage categories. Obtain a strongly robust XGBoost classification and recognition model, and save the model tree structure and hyperparameters for subsequent use;
[0080] S3: Import the ultra-long small-diameter thin-walled heat-absorbing tube damage waveform database and the strongly robust XGBoost classification and recognition model into the ultrasonic guided wave detection system, and install the ultrasonic guided wave detection system on the heat-absorbing tube to be measured;
[0081] In S3, the ultrasonic guided wave detection system includes a flexible magnetostrictive probe (1) for both transmitting and receiving, which are connected in sequence, an ultrasonic guided wave excitation module (2), a D / A converter (3), an FPGA main control system (4), and an ARM controller (5). The USB communication circuit module (6), the host computer software (7), an A / D converter (8), and the ultrasonic guided wave echo receiving module (9) are also included. The damage waveform database of the ultra-long small-diameter thin-walled heat absorption tube and the constructed XGBoost algorithm model are uploaded to the ARM controller (5), and the flexible magnetostrictive probe (1) is installed at a convenient position on one side of the heat absorption tube screen to be detected;
[0082] The ultrasonic guided wave excitation module (2) includes a power amplification circuit and an ultrasonic guided wave excitation circuit, and is connected to the FPGA main control system (4) through the D / A converter (3);
[0083] The ARM controller (5) has an FSMC interface, and the FPGA main control system (4) is connected to the ARM controller (5) through this interface. The ultrasonic guided wave excitation control module, data acquisition control module, digital filtering enhancement module, channel switching module, etc. are provided inside the FPGA main control system (4);
[0084] S4: Use the ultrasonic guided wave detection system to detect the tower-type solar heat absorption tube to be measured, generate an ultrasonic guided wave signal, and after processing this signal, match it with the waveform data to extract the matching waveform data;
[0085] S4.1: Set the detection parameters through the host computer software (7), and send instructions to the ARM controller (5) through the connected USB communication circuit (6). The ARM controller (5) sends an excitation signal to the FPGA main control system (4). After being processed by the excitation control module of the FPGA main control system (4), the signal is transmitted to the flexible magnetostrictive probe (1) through the action of the ultrasonic guided wave excitation module (2) and the power amplification module, prompting the flexible magnetostrictive probe (1) to generate an ultrasonic guided wave signal;
[0086] S4.2: The ultrasonic guided wave signal propagates in the heat absorption tube and generates an echo that returns to the flexible magnetostrictive probe (1), and the echo signal is converted into a voltage signal. Subsequently, the voltage signal is sequentially conditioned by the limiting circuit, preamplification circuit, and band-pass filter circuit of the ultrasonic guided wave echo receiving module (9). Finally, the voltage signal is subjected to analog-to-digital conversion through the A / D converter (8), and the data is stored in the data acquisition control module in the FPGA main control system (4);
[0087] S4.3. Transfer the data cached in the FPGA main control system (4) to the ARM controller (5) by using the SPI communication method. Analyze and process the incoming data through the ARM processor (5), match the incoming data with the waveform data, and extract the matched waveform data.
[0088] S5: Discretize the matched waveform data to obtain discrete waveform data.
[0089] Specifically, S5 is to perform digital filtering processing on the matched waveform data through the FIR digital filtering module in the FPGA main control system (4) to convert it into discrete waveform data.
[0090] S6: Send the discrete waveform data into the strongly robust XGBoost classification and recognition model to obtain the output result, and judge the damage of the ultra-long small-diameter thin-walled heat-absorbing tube according to the output result.
[0091] S6.1. Buffer the discrete waveform data through the recognition algorithm module in the FPGA main control system (4), and then transfer it to the strongly robust XGBoost classification and recognition model in the ARM controller (5) through cross-clock domain processing.
[0092] S6.2. If the output result of the strongly robust XGBoost classification and recognition model after calculating the discrete waveform data is 0, the final damage detection result of the ultra-long small-diameter thin-walled heat-absorbing tube is that the heat-absorbing tube is not damaged.
Claims
1. A method for identifying damage types of ultra-long, small-diameter, thin-walled heat absorber pipelines based on the XGBoost algorithm, characterized in that: Follow the steps below to implement it: S1: Collect ultrasonic guided wave echo waveform data corresponding to typical damage forms of various heat absorbing tubes, and store these waveforms in data form to form a typical damage waveform database of ultra-long, small-diameter, thin-walled heat absorbing tubes; S2: Establish the initial XGBoost algorithm, and use the waveform data of the ultra-long small-diameter thin-walled heat-absorbing tube damage waveform database to train the parameters of the XGBoost algorithm to obtain a highly robust XGBoost classification and recognition model; S3: Import the damage waveform database of ultra-long small-diameter thin-walled heat absorber tubes and the highly robust XGBoost classification and recognition model into the ultrasonic guided wave detection system, and install the ultrasonic guided wave detection system on the heat absorber tube to be tested; The ultrasonic guided wave detection system is composed of a flexible magnetostrictive probe (1), an ultrasonic guided wave excitation module (2), a D / A converter (3), an FPGA main control system (4), an ARM controller (5), a USB communication circuit module (6), a host computer software (7), an A / D converter (8) and an ultrasonic guided wave echo receiving module (9) connected in sequence. The ultra-long small-diameter thin-walled heat-absorbing tube damage waveform database and the constructed XGBoost algorithm model will be uploaded to the ARM controller (5), and the flexible magnetostrictive probe (1) is installed at a convenient location on one side of the heat-absorbing tube panel to be detected; The ultrasonic guided wave excitation module (2) comprises a power amplifier circuit and an ultrasonic guided wave excitation circuit, and is connected to an FPGA main control system (4) via a D / A converter (3); The ARM controller (5) has an FSMC interface, and the FPGA main control system (4) is connected to the ARM controller (5) via the interface. The FPGA main control system (4) is internally provided with an ultrasonic guided wave excitation control module, a data acquisition control module, a digital filter enhancement module, a channel switching module, etc.; S4: using the ultrasonic guided wave detection system to detect the tower-type solar thermal absorber tube to be detected, generating an ultrasonic guided wave signal, and matching the signal with the waveform data after processing to extract the matching waveform data; The implementation of S4 is carried out according to the following process: S4.1, setting detection parameters through the host computer software (7), and sending instructions to the ARM controller (5) through the USB communication circuit (6) connected thereto, the ARM controller (5) sends an excitation signal to the FPGA main control system (4), the signal is processed by the excitation control module of the FPGA main control system (4), and then transmitted to the flexible magnetostrictive probe (1) through the action of the ultrasonic guided wave excitation module (2) and the power amplification module, so as to prompt the flexible magnetostrictive probe (1) to generate an ultrasonic guided wave signal; S4.2, the ultrasonic guided wave signal propagates in the heat absorbing tube and generates an echo that returns to the flexible magnetostrictive probe (1), and the echo signal is converted into a voltage signal, and then the voltage signal is sequentially conditioned by the limiting circuit, preamplifier circuit and bandpass filter circuit of the ultrasonic guided wave echo receiving module (9). Finally, the voltage signal is converted into digital by the A / D converter (8), and the data is stored in the data acquisition control module in the FPGA main control system (4); S4.3, using the SPI communication method to transmit the data cached in the FPGA main control system (4) to the ARM controller (5), analyzing and processing the incoming data through the ARM processor (5), matching the incoming data with the waveform data, and extracting the matching waveform data; S5: performing discrete processing on the matching waveform data to obtain discrete waveform data; S6: The discrete waveform data is sent to the robust XGBoost classification and recognition model to obtain the output results, and the damage of the ultra-long small-diameter thin-walled heat absorption tube is judged based on the output results.
2. The damage type identification method for ultra-long small-diameter thin-walled heat absorber pipeline based on XGBoost algorithm according to claim 1 is characterized in that: The S1 specifically includes the following implementation steps: S1.
1. Use a flaw detector to collect waveform data corresponding to various types of heat absorption tube damage. The damage types include no damage, external ablation, internal corrosion, weld defects, pipeline expansion, pipeline blockage and other damage defect types; S1.
2. Convert the collected waveform into binary format, store the converted waveform data, and construct a waveform database of damage of ultra-long small-diameter thin-walled heat-absorbing tubes.
3. The damage type identification method for ultra-long small-diameter thin-walled heat absorber pipeline based on XGBoost algorithm according to claim 1 is characterized in that: The S2 specifically includes the following implementation steps: S2.
1. Establish the initial XGBoost algorithm based on Python software. The initial XGBoost model framework includes input feature data, weak classifier (tree model) and output label results; S2.2, based on Python software, load and process the waveform data in the heat absorber damage waveform database, extract a variety of time domain characteristic parameters, frequency domain characteristic parameters and information entropy characteristic parameters, divide the characteristic values and damage category label values, and map the waveform characteristic data to the [0,1] interval for normalization processing as the initial model input characteristic data; S2.
3. Set the initial hyperparameters of the XGBoost classification model, such as loss function, evaluation index, and number of classifications, and perform initial training on the model based on the training set data. The training cycle is dynamically adjusted according to the convergence of the model, and usually at least 8 rounds of iterations are performed. S2.
4. Perform SHAP-based interpretability analysis based on the model output results, calculate the SHAP value and quantify the features, and use the high-contribution feature subset as the input of the XGBoost model for the final model training; S2.
5. Optimize the hyperparameters of the XGBoost model, such as the number of iterations, learning rate, maximum tree depth, and base estimator. Use the learning curve method to preliminarily locate important hyperparameters; establish a parameter space and use the enumeration grid search method to return the best hyperparameter value results; S2.6, model training uses accuracy, precision, recall and F 1-score The classification performance of the model was evaluated by four indicators, and the model was verified to have strong classification and recognition performance for all defect types. A highly robust XGBoost classification and recognition model was obtained, and the model tree structure and hyperparameters were saved for subsequent use.
4. The damage type identification method for ultra-long small-diameter thin-walled heat absorber pipeline based on XGBoost algorithm according to claim 1 is characterized in that: S5 specifically includes performing digital filtering processing on the matching waveform data through a FIR digital filtering module in the FPGA main control system (4) to convert the matching waveform data into discrete waveform data.
5. The damage type identification method for ultra-long small-diameter thin-walled heat absorber pipeline based on XGBoost algorithm according to claim 1 is characterized in that: The S6 specifically includes the following implementation steps: S6.1, after the discrete waveform data is buffered by the recognition algorithm module in the FPGA main control system (4), it is transmitted to the robust XGBoost classification recognition model in the ARM controller (5) through cross-clock threshold processing; S6.2, the robust XGBoost classification recognition model calculates the discrete waveform data and outputs the results, and judges the output results. The output results will be interpreted according to the following labels: 0 means no damage, 1 means external ablation, 2 means internal corrosion, 3 means weld defect, 4 means pipe expansion, and 5 means pipe blockage. Finally, the damage detection results of the ultra-long small diameter thin-walled heat absorber tube are obtained.