A method and system for identifying weld porosity defects in steel bridge decks

Abstract

This invention discloses a method and system for identifying porosity defects in steel bridge deck welding. The system includes a signal acquisition module, a signal processing module, a signal classification module, an initial classification module, an intermediate classification module, a target classification module, a final classification module, and a result output module. The signal acquisition module collects multiple acoustic emission signals, including those with and without porosity defects, from the test steel bridge deck, and the acoustic emission signals collected from the test steel bridge deck constitute the source domain dataset. Multiple acoustic emission signals are also collected from the steel bridge deck to be tested, and the acoustic emission signals collected from the test steel bridge deck constitute the target domain dataset. The signal processing module processes the acoustic emission signals using an adaptive analytical mode and a Hilbert transform to obtain an adaptive marginal spectrum of the acoustic emission signals. This invention enables online monitoring of the welding process of steel bridge decks, accurately identifies porosity defects during welding, and has high identification efficiency.

Description

Technical Field

[0001] This invention relates to a defect identification method, and more particularly to a method and system for identifying weld porosity defects in steel bridge decks. Background Technology

[0002] Orthotropic steel bridge decks are complex thin-walled spatial structures welded together from panels, diaphragms, and longitudinal U-ribs. Due to their complex structural characteristics, they are prone to porosity defects during the welding process. Under the service loads of the bridge, these welding porosity defects may lead to stress concentration and fatigue crack propagation, seriously threatening the safe service performance of the bridge structure.

[0003] Existing technologies for detecting common weld porosity defects include visual inspection, ultrasonic testing, eddy current testing, infrared testing, and X-ray testing. Among these, visual inspection can only detect surface defects in welds and is difficult to identify minute defects, resulting in low overall accuracy. Ultrasonic, eddy current, infrared, and X-ray testing methods have the characteristics of penetration and non-destructive testing, but they have low detection efficiency and high labor costs. Summary of the Invention

[0004] Purpose of the invention: The first purpose of this invention is to provide a method for identifying porosity defects in steel bridge deck welding with high accuracy and high efficiency.

[0005] The second objective of this invention is to provide a system for identifying weld porosity defects in steel bridge decks.

[0006] Technical solution: This invention discloses a method for identifying weld porosity defects in steel bridge decks, comprising the following steps:

[0007] S1: Collect multiple acoustic emission signals, including those with no damage and those with porosity defects, on the test steel bridge deck to form a source domain dataset; collect multiple acoustic emission signals on the steel bridge deck to be tested to form a target domain dataset.

[0008] S2: The acoustic emission signal is processed by combining the adaptive analytical mode and Hilbert transform to obtain the adaptive marginal spectrum of the acoustic emission signal;

[0009] S3: Divide the target domain dataset into a target domain labeled dataset and a target domain unlabeled dataset. Use "non-destructive" and "with porosity defects" to label the acoustic emission signals contained in the source domain labeled dataset and the target domain labeled dataset. The target domain labeled dataset contains acoustic emission signals with both non-destructive and porosity defects.

[0010] S4: Adjust the architecture of Wav2Vec2.0 to obtain the improved AEWav2Vec2.0 model, and use the adaptive marginal spectrum of the source domain labeled dataset to train the AEWav2Vec2.0 model to obtain an initial classification model for identifying the label category of acoustic emission signals.

[0011] S5: Divide the target domain labeled dataset into a training set and a validation set according to a preset ratio, and ensure that both the training set and the validation set contain acoustic emission signals of non-destructive working conditions and those with porosity defects. The training set and the source domain labeled dataset constitute a hybrid labeled dataset, and the initial model is trained using the adaptive marginal spectrum of the hybrid labeled dataset to obtain an intermediate classification model.

[0012] S6: Input the adaptive marginal spectrum of the unlabeled dataset in the target domain into the intermediate classification model. The intermediate classification model outputs two classification results with confidence for the corresponding acoustic emission signal. Select acoustic emission signals with confidence higher than the preset value Q1. The selected acoustic emission signals and the mixed labeled dataset constitute a joint training dataset. Use the joint training dataset to iteratively optimize the intermediate classification model until the preset condition is met and the iterative optimization stops, and the target classification model is obtained.

[0013] S7: Input the adaptive marginal spectrum of the validation set into the target classification model. The target classification model outputs two classification results with confidence for the corresponding acoustic emission signal. Select acoustic emission signals with confidence higher than the preset value Q4 to form a prediction label dataset. Add the prediction label dataset to the joint training dataset to form the final training dataset. Use the final training dataset to iteratively optimize the target classification model until the same preset condition as in step S6 is reached, and stop the iterative optimization to obtain the final classification model.

[0014] S8: Input the adaptive marginal spectrum of the unlabeled dataset in the target domain into the final classification model. The final classification model outputs the classification result for each acoustic emission signal in the unlabeled dataset in the target domain.

[0015] Furthermore, in step S1, the method for collecting acoustic emission signals of welds with porosity defects on the test steel bridge deck is to perform a wet spraying treatment on the surface of the weld before welding.

[0016] Furthermore, the step in step S2 to obtain the adaptive marginal spectrum of the acoustic emission signal is as follows:

[0017] An adaptive analytical mode is used to decompose the acoustic emission signal into mutually exclusive spectral components;

[0018] Perform a Hilbert transform on all components of the acoustic emission signal to generate an analytic signal;

[0019] Calculate the instantaneous amplitude and instantaneous frequency of all analytical signals, and construct the time spectrum of the corresponding acoustic emission signal based on all instantaneous amplitudes and instantaneous frequencies of each acoustic emission signal;

[0020] Integrating the time spectrum along the time axis yields the adaptive marginal spectrum for each acoustic emission signal.

[0021] In step S4, the architecture of Wav2Vec2.0 is adjusted as follows: based on the dimension of the adaptive marginal spectrum of the acoustic emission signal, the kernel, stride, and number of input channels of the input convolutional layer of the Wav2Vec2.0 model are modified; Dropout and Pooling are added after the convolutional layer of Wav2Vec2.0; the number of encoder layers of the Transformer is set to 8; and the output layer of the Transformer is replaced with a binary classification fully connected layer.

[0022] Furthermore, the initial classification model in step S4 is optimized, with the optimization objective being to minimize the cross-entropy loss function L. cls L cls The expression is: Where y i For the actual labels, p(y) i |x i ) represents the y predicted by the initial classification model. i The label N refers to the number of acoustic emission signals contained in the source domain labeled dataset.

[0023] Furthermore, in step S5, during the training of the initial model, the feature extraction layer parameters of the initial classification model are retained, the last three layers of the Transformer encoder are unfrozen, and the parameters of the last three layers of the unfrozen Transformer encoder are allowed to be updated during training.

[0024] Furthermore, the steps for setting the preset conditions in step S6 are as follows:

[0025] Set a joint loss function L for supervised loss and label prediction loss of the intermediate classification model. joint And L joint The expression is L joint =λL sup +(1-λ)L predict ; among which, L predict For the cross-entropy loss of the acoustic emission signals selected based on confidence level, L sup The cross-entropy loss is calculated for the mixed labeled dataset, where λ is the dynamic weight.

[0026] When the dynamic weight λ decays to the preset value Q2, the iterative optimization stops.

[0027] Furthermore, the final classification model in step S8 also outputs whether each acoustic emission signal in the target domain dataset belongs to the predicted labeled dataset.

[0028] Furthermore, in step S8, acoustic emission signals whose confidence level output by the final classification model is higher than the preset value Q4 are selected, and the selected acoustic emission signals are added to the final training dataset. The target classification model is iteratively optimized again using the final training dataset until the same preset condition as in step S6 is reached and the iterative optimization stops.

[0029] Based on the same inventive concept, this invention also discloses a system for identifying weld porosity defects in steel bridge decks, comprising,

[0030] The signal acquisition module is used to acquire multiple acoustic emission signals, including those with no damage and those with porosity defects, on the test steel bridge deck. The acoustic emission signals acquired on the test steel bridge deck constitute the source domain dataset. Multiple acoustic emission signals are acquired on the steel bridge deck to be tested. The acoustic emission signals acquired on the steel bridge deck to be tested constitute the target domain dataset.

[0031] The signal processing module is used to process the acoustic emission signal by combining the adaptive analytical mode and Hilbert transform to obtain the adaptive marginal spectrum of the acoustic emission signal;

[0032] The signal classification module is used to divide the target domain dataset into a target domain labeled dataset and a target domain unlabeled dataset. It uses "non-destructive" and "with porosity and defects" to label the acoustic emission signals contained in the source domain labeled dataset and the target domain labeled dataset. The target domain labeled dataset contains acoustic emission signals with both non-destructive and porosity and defects.

[0033] The initial classification module is used to adjust the architecture of Wav2Vec2.0 to obtain the improved AEWav2Vec2.0 model. The AEWav2Vec2.0 model is trained using the adaptive marginal spectrum of the source domain labeled dataset to obtain the initial classification model for identifying the label category of acoustic emission signals.

[0034] The intermediate classification module is used to divide the target domain labeled dataset into a training set and a validation set according to a preset ratio, and ensure that both the training set and the validation set contain acoustic emission signals of non-destructive conditions and those with porosity defects. The training set and the source domain labeled dataset constitute a hybrid labeled dataset, and the initial model is trained using the adaptive marginal spectrum of the hybrid labeled dataset to obtain the intermediate classification model.

[0035] The target classification module is used to input the adaptive marginal spectrum of the unlabeled dataset of the target domain into the intermediate classification model. The intermediate classification model outputs two classification results with confidence levels for the corresponding acoustic emission signals. Acoustic emission signals with confidence levels higher than the preset value Q1 are selected, and the selected acoustic emission signals and the mixed labeled dataset constitute a joint training dataset. The intermediate classification model is iteratively optimized using the joint training dataset until the preset conditions are met and the iterative optimization stops, and the target classification model is obtained.

[0036] The final classification module is used to input the adaptive marginal spectrum of the validation set into the target classification model. The target classification model outputs two classification results with confidence levels for the corresponding acoustic emission signals. Acoustic emission signals with confidence levels higher than the preset value Q4 are selected to form a prediction label dataset. The prediction label dataset is added to the joint training dataset to form the final training dataset. The target classification model is iteratively optimized using the final training dataset. After the preset conditions are met, the iterative optimization stops and the final classification model is obtained.

[0037] The results output module is used to input the adaptive marginal spectrum of the unlabeled dataset in the target domain into the final classification model. The final classification model outputs the classification result and confidence level of each acoustic emission signal in the unlabeled dataset in the target domain.

[0038] Beneficial Effects: Compared with existing technologies, this invention has the following significant advantages: By collecting acoustic emission signals during the welding process, this invention enables online monitoring of the welding process of steel bridge decks, accurately identifying porosity defects during welding with high efficiency. This can be used to guide the welding process of steel bridge decks, thus improving the overall welding quality. Furthermore, the acoustic emission signals utilize transient elastic waves released during plastic deformation and fracture processes to diagnose structural damage, offering advantages of high detection efficiency and high sensitivity.

[0039] This invention combines adaptive analytical mode and Hilbert transform to obtain the adaptive marginal spectrum of acoustic emission signals. On the one hand, it avoids the mode mixing phenomenon of traditional empirical mode analysis. On the other hand, it adaptively extracts different frequency components contained in the acoustic emission signal, which can obtain a time spectrum with clearer physical meaning and effectively characterize the frequency domain energy distribution of the acoustic emission signal.

[0040] The acoustic emission signal of the steel bridge deck to be tested in this invention is divided into three parts to participate in the training and optimization of the classification model. On the one hand, it can improve the overall feature recognition of the acoustic emission signal of the steel bridge deck to be tested, which is conducive to improving the accuracy of the final recognition result. On the other hand, it can train the model during the testing process and realize the continuous optimization of the classification model parameters, which is conducive to improving the long-term stability and accuracy of the model in the application process. That is, as the number of uses and the duration of use increase, the accuracy of the classification result of this invention also increases. Attached Figure Description

[0041] Figure 1 This is a flowchart of the method of the present invention;

[0042] Figure 2 This is a framework diagram of the initial classification model of an embodiment of the present invention;

[0043] Figure 3 This is a top view of the steel bridge deck according to an embodiment of the present invention;

[0044] Figure 4 This is a front view of the steel bridge deck according to an embodiment of the present invention;

[0045] Figure 5 The time-domain waveform of the non-destructive acoustic emission signal of the steel bridge deck in an embodiment of the present invention;

[0046] Figure 6 The time-domain waveform of the acoustic emission signal of a steel bridge deck with porosity defects is shown in an embodiment of the present invention.

[0047] Figure 7 For the present invention Figure 5 The original waveform and component waveforms of the undamaged acoustic emission signal;

[0048] Figure 8 For the present invention Figure 6 The original waveform and component waveforms of the acoustic emission signal with pore defects are shown.

[0049] Figure 9 This is an adaptive marginal spectrum of the acoustic emission signal under non-destructive operating conditions according to an embodiment of the present invention;

[0050] Figure 10 This is an adaptive marginal spectrum of the acoustic emission signal under non-destructive operating conditions according to an embodiment of the present invention. Detailed Implementation

[0051] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0052] Example 1

[0053] like Figure 1 As shown, the present invention discloses a method for identifying weld porosity defects in steel bridge decks, comprising the following steps:

[0054] S1: Collect multiple acoustic emission signals, including those with no damage and those with porosity defects, on the test steel bridge deck to form a source domain dataset; collect multiple acoustic emission signals on the steel bridge deck to be tested to form a target domain dataset.

[0055] Preferably, when acquiring the initial sample acoustic emission signal, non-destructive welding and welding with porosity defects are performed on the test steel bridge deck. The method for acquiring the acoustic emission signal with porosity defects is to spray the surface of the weld with moisture before welding. The water vapor generated during the welding process will then be used to acquire the acoustic emission signal with porosity defects. Acquiring the acoustic emission signal with porosity defects is not limited to this method; other methods can also be used to create porosity defects in actual use.

[0056] Preferably, acoustic emission sensors are installed at the four corners of both the test steel bridge deck and the steel bridge deck to be tested, and the frequency range of the acoustic emission sensors is 100-1000 kHz. Preferably, a preamplifier is used to improve the signal-to-noise ratio of the acoustic emission signal.

[0057] S2: Combine the adaptive analytical mode and Hilbert transform to process all acoustic emission signals to obtain the adaptive marginal spectrum of the acoustic emission signals.

[0058] S21: The acoustic emission signal is decomposed into mutually exclusive spectral components using an adaptive analytical mode. The decomposition expression is as follows:

[0059]

[0060] Where x(t) refers to the acoustic emission signal, Let i be the i-th mutually exclusive component of the spectrum of x(t), where 1 ≤ i ≤ n, and n is the total number of components. In practical applications, the cutoff frequency of the mode decomposition can be adaptively adjusted according to the situation or requirements.

[0061] S22: Perform a Hilbert transform on each component of the acoustic emission signal to generate an analytic signal.

[0062]

[0063] Among them, z i (t) refers to the analytic signal, and H[x] refers to the analytic expression of the acoustic emission signal x(t). Perform a Hilbert transform, where j is the imaginary unit.

[0064] S23: Calculate the instantaneous amplitude A of all analytic signals. i (t) and instantaneous frequency f i (t):

[0065]

[0066] The instantaneous amplitude and frequency domain information of all acoustic emission signals are superimposed on the time-frequency plane to construct the time spectrum, which is also the construction of the analytic mode decomposition-Hilbert spectrum:

[0067]

[0068] Where f is frequency; t is time; H(f,t) is the time spectrum; and δ[·] is the Dirac function, representing that energy is concentrated at the instantaneous frequency. This invention obtains mutually exclusive frequency components by combining analytical mode decomposition and Hilbert transform, and adaptively extracts different frequency components contained in the acoustic emission signal, thus obtaining a time spectrum with clearer physical meaning.

[0069] S24: Integrate the time spectrum along the time axis to obtain the adaptive marginal spectrum of each acoustic emission signal, and the expression for the adaptive marginal spectrum M(f) of the acoustic emission signal is:

[0070]

[0071] Where T is the total duration of the acoustic emission signal; the adaptive marginal spectrum can reflect the cumulative energy-frequency distribution of the acoustic emission signal within the total duration, and can be used to identify the characteristic frequencies of welding porosity defects.

[0072] S3: Divide the target domain dataset into a target domain labeled dataset and a target domain unlabeled dataset. Label the acoustic emission signals contained in the source domain labeled dataset and the target domain labeled dataset using the criteria of "non-destructive" and "with porosity and defects." The target domain labeled dataset contains both non-destructive and porosity-defective acoustic emission signals. The labeling method for the target domain labeled dataset is, in actual use, to label the acoustic emission signals in the target domain dataset that are obviously non-destructive or obviously defective based on human experience.

[0073] Preferably, the ratio of acoustic emission signal data contained in the source domain standard dataset, the target domain labeled dataset, and the target domain unlabeled dataset is 6:1:3. In actual use, this preset ratio can be adjusted according to the actual situation and needs. For example, in actual use, when labeling acoustic emission signals in the target domain dataset, obviously undamaged and defective acoustic emission signals are usually labeled, that is, the target domain labeled dataset accounts for a smaller proportion.

[0074] S4: Adjust the architecture of Wav2Vec2.0 to obtain the improved AEWav2Vec2.0 model, and train the AEWav2Vec2.0 model using the adaptive marginal spectrum of the source domain labeled dataset to obtain an initial classification model for identifying the label category of acoustic emission signals.

[0075] Preferably, the adaptive marginal spectra of all acoustic emission signals are normalized beforehand to obtain the model input tensor of the Wav2Vec2.0 model.

[0076] The architecture of Wav2Vec2.0 was adjusted as follows:

[0077] Based on the dimension of the adaptive marginal spectrum of the acoustic emission signal, the convolution kernel of the input convolutional layer of the Wav2Vec2.0 model is modified to adapt to the long sequence characteristics of the adaptive marginal spectrum; the stride of the input convolutional layer of the Wav2Vec2.0 model is modified to reduce computation; and the number of input channels of the input convolutional layer of the Wav2Vec2.0 model is modified. The convolution kernel size is typically chosen to be relatively small, such as 30, to extract detailed features from longer acoustic emission signals, such as when extracting detailed features from an adaptive marginal spectrum with a dimension of 32768×1. Preferably, a stride of 10 is chosen, which strikes a balance between computational efficiency and information preservation. The number of input channels is usually set to 1 because each acoustic emission signal is typically a single channel.

[0078] Considering the limited number of acoustic emissions, which can easily lead to overfitting in the early stages, Dropout is added after the convolutional layers of Wav2Vec2.0 to prevent overfitting and enhance the model's generalization ability. Preferably, the Dropout dropout rate is 0.1. A Pooling operation is added after the convolutional layers to reduce dimensionality through downsampling, allowing subsequent layers of Wav2Vec2.0 to process only the refined features. This reduces the original 7 convolutional layers in the Wav2Vec2.0 model to 3 without affecting its performance. The Transformer encoder is set to 8 layers to reduce the complexity of the Wav2Vec2.0 model and adapt to situations with fewer acoustic emission signal samples. The Transformer output layer is replaced with a binary classification fully connected layer to distinguish between the "no damage" and "porosity defects" labels.

[0079] Preferably, the AEWav2Vec2.0 model is trained using the adaptive marginal spectrum of the source domain labeled dataset for supervised training, and the training optimization objective is to minimize the cross-entropy loss function L. cls The cross-entropy loss function L cls The expression is

[0080]

[0081] Among them, y i For the acoustic emission signal, a real label is pre-labeled, p(y) i |x i ) represents the y predicted by the initial classification model. i The label N represents the number of acoustic emission signals contained in the source domain labeled dataset. The cross-entropy loss function L is minimized. cls It can optimize the weights and biases of convolutional layers in the feature extractor, the weight matrix of the attention layer in the Transformer encoder, and the classification weights of the output layer.

[0082] In summary, as Figure 2 As shown, the initial classification model obtained after training includes an input one-dimensional convolutional module, a Transformer encoder, a Pooling layer, and a binary classification output layer, which can perform preliminary identification and classification of the adaptive marginal spectrum of acoustic emission signals.

[0083] S5: Divide the target domain labeled dataset into a training set and a validation set according to a preset ratio, and ensure that both the training set and the validation set contain acoustic emission signals of non-destructive working conditions and those with porosity defects. The training set and the source domain labeled dataset constitute a hybrid labeled dataset, and the initial model is trained using the adaptive marginal spectrum of the hybrid labeled dataset to obtain an intermediate classification model.

[0084] Preferably, the preset ratio of the target domain labeled dataset is training set:validation set = 1:1; in actual use, the ratio can be adjusted according to needs.

[0085] Because the acoustic emission signals of the source domain labeled dataset and the target domain labeled dataset differ, training with only the source domain labeled dataset can easily lead to overfitting to the acoustic emission signal features of the source domain labeled dataset, resulting in poor feature recognition ability for the acoustic emission signals of the target domain labeled dataset. Merging the training set of the target domain labeled dataset with the source domain labeled dataset to construct a hybrid labeled dataset helps the model gradually adapt to the characteristics of the target domain data, improving the accuracy and stability of the intermediate classification model in cross-domain recognition.

[0086] Preferably, during the training of the initial model, the feature extraction layer parameters of the initial classification model are retained, the last three layers of the Transformer encoder are unfrozen, and the parameters of the last three layers of the unfrozen Transformer encoder are allowed to be updated during training, so as to improve the adaptability of the intermediate classification model to the features of acoustic emission signals in the target domain labeled dataset.

[0087] S6: Input the adaptive marginal spectrum of the unlabeled dataset in the target domain into the intermediate classification model. The intermediate classification model outputs two classification results with confidence levels for the corresponding acoustic emission signals. Select acoustic emission signals with confidence levels higher than the preset value Q1. The selected acoustic emission signals and the mixed labeled dataset constitute a joint training dataset. Use the joint training dataset to iteratively optimize the intermediate classification model until the preset conditions are met and the iterative optimization stops, and the target classification model is obtained.

[0088] Preferably, the preset confidence level Q1 is 0.9; in actual use, the preset value Q1 can be adjusted according to the needs.

[0089] A joint loss function is set up, which combines the supervised loss and the label prediction loss of the intermediate classification model, and a dynamic weight adjustment mechanism is defined. The joint loss function L...joint The expression is

[0090] L joint =λL sup +(1-λ)L predict

[0091] Among them, L predict For the cross-entropy loss of the acoustic emission signals selected based on confidence level, L sup The cross-entropy loss is used for the mixed labeled dataset, and λ represents the dynamic weights. The intermediate classification model stops iterative optimization when the dynamic weights λ decay to a preset value Q2, preferably 0.3. Using the joint loss function for iterative training of the fine-tuned model helps preserve the model parameters that yield the best validation performance.

[0092] S7: Input the adaptive marginal spectrum of the validation set into the target classification model. The target classification model outputs two classification results with confidence levels for the corresponding acoustic emission signals. Select acoustic emission signals with confidence levels higher than the preset value Q4 to form a prediction label dataset. Add the prediction label dataset to the joint training dataset to form the final training dataset. Use the final training dataset to iteratively optimize the target classification model until the same preset conditions as in step S6 are met, and stop the iterative optimization to obtain the final classification model.

[0093] Preferably, the preset value Q4 is set to 0.9, but Q4 can be adjusted according to needs during actual use. Preferably, Q4 is equal to Q1, but their actual values ​​can be set separately in actual applications.

[0094] S8: Input the adaptive marginal spectrum of the unlabeled dataset in the target domain into the final classification model. The final classification model outputs the classification result and confidence level of each acoustic emission signal in the unlabeled dataset in the target domain.

[0095] If the confidence level of an acoustic emission signal having a porosity defect exceeds a preset value Q3, then the acoustic emission signal is determined to have a porosity defect. The final classification model also outputs whether each acoustic emission signal in the target domain dataset belongs to the predicted labeled dataset.

[0096] Preferably, acoustic emission signals with a confidence level higher than a preset value Q4 are selected from the final classification model output and added to the final training dataset. The target classification model is then iteratively optimized again using the final training dataset until the same preset condition as in step S6 is met. After the final classification model outputs its results, acoustic emission signals with high confidence levels higher than the preset value Q4 from the unlabeled dataset of the target domain are selected and used again to iteratively optimize the target classification model, which can further improve the accuracy of the final recognition results.

[0097] In this embodiment, step S5 uses a hybrid labeled dataset containing the training set to optimize the initial model; step S6 uses a hybrid labeled dataset containing unlabeled data from the target domain to optimize the intermediate classification model; and step S7 uses a joint hybrid labeled dataset containing the validation set to iteratively optimize the target classification model. This joint training can effectively reduce the distribution differences between domains, which is beneficial to improving the overall accuracy of identifying porosity defects. This embodiment divides the acoustic emission signal of the steel bridge deck to be tested into three parts: a labeled training set, a labeled validation set, and an unlabeled dataset from the target domain. These three parts of the acoustic emission signal participate in the training and optimization of the classification model step by step. On the one hand, this can improve the overall feature recognition of the acoustic emission signal of the steel bridge deck to be tested, which is beneficial to improving the accuracy of the final identification result; on the other hand, it can train the model during the testing process, realizing continuous optimization of the classification model parameters, which is beneficial to improving the long-term stability and accuracy of the model during application. That is, as the number of uses and the duration of use increase, the accuracy of the classification result output by the final classification model also increases.

[0098] Example 2

[0099] The experimental verification process of a method for identifying weld porosity defects in steel bridge decks disclosed in this invention.

[0100] S1: Collect multiple acoustic emission signals, including those with no damage and those with porosity defects, on the test steel bridge deck to form a source domain dataset; collect multiple acoustic emission signals on the steel bridge deck to be tested to form a target domain dataset.

[0101] In this embodiment, the welding type was robotic gas shielded welding. Seven steel bridge deck panels were used in the experiment. Welding was performed on five of these panels under non-destructive conditions, and 700 non-destructive acoustic emission signals were collected. The weld surfaces of the remaining two steel bridge deck panels underwent wet spraying treatment, and 300 acoustic emission signals with porosity defects were collected. A schematic diagram of the steel bridge deck structure is shown below. Figure 3 and Figure 4 As shown, the dimensions of the steel bridge deck are as follows: length and width are 500mm × 600mm, and the plate thickness is 20mm; the lower edge of the U-shaped longitudinal rib is 300mm wide, 280mm high, and 8mm thick.

[0102] In this embodiment, four broadband acoustic emission sensors with a frequency range of 100-1000kHz are used to collect acoustic emission signals, and the four acoustic emission sensors are as follows: Figure 3 As shown in S1-S4, the acoustic emission sensors are installed at the top corners of the steel bridge deck, with the installation position of the acoustic emission sensor 30mm away from the edge of the steel bridge deck. They are fixed by coupling agent and magnetic clamps. The signal-to-noise ratio of the acoustic emission signal is improved by a preamplifier, with the amplification factor set to 40dB, the trigger sampling threshold set to 40dB, and the sampling frequency set to 5MHz.

[0103] In this embodiment, a total of 700 undamaged acoustic emission signals and 300 acoustic emission signals with pore defects were obtained. 600 of these acoustic emission signals were randomly selected to form a source domain dataset, which includes both undamaged and pore-defect acoustic emission signals. The remaining 400 acoustic emission signals formed a target domain dataset, which also includes both undamaged and pore-defect acoustic emission signals. The time-domain waveform of one of the undamaged acoustic emission signals received by the acoustic emission sensor in this embodiment is shown below. Figure 5 As shown, the time-domain waveform of one of the acoustic emission signals with porosity defects received by the acoustic emission sensor in this embodiment is as follows: Figure 6 As shown.

[0104] S2: Combine the adaptive analytical mode and Hilbert transform to process all acoustic emission signals to obtain the adaptive marginal spectrum of the acoustic emission signals.

[0105] The acoustic emission signal acquired in step S1 is decomposed into multiple mutually exclusive frequency components using an adaptive analytical mode:

[0106]

[0107] Where x(t) is the acoustic emission signal, For n mutually exclusive spectral components; Figure 5 Taking the non-destructive acoustic emission signal as an example, the adaptive analytical mode decomposition method is used to decompose it into 5 components. The original waveform and the waveforms of the 5 components of the non-destructive acoustic emission signal are as follows: Figure 7 As shown; Figure 6 Taking the acoustic emission signal with pore defects as an example, the adaptive analytical mode decomposition method is used to decompose it into 5 components. The original waveform and the waveforms of the 5 components of the acoustic emission signal with pore defects are as follows. Figure 8 As shown.

[0108] Perform a Hilbert transform on each component of the decomposed acoustic emission signal to generate an analytic signal:

[0109]

[0110] Among them, z i (t) is an analytic signal, and H[x] represents the Hilbert transform of the signal x.

[0111] Calculate the instantaneous amplitude A of the analytic signal i (t) and instantaneous frequency f i (t):

[0112]

[0113] The instantaneous amplitudes of all components of the acoustic emission signal are superimposed with the frequency domain information on the time-frequency plane to construct the time spectrum (analytical mode decomposition-Hilbert spectrum):

[0114]

[0115] Where H(f,t) is the analytical mode decomposition-Hilbert spectrum, and δ[·] is the Dirac function, indicating that the energy is concentrated at the instantaneous frequency.

[0116] Integrating the time spectrum over the time dimension yields the adaptive marginal spectrum M(f) of the acoustic emission signal, which is expressed as:

[0117]

[0118] Where T is the total signal duration; the adaptive marginal spectrum reflects the cumulative energy-frequency distribution of the acoustic emission signal over the entire time period and is used to identify the characteristic frequencies of welding porosity defects. The adaptive marginal spectrum of the acoustic emission signal under the non-destructive welding condition of the steel bridge deck in this embodiment is as follows: Figure 9 As shown, the adaptive marginal spectrum of acoustic emission signal under the condition of porosity defects in the welded steel bridge deck is as follows: Figure 10 As shown.

[0119] S3: The target domain dataset is divided into a target domain labeled dataset and a target domain unlabeled dataset. The acoustic emission signals contained in the source domain labeled dataset and the target domain labeled dataset are labeled as "non-damaging" and "with porosity defects." The target domain labeled dataset contains both non-damaging and porosity defect acoustic emission signals. The target domain labeled dataset includes 100 acoustic emission signals, and the target domain unlabeled dataset includes 300 acoustic emission signals. In this embodiment, the user actually knows whether the 1000 acoustic emission signals are non-damaging or porosity defective, but the user only labels 600 acoustic emission signals in the source domain dataset and 100 acoustic emission signals in the target domain labeled dataset, and does not label the 300 acoustic emission signals in the target domain unlabeled dataset. This data label setting simulates a scenario in a real-world application.

[0120] S4: Adjust the architecture of Wav2Vec2.0 to obtain the improved AEWav2Vec2.0 model, and train the AEWav2Vec2.0 model using the adaptive marginal spectrum of the source domain labeled dataset to obtain an initial classification model for identifying the label category of acoustic emission signals.

[0121] The adaptive marginal spectra of all acoustic emission signals are normalized beforehand to obtain the model input tensor of the Wav2Vec2.0 model.

[0122] The architecture of Wav2Vec2.0 was adjusted as follows:

[0123] Based on the dimensionality of the adaptive marginal spectrum of acoustic emission signals, the convolution kernels of the input convolutional layers of the Wav2Vec2.0 model are modified to adapt to the long sequence characteristics of the adaptive marginal spectrum; the stride of the input convolutional layers of the Wav2Vec2.0 model is modified to reduce computation; and the number of input channels of the input convolutional layers of the Wav2Vec2.0 model is modified. Considering the limited number of acoustic emissions, which can easily lead to overfitting in the early stages, Dropout is added after the convolutional layers of Wav2Vec2.0 to prevent overfitting and enhance the model's generalization ability. Preferably, the dropout rate is 0.1. A pooling operation is added after the convolutional layers to reduce the dimensionality through downsampling, so that subsequent layers of Wav2Vec2.0 only need to process the refined features, thereby reducing the original 7 convolutional layers in the Wav2Vec2.0 model to 3 layers without affecting the performance of the Wav2Vec2.0 model. The number of encoder layers in the Transformer is set to 8 to reduce the complexity of the Wav2Vec2.0 model and adapt to the case of fewer acoustic emission signal samples. The output layer of the Transformer is replaced with a binary classification fully connected layer to distinguish between the two categories of "no damage" and "porosity defects".

[0124] Preferably, the Wav2Vec2.0 model is trained using the adaptive marginal spectrum of the source domain labeled dataset for supervised training, and the training optimization objective is to minimize the cross-entropy loss function L. cls The cross-entropy loss function L cls The expression is

[0125]

[0126] Among them, y i For the acoustic emission signal, a real label is pre-labeled, p(y) i |x i ) represents the y predicted by the initial classification model. i The label N represents the number of acoustic emission signals contained in the source domain labeled dataset. The cross-entropy loss function L is minimized. cls It can optimize the weights and biases of convolutional layers in the feature extractor, the weight matrix of the attention layer in the Transformer encoder, and the classification weights of the output layer.

[0127] S5: Divide the target domain labeled dataset into a training set and a validation set according to a preset ratio, ensuring that both the training and validation sets contain acoustic emission signals from both non-destructive and porosity-defect conditions. The training set and the source domain labeled dataset constitute a hybrid labeled dataset. The initial model is trained using the adaptive marginal spectrum of the hybrid labeled dataset to obtain an intermediate classification model. The preset ratio of the target domain labeled dataset is training set:validation set = 1:1, meaning that the training set and validation set each contain 150 acoustic emission signals.

[0128] During the training of the initial model, the feature extraction layer parameters of the initial classification model are retained, the last three layers of the Transformer encoder are unfrozen, and the parameters of the last three layers of the unfrozen Transformer encoder are allowed to be updated during training to improve the adaptability of the intermediate classification model to the features of acoustic emission signals in the target domain labeled dataset.

[0129] S6: Input the adaptive marginal spectrum of the unlabeled dataset in the target domain into the intermediate classification model. The intermediate classification model outputs two classification results with confidence levels for the corresponding acoustic emission signals. Select acoustic emission signals with confidence levels higher than a preset value Q1. The selected acoustic emission signals and the mixed labeled dataset constitute a joint training dataset. Use the joint training dataset to iteratively optimize the intermediate classification model until a preset condition is met, at which point the iteration stops, and the target classification model is obtained. The preset value Q1 is 0.9.

[0130] A joint loss function is set up, which combines the supervised loss and the label prediction loss of the intermediate classification model, and a dynamic weight adjustment mechanism is defined. The joint loss function L... joint The expression is

[0131] L joint =λL sup +(1-λ)L predict

[0132] Among them, L predict For the cross-entropy loss of the acoustic emission signals selected based on confidence level, L sup The cross-entropy loss is used for the mixed labeled dataset, and λ represents the dynamic weights. The intermediate classification model stops iterative optimization when the dynamic weights λ decay to a preset value Q2, and Q2 is set to 0.3. In this embodiment, the initial value of the dynamic weights λ is 0.7, and iterative optimization stops when the dynamic weights λ decay to 0.3.

[0133] S7: Input the adaptive marginal spectrum of the validation set into the target classification model. The target classification model outputs two classification results with confidence for the corresponding acoustic emission signal. Select acoustic emission signals with confidence higher than the preset value Q4 to form a prediction label dataset. Add the prediction label dataset to the joint training dataset to form the final training dataset. Use the final training dataset to iteratively optimize the target classification model until the same preset condition as in step S6 is reached, and stop the iterative optimization to obtain the final classification model.

[0134] S8: Input the adaptive marginal spectrum of the unlabeled dataset in the target domain into the final classification model. The final classification model outputs the classification result and confidence score for each acoustic emission signal in the unlabeled dataset of the target domain. The final classification model also outputs whether each acoustic emission signal in the target domain dataset belongs to the predicted labeled dataset.

[0135] When the confidence level of an acoustic emission signal having a porosity defect exceeds 0.5, the acoustic emission signal is determined to have a porosity defect. The classification results ("no damage" or "with porosity defect") of 300 acoustic emission signals in the unlabeled dataset of the target domain output by the final classification model are compared with the true labels of the 300 acoustic emission signals in step S1. In this embodiment, the recognition rate of no-damage samples reaches 96% and the recognition rate of porosity defect samples reaches 95% on the unlabeled dataset of the target domain, verifying the accuracy, stability and adaptability of the method of the present invention in the identification of porosity defects in steel bridge deck welding.

[0136] Example 3

[0137] The present invention discloses a steel bridge deck welding porosity defect identification system, which includes a signal acquisition module, a signal processing module, a signal classification module, an initial classification module, an intermediate classification module, a target classification module, a final classification module, and a result output module.

[0138] The signal acquisition module is used to acquire multiple acoustic emission signals, including those with no damage and those with porosity defects, on the test steel bridge deck, and the acoustic emission signals acquired on the test steel bridge deck constitute the source domain dataset; multiple acoustic emission signals are acquired on the steel bridge deck to be tested, and the acoustic emission signals acquired on the steel bridge deck to be tested constitute the target domain dataset. The signal acquisition module performs step S1 in Example 1.

[0139] The signal processing module is used to process the acoustic emission signal by combining the adaptive analytical mode and the Hilbert transform to obtain the adaptive marginal spectrum of the acoustic emission signal. The signal processing module executes step S2 in Example 1.

[0140] The signal classification module is used to divide the target domain dataset into a target domain labeled dataset and a target domain unlabeled dataset. It labels the acoustic emission signals contained in the source domain labeled dataset and the target domain labeled dataset using the criteria of "non-destructive" and "with porosity and defects." The target domain labeled dataset contains both non-destructive and porosity-defective acoustic emission signals. The signal classification module executes step S3 in Example 1.

[0141] The initial classification module adjusts the architecture of Wav2Vec2.0 to obtain an improved AEWav2Vec2.0 model. It then trains the AEWav2Vec2.0 model using the adaptive marginal spectrum of the source domain labeled dataset, resulting in an initial classification model for identifying the label categories of acoustic emission signals. The initial classification module executes step S4 in Example 1.

[0142] The intermediate classification module is used to divide the target domain labeled dataset into a training set and a validation set according to a preset ratio, ensuring that both the training set and the validation set contain acoustic emission signals from both non-destructive conditions and those with porosity defects. The training set and the source domain labeled dataset constitute a hybrid labeled dataset, and the initial model is trained using the adaptive marginal spectrum of the hybrid labeled dataset to obtain the intermediate classification model. The intermediate classification module executes step S5 in Example 1.

[0143] The target classification module is used to input the adaptive marginal spectrum of the unlabeled dataset in the target domain into the intermediate classification model. The intermediate classification model outputs two classification results with confidence levels for the corresponding acoustic emission signals. Acoustic emission signals with confidence levels higher than a preset value Q1 are selected, and the selected acoustic emission signals and the mixed labeled dataset constitute a joint training dataset. The intermediate classification model is iteratively optimized using the joint training dataset until a preset condition is met, at which point the iterative optimization stops, and the target classification model is obtained. The target classification module executes step S6 in Example 1.

[0144] The final classification module inputs the adaptive marginal spectrum of the validation set into the target classification model. The target classification model outputs two classification results with confidence levels for the corresponding acoustic emission signals. Acoustic emission signals with confidence levels higher than a preset value Q4 are selected to form a prediction labeled dataset. The prediction labeled dataset is added to the joint training dataset to form the final training dataset. The target classification model is iteratively optimized using the final training dataset. After reaching a preset condition, the iterative optimization stops and the final classification model is obtained. The final classification module executes step S7 in Example 1.

[0145] The result output module is used to input the adaptive marginal spectrum of the target domain dataset into the final classification model. The final classification model outputs two classification results with confidence levels. When the confidence level of a certain acoustic emission signal having porosity defects exceeds the preset value Q3, the acoustic emission signal is determined to have porosity defects. The result output module executes step S8 in Example 1.

Claims

1. A method for identifying weld porosity defects in steel bridge decks, characterized in that: Includes the following steps, S1: Collect multiple acoustic emission signals, including those with no damage and those with porosity defects, on the test steel bridge deck to form a source domain dataset; collect multiple acoustic emission signals on the steel bridge deck to be tested to form a target domain dataset. S2: The acoustic emission signal is processed by combining the adaptive analytical mode and Hilbert transform to obtain the adaptive marginal spectrum of the acoustic emission signal; the steps to obtain the adaptive marginal spectrum of the acoustic emission signal are as follows: An adaptive analytical mode is used to decompose the acoustic emission signal into mutually exclusive spectral components; Perform a Hilbert transform on all components of the acoustic emission signal to generate an analytic signal; Calculate the instantaneous amplitude and instantaneous frequency of all analytical signals, and construct the time spectrum of the corresponding acoustic emission signal based on all instantaneous amplitudes and instantaneous frequencies of each acoustic emission signal; Integrating the time spectrum along the time axis yields the adaptive marginal spectrum for each acoustic emission signal. S3: Divide the target domain dataset into a target domain labeled dataset and a target domain unlabeled dataset. Use "non-destructive" and "with porosity defects" to label the acoustic emission signals contained in the source domain labeled dataset and the target domain labeled dataset. The target domain labeled dataset contains acoustic emission signals with both non-destructive and porosity defects. S4: Adjust the architecture of Wav2Vec2.0 to obtain the improved AEWav2Vec2.0 model, and use the adaptive marginal spectrum of the source domain labeled dataset to train the AEWav2Vec2.0 model to obtain an initial classification model for identifying the label category of acoustic emission signals. S5: Divide the target domain labeled dataset into a training set and a validation set according to a preset ratio, and ensure that both the training set and the validation set contain acoustic emission signals of non-destructive working conditions and those with porosity defects. The training set and the source domain labeled dataset constitute a hybrid labeled dataset, and the initial model is trained using the adaptive marginal spectrum of the hybrid labeled dataset to obtain an intermediate classification model. S6: Input the adaptive marginal spectrum of the unlabeled dataset in the target domain into the intermediate classification model. The intermediate classification model outputs two classification results with confidence for the corresponding acoustic emission signal. Select acoustic emission signals with confidence higher than the preset value Q1. The selected acoustic emission signals and the mixed labeled dataset constitute a joint training dataset. Use the joint training dataset to iteratively optimize the intermediate classification model until the preset condition is met and the iterative optimization stops, and the target classification model is obtained. S7: Input the adaptive marginal spectrum of the validation set into the target classification model. The target classification model outputs two classification results with confidence for the corresponding acoustic emission signal. Select acoustic emission signals with confidence higher than the preset value Q4 to form a prediction label dataset. Add the prediction label dataset to the joint training dataset to form the final training dataset. Use the final training dataset to iteratively optimize the target classification model until the same preset condition as in step S6 is reached, and stop the iterative optimization to obtain the final classification model. S8: Input the adaptive marginal spectrum of the unlabeled dataset in the target domain into the final classification model. The final classification model outputs the classification result and confidence level of each acoustic emission signal in the unlabeled dataset in the target domain.

2. The method for identifying weld porosity defects in steel bridge decks according to claim 1, characterized in that: The method for collecting acoustic emission signals of welds with porosity defects on the test steel bridge deck in step S1 is to perform a wet spraying treatment on the surface of the weld before welding.

3. The method for identifying weld porosity defects in steel bridge decks according to claim 1, characterized in that: In step S4, the architecture of Wav2Vec2.0 is adjusted as follows: based on the dimension of the adaptive marginal spectrum of the acoustic emission signal, the kernel, stride, and number of input channels of the input convolutional layer of the Wav2Vec2.0 model are modified; Dropout and Pooling are added after the convolutional layer of Wav2Vec2.0; the number of encoder layers of the Transformer is set to 8; and the output layer of the Transformer is replaced with a binary classification fully connected layer.

4. The method for identifying weld porosity defects in steel bridge decks according to claim 1, characterized in that: The initial classification model in step S4 is optimized, and the optimization objective is to minimize the cross-entropy loss function. , The expression is: ; where y i For the actual labels, p(y) i |x i ) represents the y predicted by the initial classification model. i The label N refers to the number of acoustic emission signals contained in the source domain labeled dataset.

5. The method for identifying weld porosity defects in steel bridge decks according to claim 1, characterized in that: In step S5, during the training of the initial model, the feature extraction layer parameters of the initial classification model are retained, the last 3 layers of the Transformer encoder are unfrozen, and the parameters of the last 3 layers of the unfrozen Transformer encoder are allowed to be updated during the training process.

6. The method for identifying weld porosity defects in steel bridge decks according to claim 1, characterized in that: The steps for setting the preset conditions in step S6 are as follows: Set a joint loss function for supervised loss and label prediction loss from intermediate classification models. ,and The expression is ;in, The cross-entropy loss of the acoustic emission signals selected based on confidence level. The cross-entropy loss is calculated for the mixed labeled dataset, where λ is the dynamic weight. When the dynamic weight λ decays to the preset value Q2, the iterative optimization stops.

7. The method for identifying weld porosity defects in steel bridge decks according to claim 1, characterized in that: The final classification model in step S8 also outputs whether each acoustic emission signal in the target domain dataset belongs to the predicted labeled dataset.

8. The method for identifying weld porosity defects in steel bridge decks according to claim 1, characterized in that: In step S8, acoustic emission signals whose confidence level output by the final classification model is higher than the preset value Q4 are selected, and the selected acoustic emission signals are added to the final training dataset. The target classification model is iteratively optimized again using the final training dataset until the same preset condition as in step S6 is reached and the iterative optimization stops.

9. A system for identifying weld porosity defects in steel bridge decks, characterized in that: include, The signal acquisition module is used to acquire multiple acoustic emission signals, including those with no damage and those with porosity defects, on the test steel bridge deck. The acoustic emission signals acquired on the test steel bridge deck constitute the source domain dataset. Multiple acoustic emission signals are acquired on the steel bridge deck to be tested. The acoustic emission signals acquired on the steel bridge deck to be tested constitute the target domain dataset. The signal processing module is used to process the acoustic emission signal by combining the adaptive analytical mode and Hilbert transform to obtain the adaptive marginal spectrum of the acoustic emission signal; the steps for obtaining the adaptive marginal spectrum of the acoustic emission signal are as follows: An adaptive analytical mode is used to decompose the acoustic emission signal into mutually exclusive spectral components; Perform a Hilbert transform on all components of the acoustic emission signal to generate an analytic signal; Calculate the instantaneous amplitude and instantaneous frequency of all analytical signals, and construct the time spectrum of the corresponding acoustic emission signal based on all instantaneous amplitudes and instantaneous frequencies of each acoustic emission signal; Integrating the time spectrum along the time axis yields the adaptive marginal spectrum for each acoustic emission signal. The signal classification module is used to divide the target domain dataset into a target domain labeled dataset and a target domain unlabeled dataset. It uses "non-destructive" and "with porosity and defects" to label the acoustic emission signals contained in the source domain labeled dataset and the target domain labeled dataset. The target domain labeled dataset contains acoustic emission signals with both non-destructive and porosity and defects. The initial classification module is used to adjust the architecture of Wav2Vec2.0 to obtain the improved AEWav2Vec2.0 model. The AEWav2Vec2.0 model is trained using the adaptive marginal spectrum of the source domain labeled dataset to obtain the initial classification model for identifying the label category of acoustic emission signals. The intermediate classification module is used to divide the target domain labeled dataset into a training set and a validation set according to a preset ratio, and ensure that both the training set and the validation set contain acoustic emission signals of non-destructive conditions and those with porosity defects. The training set and the source domain labeled dataset constitute a hybrid labeled dataset, and the initial model is trained using the adaptive marginal spectrum of the hybrid labeled dataset to obtain the intermediate classification model. The target classification module is used to input the adaptive marginal spectrum of the unlabeled dataset of the target domain into the intermediate classification model. The intermediate classification model outputs two classification results with confidence levels for the corresponding acoustic emission signals. Acoustic emission signals with confidence levels higher than the preset value Q1 are selected, and the selected acoustic emission signals and the mixed labeled dataset constitute a joint training dataset. The intermediate classification model is iteratively optimized using the joint training dataset until the preset conditions are met and the iterative optimization stops, and the target classification model is obtained. The final classification module is used to input the adaptive marginal spectrum of the validation set into the target classification model. The target classification model outputs two classification results with confidence levels for the corresponding acoustic emission signals. Acoustic emission signals with confidence levels higher than the preset value Q4 are selected to form a prediction label dataset. The prediction label dataset is added to the joint training dataset to form the final training dataset. The target classification model is iteratively optimized using the final training dataset. After the preset conditions are met, the iterative optimization stops and the final classification model is obtained. The results output module is used to input the adaptive marginal spectrum of the unlabeled dataset in the target domain into the final classification model. The final classification model outputs the classification result and confidence level of each acoustic emission signal in the unlabeled dataset in the target domain.

Images