Ultrasonic guided wave array detection method and system for orthotropic steel bridge deck cracks

Abstract

The invention relates to an ultrasonic guided wave array detection method and system for orthotropic steel bridge deck cracks, and the method comprises the steps: employing an ultrasonic guided wave generator to apply ultrasonic guided waves propagating in a second direction to a plurality of set areas delimited in a first direction on a steel bridge deck one by one, a signal receiver is adopted to receive a reflection signal of the ultrasonic guided wave; a preset signal analysis model is adopted to carry out intelligent processing on the reflection signals, the intelligent processing comprises signal data preprocessing, first module strength feature analysis based on a dense block convolutional neural network and second module time sequence dependency analysis based on a bidirectional long-short-term memory neural network, and therefore cracks are accurately recognized; in addition, the invention further provides a detection system for executing the method. The crack detection method provided by the invention is simple and convenient to operate, solves the problem that a traditional detection method is difficult to fully cover, and improves the accuracy and efficiency of detection.

Description

Technical Field

[0001] The present application relates to the field of steel bridge decks, and specifically to an ultrasonic guided wave array detection method and system for orthotropic steel bridge deck cracks. Background Art

[0002] Orthotropic steel bridge decks have been widely used in highway and railway bridge construction due to their advantages of light weight, large bearing capacity, and high degree of factory production. Orthotropic steel bridge decks are composed of top plates, bottom plates, U-shaped ribs in the longitudinal direction of the bridge, longitudinal diaphragms, and transverse diaphragms in the transverse direction of the bridge. Figure 2 As shown in the figure, 1 is the diaphragm, 2 is the U rib, and 3 is the top plate. However, its complex structural design and stress characteristics, coupled with the residual stress generated during the welding process and the long-term vehicle load and environmental corrosion during the operation stage, make the orthotropic steel bridge deck prone to fatigue cracks. These cracks are mainly distributed in the connection between the U rib and the top plate, the connection between the U rib and the diaphragm, the U rib butt weld, and the U rib through-weld hole, etc., posing a serious threat to the safety and life of the bridge, especially the bridge deck pavement on the top surface of the orthotropic steel bridge deck "repaired repeatedly, and repaired repeatedly, and damaged again and again", causing adverse social impacts.

[0003] At present, there are mainly the following methods for detecting fatigue cracks in orthotropic steel bridge decks: 1) Machine vision method, which uses industrial cameras to capture images and import computer vision algorithms to identify cracks. It is widely used in bridge engineering, machinery manufacturing and other fields. Its advantages are fast recognition speed and disadvantages of only identifying superficial damage; 2) Acoustic emission method, which is mainly based on the principle of acoustic elasticity. It uses acoustic emission probes and signal acquisition systems to detect cracks by collecting transient elastic wave signals excited when cracks extend. Its advantages are high accuracy and disadvantages. As a passive detection method, it cannot be used to detect fatigue cracks that already exist and are not at the moment of extension; 3) Ultrasonic detection technology, especially technology based on ultrasonic body waves, occupies an important position in the field of industrial non-destructive testing. However, the propagation characteristics of ultrasonic body waves limit their detection range, resulting in the need to adopt point-by-point scanning in practical applications, especially in crack detection of orthotropic steel bridge decks, and the probe can only detect its coverage area or a small range around it, which not only reduces the detection efficiency, but also significantly increases the detection cost.

[0004] At the same time, for the crack detection at the connection between the U-rib inside the orthotropic steel bridge deck and the top plate, the existing TOFD ultrasonic flaw detector and ultrasonic phased array technology are also based on the detection principle of ultrasonic body waves. The point-by-point scanning detection method is not only time-consuming and labor-intensive, but also difficult to ensure the comprehensiveness and accuracy of the detection. There are still shortcomings such as too small scanning range, scanning blind spots, cumbersome detection process and extremely low efficiency.

[0005] In summary, the existing technology has many shortcomings in ultrasonic guided wave array detection of orthotropic steel bridge deck cracks, especially in terms of detection efficiency, cost and comprehensiveness. In particular, there is no effective detection method for invisible cracks in hidden parts of orthotropic steel bridge decks.

[0006] However, ultrasonic guided waves are stress waves propagating along the structural interface formed by the reflection and superposition of body waves at the structural interface. Compared with body waves, ultrasonic guided waves have the characteristics of low attenuation and long propagation distance, which can realize rapid non-destructive testing of large structural parts with regular shapes; however, ultrasonic guided waves are more complex than body waves, which is mainly manifested in two aspects: on the one hand, the multi-modal characteristics of guided waves, that is, there are multiple guided wave modes at the same frequency; on the other hand, the dispersion characteristics, that is, the propagation speed of the same mode guided wave at different frequencies. The complexity of ultrasonic guided waves puts higher requirements on the detection platform and detection methods. Summary of the invention

[0007] The present application provides an ultrasonic guided wave array detection method and system for orthogonal anisotropic steel bridge deck cracks, which can solve the technical problems existing in the above-mentioned prior art. The ultrasonic guided waves used have the characteristics of wide propagation range and long distance. At the same time, they can reflect the real information of structural changes, and can effectively detect and judge whether there are cracks in the unobservable structural area and evaluate the length of the cracks, solving the problem that traditional detection methods are difficult to fully cover, and improving the accuracy and efficiency of detection. At the same time, a preset signal analysis model is used to intelligently process the reflected signal, including signal data preprocessing, first module strength feature analysis based on dense block convolutional neural network, and second module timing dependency analysis based on bidirectional long short-term memory neural network, so as to accurately identify cracks; solving the signal processing problems caused by the multimodal characteristics and dispersion characteristics of ultrasonic guided waves.

[0008] In a first aspect, the present application provides an orthogonal special-shaped steel bridge deck crack detection method, which adopts the following technical solution: An ultrasonic guided wave array detection method for orthotropic steel bridge deck cracks comprises the following steps: Pre-treat the steel structure, including at least removing the surface coating, rust or dirt, to ensure that the ultrasonic guided wave can propagate smoothly and accurately reflect the crack information; Arrange multiple ultrasonic guided wave sensors on the surface of the orthotropic steel bridge deck to be tested, select one ultrasonic guided wave sensor as a transmitting end to transmit ultrasonic guided waves, and the remaining ultrasonic sensors as receiving ends to receive ultrasonic guided waves; wherein the arrangement of the ultrasonic guided wave sensors is determined according to the shape and size of the steel structure to be tested, and the first direction and the second direction are selected to ensure that potential crack areas can be fully covered and detected; The transmitting frequency and receiving sensitivity of the ultrasonic guided wave sensor can be dynamically adjusted according to the material properties of the steel structure and the expected crack size to achieve noise suppression of the received ultrasonic guided wave signal to optimize the detection effect; the received analog ultrasonic guided wave signal is converted into a digital signal through a data acquisition card, and the digital signal is multiplied by a multiplier with an adjustable frequency standard signal to achieve signal modulation, and then low-pass filtering and normalization processing are performed in sequence, and the normalization processing includes at least one of standardization processing, normalization processing, averaging processing or logarithmic transformation; the received multiple groups of ultrasonic guided wave signals are aggregated into a signal matrix; Adaptive filtering algorithm is used to filter the received ultrasonic guided wave signal to dynamically adjust the filtering parameters, so as to more effectively remove noise interference and improve the accuracy of crack identification. Chaotic oscillator analysis is performed on the signal matrix, and the state variables are obtained by solving the Duffing equation and drawing a timing diagram. If the timing diagram shows regular intermittent chaotic phenomena, it is determined that the signal corresponding to the state variable is a weak signal. Extract the frequency and amplitude features of weak signals, and further extract these features using the first module DenseBlock convolution module in the preset signal analysis model. Each convolution layer connects the input features with the output of the previous layer and compares them with the preset standard signal pattern to form a deeper feature representation. By utilizing the time series dependency of the signal, the correlation between these features and the crack length is learned by using the second module BiLSTM module in the preset signal analysis model, so as to obtain the change of the crack signal in the time dimension of the reflection signal and identify the existence of potential cracks; According to the comparison between the change of the crack signal in the time dimension of the reflection signal and the preset standard signal pattern, the crack length of the orthotropic steel bridge deck is evaluated; Determine whether the crack length exceeds the preset threshold, and if so, issue a warning message; When the crack length exceeds the preset threshold, the distance between the crack and the transmitting sensor is calculated based on the flight time of the ultrasonic guided wave (i.e. the time difference between transmission and reception) and the propagation speed of ultrasound in the steel structure, thereby preliminarily determining the crack location; Adopting multi-sensor fusion technology, the ultrasonic guided wave sensor data from different positions and directions are comprehensively processed and analyzed to accurately locate the crack position; After determining the crack location and length, the steel structure model of the crack area is reconstructed using 3D modeling technology to more intuitively display the crack situation and provide a reference for subsequent repair or reinforcement work; Use machine learning algorithms to intelligently classify and identify crack features, and automatically determine the type of crack (including fatigue cracks, welding cracks, etc.) based on the crack's shape, depth, direction and other characteristics; Conduct risk assessment on detected cracks, evaluate the impact of cracks on structural safety based on the size, location, type of cracks, and the service environment and stress state of the steel structure, and provide corresponding maintenance recommendations or early warning information; Establish a crack detection database to record and analyze the results of each test (including crack location, size, type, risk assessment and other information) in order to track the development of cracks, evaluate the long-term effectiveness of detection methods, and provide reference for future detection work; After crack detection is completed, the steel structure is marked or annotated so that the crack position can be quickly located during subsequent maintenance, repair or reinforcement work.

[0009] In combination with the first aspect, in some embodiments, a phased array including a plurality of ultrasonic guided wave sensors is used to excite and receive ultrasonic guided waves, preferably 16 Φ7×0.2mm piezoelectric chip ultrasonic guided wave sensors are used, and each sensor is independent of each other. Preferably, when using ultrasonic guided wave radar for defect imaging detection, it is necessary to first excite one sensor, then record the guided wave signals received by the 16 sensors, then excite another sensor, and then record the wave signals received by the 16 sensors, and finally obtain 16×16 time domain signals, each of which corresponds to an excitation-receiving sensor combination.

[0010] In combination with the first aspect, in some embodiments, when an adaptive filtering algorithm is used to filter the received ultrasonic guided wave signal, it includes signal denoising, normalization and time window segmentation, preferably a 5-cycle sine wave modulated by a Hanning window. By windowing and modulating the excitation signal, the bandwidth of the excitation signal can be reduced, thereby reducing the dispersion effect.

[0011] In combination with the first aspect, in some embodiments, the preset signal analysis model adopts the first module DenseBlock convolution module, which is designed through dense connection, feature map splicing, Bottleneck structure, growth rate control and transition layer. Each convolution layer in it connects the input features with the output of the previous layer to form a deeper feature representation for analyzing the intensity characteristics of the reflected signal.

[0012] In combination with the first aspect, in some embodiments, in a classification task, preferably, the loss function is a cross-entropy loss, the performance optimizer is an RMSProp algorithm optimizer, and the hyperparameter adopts a growth rate as the first parameter; In combination with the first aspect, in some embodiments, the second module BiLSTM module used in the preset signal analysis model includes a neural network model of an embedding layer, a bidirectional LSTM layer, and a fully connected output layer; the analysis model database is composed of a characteristic crack preset set, a verification set, and a test set, the preset set is used to update the parameters of the model, the verification set is used to adjust the hyperparameters of the analysis model, and the test set is used to evaluate the performance of the analysis model. It uses the time series dependency of the signal to analyze the changes of the crack signal of the reflection signal in the time dimension through forward and backward propagation.

[0013] In combination with the first aspect, in some embodiments, the BiLSTM module preferably uses the following code based on Python TensorFlow: model = Sequential([ Embedding(input_dim=10000, output_dim=64, input_length=max_length), #embedding layer Bidirectional(LSTM(64, return_sequences=True)), # Bidirectional LSTM layer, return_sequences=True means outputting the hidden state of each time step Dropout(0.5), # Dropout layer is used to reduce overfitting Bidirectional(LSTM(32)), # Another bidirectional LSTM layer, this time without outputting the hidden state at each time step Dense(len(label_encoder.classes_), activation='softmax') # Fully connected layer, number of output categories]) model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=['accuracy']).

[0014] In combination with the first aspect, in some embodiments, preferably, a penalty is imposed on the weights of the model to prevent the weights from becoming too large, and a Dropout layer and regularization technique are used to reduce overfitting; the Python code using the Dropout layer and L2 regularization in the BiLSTM model is as follows: import tensorflow as tf # Assuming your data has been preprocessed and has the following variables from tensorflow.keras.models import Sequential from tensorflow.keras.layers import Embedding, Bidirectional, LSTM,Dense, Dropout from tensorflow.keras.regularizers import l2 vocab_size = len(tokenizer.word_index) + 1 embedding_dim = 64 max_length = 100 # Example value, should be set according to your data num_classes = len(label_encoder.classes_) # Build the model model = Sequential([ Embedding(input_dim=vocab_size, output_dim=embedding_dim, input_length=max_length, embeddings_regularizer=l2(0.01)), # Use L2 regularization, where 0.01 is the regularization strength.

[0015] Bidirectional(LSTM(64, return_sequences=True, dropout=0.2, recurrent_dropout=0.2)), Dropout(0.5), # Added a separate Dropout layer after the first bidirectional LSTM layer Bidirectional(LSTM(32, dropout=0.2, recurrent_dropout=0.2)), Dense(num_classes, activation='softmax') # Output layer ]) # Compile the model model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=['accuracy']) # Train the model (assuming X_train_pad and y_train are your training data) model.fit(X_train_pad, y_train, epochs=10, batch_size=32, validation_split=0.2).

[0016] In the second aspect, an ultrasonic guided wave array detection system for orthotropic steel bridge deck cracks is provided, which adopts the following technical solutions: An ultrasonic guided wave array detection system for orthotropic steel bridge deck cracks, comprising: A plurality of ultrasonic sensors are arranged on the surface of the orthotropic steel bridge deck to be measured, and are used to apply ultrasonic guided waves propagating in a second direction to a plurality of set areas demarcated along a first direction on the steel structure one by one, and receive reflected signals; The signal processing unit is used to aggregate the received multiple groups of ultrasonic guided wave signals into a signal matrix, perform chaotic oscillator analysis, obtain state variables by solving the Duffing equation, and draw a timing diagram to determine weak signals; The crack assessment module uses a convolutional neural network to extract the frequency and amplitude characteristics of weak signals, and learns the correlation between these characteristics and crack length through a bidirectional long short-term memory network. It processes the reflection signals corresponding to multiple set areas according to a preset signal analysis model to assess the crack length. The early warning module is used to determine whether the crack length exceeds a preset threshold and issue an early warning message when it exceeds the threshold.

[0017] In combination with the second aspect, in some embodiments, the application position and the receiving position of the ultrasonic guided wave are at the same position in the first direction, and are located at two ends of the steel bridge deck in the second direction.

[0018] The beneficial effects brought by the technical solution provided in the embodiments of the present application include: The crack detection method proposed in the present invention is simple and convenient to operate, solves the problem that traditional detection methods are difficult to fully cover, and improves the accuracy and efficiency of detection; the ultrasonic guided waves used have the characteristics of wide propagation range and long distance, and at the same time it can reflect the real information of structural changes, and thus can effectively detect and judge whether cracks exist in unobservable structural areas. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1This is a schematic diagram of a process of an embodiment of a method for detecting cracks in an orthogonal special-shaped steel bridge deck of the present application; Figure 2 This is a schematic diagram of the detection direction of the orthogonal special-shaped steel bridge deck of this application; Figure 3 This is a cross-sectional view of an orthogonal profiled steel bridge deck crack detection method in a detection state in an embodiment of the present application; Figure 4 This is a cross-sectional view of another embodiment of the orthogonal special-shaped steel bridge deck crack detection method in the present application under detection status; Figure 5 A schematic diagram of a deep neural network architecture combining a DenseBlock convolution module and a BiLSTM module used in the orthogonal special-shaped steel bridge deck crack detection method of the present application; Figure 6 This is a schematic diagram of the hardware structure of an orthogonal special-shaped steel bridge deck crack detection device involved in the embodiment of the present application; Reference numerals: 1. Partition; 2. U-rib; 3. Top plate; 4. Top plate pavement layer; 5. Excitation end probe; 6. Receiving end probe; 7. Ultrasonic guided wave excitation system; 8. Receiving system; 9. Self-excited and self-receiving probe; 10. Excitation-receiving system. DETAILED DESCRIPTION

[0020] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0021] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0022] In a first aspect, an embodiment of the present application provides a method for detecting cracks in an orthogonal special-shaped steel bridge deck.

[0023] In one embodiment, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the orthogonal profiled steel bridge deck crack detection method of the present application. Figure 1 As shown in the figure, the crack detection method of orthogonal profiled steel bridge deck includes: S1. Pre-treating the steel structure of the orthogonal profiled steel bridge deck, at least including removing the surface coating, rust or dirt, to ensure that the ultrasonic guided wave can propagate smoothly and accurately reflect the crack information; S2. Arrange multiple ultrasonic guided wave sensors on the surface of the orthotropic steel bridge deck to be tested, select one ultrasonic guided wave sensor as a transmitting end to transmit ultrasonic guided waves, and the remaining ultrasonic sensors as receiving ends to receive ultrasonic guided waves; wherein the arrangement of the ultrasonic guided wave sensors is determined according to the shape and size of the steel structure to be tested, and the first direction and the second direction are selected to ensure that the potential crack area can be fully covered and detected; Specifically, a phased array including multiple ultrasonic guided wave sensors is used to excite and receive ultrasonic guided waves, preferably 16 Φ7×0.2mm piezoelectric chip ultrasonic guided wave sensors are used, and each sensor is independent of each other. Preferably, when using ultrasonic guided wave radar for defect imaging detection, it is necessary to first excite one sensor, then record the guided wave signals received by 16 sensors, then excite another sensor, and then record the wave signals received by 16 sensors, and finally obtain 16×16 time domain signals, each of which corresponds to an excitation-receiving sensor combination.

[0024] S3. The transmitting frequency and receiving sensitivity of the ultrasonic guided wave sensor can be dynamically adjusted according to the material properties of the steel structure and the expected crack size to achieve noise suppression of the received ultrasonic guided wave signal to optimize the detection effect; the received analog ultrasonic guided wave signal is converted into a digital signal through a data acquisition card, and the digital signal is multiplied by a multiplier with an adjustable frequency standard signal to achieve signal modulation, and then low-pass filtering and normalization are performed in sequence, and the normalization processing includes at least one of standardization processing, normalization processing, averaging processing or logarithmic transformation; the received multiple groups of ultrasonic guided wave signals are aggregated into a signal matrix; In the above-mentioned process of transmitting ultrasonic guided waves and receiving reflected signals, ultrasonic guided waves propagating in the second direction are applied one by one to a plurality of set areas demarcated along the first direction on the steel bridge deck, and reflected signals of the ultrasonic guided waves are received; Figure 2 In this embodiment, the first direction is the length direction of the steel bridge deck, that is, the probe moving direction in the figure, and the second direction is the width direction of the steel bridge deck, that is, the waveguide propagation direction in the figure. Figure 3The figure shows a cross-sectional schematic diagram of the on-site arrangement, wherein the steel bridge deck and the detection device involved include U-rib 2, top plate 3, top plate 3 pavement layer excitation end probe 5 and receiving end probe 6. The ultrasonic guided wave excited by the excitation end probe 5 propagates within a certain range and is received by the receiving end probe 6, and fatigue damage cracks within the range are detected. By designing the ultrasonic guided wave excitation system 7 and the receiving system 8, an integrated guided wave detection device with one excitation and one reception is realized. Among them, the excitation system includes a signal generator and a voltage amplifier, which are connected to the excitation end probe 5. The excitation waveform adopts a five-cycle Hanning window amplitude modulated pulse wave, and the specific voltage waveform can be generated by formula (1). The receiving system 8 is mainly composed of an oscilloscope, which is connected to the receiving end probe 6 to realize the display and access of the received signal time course.

[0025] (1) Where t is time, V(t) is the voltage waveform that changes with time, A is the maximum amplitude of the voltage pulse, H(t) is the unit step function, n is the number of cycles contained in the waveform, and fc is the center frequency of the narrowband waveform. The frequency is mainly selected based on the principle that the corresponding waveguide wavelength matches the crack size. For steel plates in actual engineering, the frequency is approximately between 30kHz and 200kHz. In addition, a self-excited and self-collected system can also be used to collect waveguide signals. Figure 4 As shown, in this state, the probe is a self-excited self-receiving probe 9, which can simultaneously excite the waveguide and receive the reflected waveguide signal. And the excitation-receiving system 10 is integrated with a signal generator, a voltage amplifier and an oscilloscope. After the waveguide excitation-receiving operation is completed once, as shown in FIG. Figure 2 , 3 As shown, the excitation end probe 5 and the receiving end probe 6 are moved longitudinally (in the embodiment of the self-excitation and self-receiving system, only the self-excitation and self-receiving probe 9 is moved, as shown in FIG. Figure 2 , 4 As shown in the figure, the moving distance is determined by the resolution of each measurement. After moving to a new position, the above measurement process is repeated to collect ultrasonic guided wave signals at different positions, and the received signal time history is stored through the oscilloscope to form a received signal database.

[0026] S4. Adopt an adaptive filtering algorithm to filter the received ultrasonic guided wave signal to dynamically adjust the filtering parameters, so as to more effectively remove noise interference and improve the accuracy of crack identification; perform chaotic oscillator analysis on the signal matrix, obtain the state variable by solving the Duffing equation, and draw a timing diagram. If the timing diagram shows regular intermittent chaotic phenomena, it is determined that the signal corresponding to the state variable is a weak signal; Specifically, when the received ultrasonic guided wave signal is filtered using an adaptive filtering algorithm, including signal denoising, normalization and time window segmentation, a 5-cycle sine wave modulated by a Hanning window is preferably used. By windowing and modulating the excitation signal, the bandwidth of the excitation signal can be reduced, thereby reducing the dispersion effect.

[0027] S5, extract the frequency and amplitude features of weak signals, and use the convolutional neural network DenseBlock convolution module to further extract these features. Each convolution layer connects the input features with the output of the previous layer and compares them with the preset standard signal pattern to form a deeper feature representation; S6. By utilizing the time dependency of the signal, the correlation between these features and the crack length is learned through the BiLSTM module of the forward and reverse propagation of ultrasonic guided waves, the change of the crack signal in the time dimension of the reflection signal is obtained, and the existence of potential cracks is identified; In step S5-S6 of this embodiment, the first module adopted by the preset signal analysis model is a DenseBlock convolution module, which connects the input features with the output of the previous layer through dense connection, feature map concatenation, Bottleneck structure, growth rate control and transition layer design, forming a deeper feature representation for analyzing the intensity characteristics of the reflected signal. In the classification task, preferably, the loss function is the cross-entropy loss, the performance optimizer is the RMSProp algorithm optimizer, and the hyperparameter adopts the growth rate as the first parameter.

[0028] The second module used in the preset signal analysis model is the BiLSTM module, which includes a neural network model of an embedding layer, a bidirectional LSTM layer, and a fully connected output layer; the analysis model database consists of a characteristic crack preset set, a verification set, and a test set. The preset set is used to update the parameters of the model, the verification set is used to adjust the hyperparameters of the analysis model, and the test set is used to evaluate the performance of the analysis model. It uses the temporal dependency of the signal to analyze the changes in the crack signal of the reflection signal in the time dimension through forward and backward propagation.

[0029] Specifically, corresponding to Figure 5The deep neural network architecture shown in the figure combines the DenseBlock convolution module and the Bidirectional Long Short-Term Memory (BiLSTM) network module to achieve efficient processing of multi-dimensional signal data with time series characteristics. The network input is the signal at the receiving end of the test piece, which can determine whether the test piece is damaged. This network architecture takes advantage of the convolutional neural network (CNN) and BiLSTM to ensure the consistency of the input and output signal shapes, thereby improving the model's expressiveness and the effect of processing complex time series data.

[0030] If the input of DenseBlock is x 0, then in l The output of the layer is expressed as formula (1):

[0031] In the formula, Indicates the concatenation of the output feature maps of all previous layers.

[0032] Nonlinear transformation It usually consists of Batch Normalization (BN), ReLU activation function and convolution operation (Conv), as shown in formula (2).

[0033]

[0034] Next, the features processed by DenseBlock are input to the BiLSTM module. BiLSTM can capture the temporal dependency of the signal and analyze the signal more comprehensively by taking into account the previous and next information through forward and backward propagation. After LSTM processing, the model can generate a sequence output that reflects the changes in the crack signal in the time dimension. The final result will be used to determine the trend and change pattern of the reflection signal and identify the potential existence of cracks by comparing it with the normal signal pattern.

[0035] BiLSTM (Bidirectional Long Short-Term Memory) is a bidirectional long short-term memory network used to process sequence data to capture the forward and backward information of each time step in the sequence. Its mathematical representation is shown in equations (3) to (8):

[0036] In the formula, , , , They are forget gate, input gate, candidate memory, and output gate. is the sigmoid activation function, Represents element-wise multiplication (Hadamard product).

[0037] BiLSTM is composed of forward LSTM and backward LSTM, which shows how BiLSTM captures the complete information of each time step in the sequence by combining forward and backward LSTM units. Its mathematical expression can be shown in equations (9) to (11):

[0038] In the formula, is the hidden state of the forward LSTM at each time step, is the hidden state of the forward LSTM at each time step.

[0039] The advantages of DenseBlock and BiLSTM modules are that DenseBlock achieves efficient reuse of features and smooth flow of gradients through dense connections between layers, while BiLSTM captures complete information of each time step in the sequence by combining forward and backward LSTM units. In the deep neural network architecture proposed in the present invention, these two modules are organically combined to give full play to their advantages in feature extraction and time series modeling.

[0040] Specifically, each input signal received by the network has the shape of L × C , indicating that the signal length is L , the number of channels is C . The output dimension of the network is 2, which are the network predicted damage probability and no-damage probability. After receiving the data, the input layer passes it to the DenseBlock module for preliminary feature extraction. DenseBlock passes the output of each layer directly to all subsequent layers through a dense connection mechanism, thereby ensuring the smooth flow of gradients during back propagation and alleviating the gradient vanishing problem. In addition, the design of DenseBlock enables the low-level features extracted by the early layers to be directly used by the subsequent layers, improving the efficiency of feature reuse and reducing the number of model parameters while maintaining efficient feature transfer.

[0041] Next, DenseBlock 2 receives the output of DenseBlock 1. In this process, the dense connection mechanism of DenseBlock further improves the efficiency of feature extraction and ensures efficient reuse and transmission of features. The output features are then passed through a convolutional layer to obtain the same number of channels as the input. In order to further enhance the effect of feature extraction and time series modeling, the Skip Connection mechanism is used in the network to connect the feature map with the input signal. This skip connection method ensures that the original input features can still be directly transmitted to the BiLSTM module after being processed by DenseBlock, thereby enhancing the diversity and integrity of the features.

[0042] Finally, the temporal dependencies in the data are captured through a bidirectional long short-term memory network. BiLSTM can not only process forward temporal dependencies, but also capture reverse temporal dependencies, providing more comprehensive temporal information. The memory cells in LSTM can store information for a long time, avoid long-distance dependency problems, and further improve the generalization ability of the model. The output of BiLSTM is converted through a fully connected layer to achieve the binary classification task.

[0043] In the classification task, preferably, the loss function is cross-entropy loss, the performance optimizer is RMSProp algorithm optimizer, and the hyperparameter adopts growth rate as the first parameter.

[0044] The DenseBlock module improves the gradient flow through the dense connection mechanism, ensures the efficient reuse of features, reduces the number of parameters, and improves the computational efficiency. The BiLSTM module enhances the model's understanding and generalization capabilities of time series data by capturing the long-term and short-term dependencies and bidirectional time information of the data. The Skip Connection mechanism further ensures the diversity and integrity of the input features. Combining these advantages, the network architecture of the present invention can efficiently process and classify complex time series data and achieve accurate prediction tasks.

[0045] Finally, DenseBlock can effectively extract features from the signal through multiple convolutional layers. Each convolutional layer connects the input features with the output of the previous layer to form a deeper feature representation. This module improves the utilization of features through dense connections, allowing the model to learn more complex signal patterns. The processed results will obtain the intensity information of different frequency components, which helps to determine the abnormal characteristics of the reflected signal and then preliminarily identify areas where cracks may exist.

[0046] Furthermore, the BiLSTM module preferably uses the following code based on Python TensorFlow: model = Sequential([ Embedding(input_dim=10000, output_dim=64, input_length=max_length), #embedding layer Bidirectional(LSTM(64, return_sequences=True)), # Bidirectional LSTM layer, return_sequences=True means outputting the hidden state of each time step Dropout(0.5), # Dropout layer is used to reduce overfitting Bidirectional(LSTM(32)), # Another bidirectional LSTM layer, this time without outputting the hidden state at each time step Dense(len(label_encoder.classes_), activation='softmax') # Fully connected layer, number of output categories]) model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=['accuracy']).

[0047] S7, evaluating the crack length of the orthotropic steel bridge deck according to the comparison between the change of the crack signal of the reflection signal in the time dimension and the preset standard signal pattern; In the crack length fitting process, preferably, a penalty is imposed on the weight of the model to prevent the weight from becoming too large, and a Dropout layer and regularization technique are used to reduce overfitting; the python code using the Dropout layer and L2 regularization in the BiLSTM model is as follows: import tensorflow as tf # Assuming your data has been preprocessed and has the following variables from tensorflow.keras.models import Sequential from tensorflow.keras.layers import Embedding, Bidirectional, LSTM,Dense, Dropout from tensorflow.keras.regularizers import l2 vocab_size = len(tokenizer.word_index) + 1 embedding_dim = 64 max_length = 100 # Example value, should be set according to your data num_classes = len(label_encoder.classes_) # Build the model model = Sequential([ Embedding(input_dim=vocab_size, output_dim=embedding_dim, input_length=max_length, embeddings_regularizer=l2(0.01)), # Use L2 regularization, where 0.01 is the regularization strength.

[0048] Bidirectional(LSTM(64, return_sequences=True, dropout=0.2, recurrent_dropout=0.2)), Dropout(0.5), # Added a separate Dropout layer after the first bidirectional LSTM layer Bidirectional(LSTM(32, dropout=0.2, recurrent_dropout=0.2)), Dense(num_classes, activation='softmax') # Output layer ]) # Compile the model model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=['accuracy']) # Train the model (assuming X_train_pad and y_train are your training data) model.fit(X_train_pad, y_train, epochs=10, batch_size=32, validation_split=0.2).

[0049] S8, determining whether the crack length exceeds a preset threshold, and if so, issuing a warning message; S9. When the crack length exceeds a preset threshold, the distance between the crack and the transmitting sensor is calculated based on the flight time of the ultrasonic guided wave (i.e., the time difference between the transmission and reception) and the propagation speed of ultrasound in the steel structure, thereby preliminarily determining the crack location; S10. Use multi-sensor fusion technology to comprehensively process and analyze ultrasonic guided wave sensor data from different positions and directions to accurately locate the crack position; S11. After determining the crack location and length, use 3D modeling technology to reconstruct the steel structure model of the crack area to more intuitively display the crack situation and provide a reference for subsequent repair or reinforcement work; S12. Use machine learning algorithms to intelligently classify and identify crack features, and automatically determine the type of crack (including fatigue cracks, welding cracks, etc.) based on the crack's morphology, depth, direction and other characteristics; S13. Conduct risk assessment on detected cracks, evaluate the impact of cracks on structural safety based on the size, location, type of cracks, and the service environment and stress state of the steel structure, and provide corresponding maintenance recommendations or early warning information; S14. Establish a crack detection database to record and analyze the results of each test (including crack location, size, type, risk assessment and other information) in order to track the development of cracks, evaluate the long-term effectiveness of detection methods, and provide reference for future detection work; S15. After crack detection is completed, the steel structure shall be marked or annotated so that the crack position can be quickly located during subsequent maintenance, inspection or reinforcement work.

[0050] In a second aspect, an embodiment of the present application also provides an orthogonal special-shaped steel bridge deck crack detection system.

[0051] In one embodiment, an orthogonal profiled steel bridge deck crack detection system includes: A plurality of ultrasonic sensors are arranged on the surface of the orthotropic steel bridge deck to be measured, and are used to apply ultrasonic guided waves propagating in a second direction to a plurality of set areas demarcated along a first direction on the steel structure one by one, and receive reflected signals; The signal processing unit is used to aggregate the received multiple groups of ultrasonic guided wave signals into a signal matrix, perform chaotic oscillator analysis, obtain state variables by solving the Duffing equation, and draw a timing diagram to determine weak signals; The crack assessment module uses a convolutional neural network to extract the frequency and amplitude characteristics of weak signals, and learns the correlation between these characteristics and crack length through a bidirectional long short-term memory network. It processes the reflection signals corresponding to multiple set areas according to a preset signal analysis model to assess the crack length. The early warning module is used to determine whether the crack length exceeds a preset threshold and issue an early warning message when it exceeds the threshold.

[0052] Furthermore, the application position and the receiving position of the ultrasonic guided wave are at the same position in the first direction, and are respectively located at two ends of the steel bridge deck in the second direction.

[0053] In a third aspect, an embodiment of the present application provides an orthogonal special-shaped steel bridge deck crack detection device, which can be a personal computer (PC), a laptop computer, a server, or other device with data processing capabilities.

[0054] Reference Figure 6 , Figure 6 The hardware structure diagram of the orthogonal profiled steel bridge deck crack detection device involved in the embodiment of the present application is shown in FIG. In the embodiment of the present application, the orthogonal profiled steel bridge deck crack detection device may include a processor, a memory, a communication interface, and a communication bus.

[0055] The communication bus may be of any type and is used to interconnect the processor, the memory, and the communication interface.

[0056] The communication interface includes input / output (I / O) interface, physical interface and logical interface, etc., which are used to realize the interconnection of devices inside the orthogonal profiled steel bridge deck crack detection device, and the interface used to realize the interconnection of the orthogonal profiled steel bridge deck crack detection device with other devices (such as other computing devices or user devices). The physical interface can be an Ethernet interface, a fiber optic interface, an ATM interface, etc.; the user device can be a display, a keyboard, etc.

[0057] The memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0058] The processor may be a general-purpose processor, which may call the orthogonal profiled steel bridge deck crack detection program stored in the memory and execute the orthogonal profiled steel bridge deck crack detection method provided in the embodiment of the present application. For example, the general-purpose processor may be a central processing unit (CPU). The method executed when the orthogonal profiled steel bridge deck crack detection program is called may refer to the various embodiments of the orthogonal profiled steel bridge deck crack detection method of the present application, which will not be described in detail here.

[0059] Those skilled in the art will understand that Figure 6 The hardware structure shown in the figure does not constitute a limitation on the present application, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0060] The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices. The terms "first", "second" and "third" are used to distinguish different objects, etc., and do not represent a sequence, nor do they limit "first", "second" and "third" to different types.

[0061] In the description of the embodiments of the present application, "exemplary", "for example" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary", "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary", "for example" or "for example" is intended to present related concepts in a specific way.

[0062] In the description of the embodiments of the present application, unless otherwise specified, “ / ” means or, for example, A / B can mean A or B; the “and / or” in the text is merely a description of the association relationship of associated objects, indicating that three relationships may exist, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, “multiple” refers to two or more than two.

[0063] In some processes described in the embodiments of the present application, multiple operations or steps that appear in a specific order are included, but it should be understood that these operations or steps may not be executed in the order in which they appear in the embodiments of the present application or in parallel, and the sequence number of the operation is only used to distinguish the different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed in sequence or in parallel, and these operations or steps may be combined.

[0064] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD) as described above, and includes a number of instructions for a terminal device to execute the methods described in each embodiment of the present application.

[0065] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. An ultrasonic guided wave array detection method for orthotropic steel bridge deck cracks, characterized in that: The following steps are involved: S1: Pre-treat the steel structure, including at least removing the surface coating, rust or dirt, to ensure that the ultrasonic guided wave can propagate smoothly and accurately reflect the crack information; S2: Arrange multiple ultrasonic guided wave sensors on the surface of the orthotropic steel bridge deck to be tested, select one ultrasonic guided wave sensor as a transmitting end to transmit ultrasonic guided waves, and the remaining ultrasonic sensors as receiving ends to receive ultrasonic guided waves; wherein the arrangement of the ultrasonic guided wave sensors is determined according to the shape and size of the steel structure to be tested, and the first direction and the second direction are selected to ensure that the potential crack area can be fully covered and detected; S3: The transmitting frequency and receiving sensitivity of the ultrasonic guided wave sensor can be dynamically adjusted according to the material properties of the steel structure and the expected crack size to achieve noise suppression of the received ultrasonic guided wave signal to optimize the detection effect; the received analog ultrasonic guided wave signal is converted into a digital signal through a data acquisition card, and the digital signal is multiplied by a multiplier with an adjustable frequency standard signal to achieve signal modulation, and then low-pass filtering and normalization processing are performed in sequence, and the normalization processing includes at least one of standardization processing, regularization processing, averaging processing or logarithmic transformation; the received multiple groups of ultrasonic guided wave signals are aggregated into a signal matrix; S4: Adopting adaptive filtering algorithm to filter the received ultrasonic guided wave signal, dynamically adjusting filtering parameters, so as to more effectively remove noise interference and improve the accuracy of crack identification; performing chaotic oscillator analysis on the signal matrix, obtaining state variables by solving Duffing equation, and drawing a timing diagram. If the timing diagram shows regular intermittent chaotic phenomenon, it is determined that the signal corresponding to the state variable is a weak signal; S5: Extract the frequency and amplitude features of weak signals, and further extract these features using the first module DenseBlock convolution module in the preset signal analysis model. Each convolution layer connects the input features with the output of the previous layer and compares them with the preset standard signal pattern to form a deeper feature representation. S6: utilizing the temporal dependency of the signal, by using the second module BiLSTM module in the preset signal analysis model to learn the correlation between these features and the crack length, obtaining the change of the crack signal in the time dimension of the reflection signal, and identifying the existence of potential cracks; S7: evaluating the crack length of the orthotropic steel bridge deck according to the comparison of the change of the crack signal of the reflection signal in the time dimension with the preset standard signal pattern; S8: judging whether the crack length exceeds a preset threshold, and if so, issuing a warning message; S9: When the crack length exceeds the preset threshold, the distance between the crack and the transmitting sensor is calculated according to the flight time of the ultrasonic guided wave (i.e., the time difference between the transmission and reception) and the propagation speed of the ultrasound in the steel structure, so as to preliminarily determine the crack location; S10: Using multi-sensor fusion technology, the ultrasonic guided wave sensor data from different positions and directions are comprehensively processed and analyzed to accurately locate the crack position; S11: After determining the crack location and length, use 3D modeling technology to reconstruct the steel structure model of the crack area to more intuitively display the crack situation and provide a reference for subsequent repair or reinforcement work; S12: Use machine learning algorithms to intelligently classify and identify crack features, and automatically determine the type of crack (including fatigue cracks, welding cracks, etc.) based on the crack's morphology, depth, direction and other characteristics; S13: Conduct risk assessment on detected cracks, evaluate the impact of cracks on structural safety based on the size, location, type of cracks, and the service environment and stress state of the steel structure, and provide corresponding maintenance recommendations or early warning information; S14: Establish a crack detection database to record and analyze the results of each test (including crack location, size, type, risk assessment and other information) in order to track the development of cracks, evaluate the long-term effectiveness of the detection method, and provide reference for future detection work; S15: After crack detection is completed, the steel structure shall be marked or annotated so that the crack position can be quickly located during subsequent maintenance, inspection or reinforcement work.

2. The ultrasonic guided wave array detection method for orthotropic steel bridge deck cracks according to claim 1, characterized in that: A phased array including multiple ultrasonic guided wave sensors is used for the excitation and reception of ultrasonic guided waves. Preferably, 16 Φ7×0.2mm piezoelectric chip ultrasonic guided wave sensors are used, and each sensor is independent of each other. When using ultrasonic guided wave radar for defect imaging detection, it is necessary to first excite one sensor, then record the guided wave signals received by the 16 sensors, then excite another sensor, and then record the wave signals received by the 16 sensors. Finally, 16×16 time domain signals will be obtained, and each time domain signal corresponds to an excitation-receiving sensor combination.

3. The ultrasonic guided wave array detection method for orthotropic steel bridge deck cracks according to claim 1, characterized in that: When the received ultrasonic guided wave signal is filtered using an adaptive filtering algorithm, including signal denoising, normalization and time window segmentation, specifically a 5-cycle sine wave modulated by a Hanning window, the frequency bandwidth of the excitation signal can be reduced by windowing the excitation signal, thereby reducing the dispersion effect.

4. The ultrasonic guided wave array detection method for orthotropic steel bridge deck cracks according to claim 1, characterized in that: The first module DenseBlock convolution module adopted by the preset signal analysis model has dense connection, feature map splicing, Bottleneck structure, growth rate control and transition layer design. Each convolution layer in it connects the input features with the output of the previous layer to form a deeper feature representation for analyzing the intensity characteristics of the reflected signal.

5. The ultrasonic guided wave array detection method for orthotropic steel bridge deck cracks according to claim 4, characterized in that: In the classification task, the loss function is the cross-entropy loss, the performance optimizer is the RMSProp algorithm optimizer, and the hyperparameter uses the growth rate as the first parameter.

6. The ultrasonic guided wave array detection method for orthotropic steel bridge deck cracks according to claim 1, characterized in that: The second module BiLSTM module adopted by the preset signal analysis model includes an embedding layer, a bidirectional LSTM layer and a neural network model of a fully connected output layer; the analysis model database consists of a characteristic crack preset set, a verification set and a test set, the preset set is used to update the parameters of the model, the verification set is used to adjust the hyperparameters of the analysis model, and the test set is used to evaluate the performance of the analysis model; it utilizes the timing dependency of the signal to analyze the changes of the crack signal of the reflection signal in the time dimension through forward and backward propagation.

7. An ultrasonic guided wave array detection system for orthotropic steel bridge deck cracks, characterized in that: include: 1) A plurality of ultrasonic sensors are arranged on the surface of the orthotropic steel bridge deck to be measured, and are used to apply ultrasonic guided waves propagating in a second direction to a plurality of set areas demarcated along a first direction on the steel structure one by one, and receive reflected signals; 2) Signal processing unit, which is used to aggregate the received multiple groups of ultrasonic guided wave signals into a signal matrix, perform chaotic oscillator analysis, obtain state variables by solving the Duffing equation, and draw a timing diagram to determine weak signals; 3) Crack assessment module, which uses convolutional neural networks to extract the frequency and amplitude characteristics of weak signals, and learns the correlation between these characteristics and crack length through a bidirectional long short-term memory network. It processes the reflection signals corresponding to multiple set areas according to the preset signal analysis model to evaluate the crack length; 4) Early warning module, used to determine whether the crack length exceeds the preset threshold and issue an early warning message when it exceeds the threshold.

8. The ultrasonic guided wave array detection system for orthotropic steel bridge deck cracks according to claim 7, characterized in that: The application position and the receiving position of the ultrasonic guided wave are at the same position in the first direction, and are respectively located at two ends of the steel bridge deck in the second direction.

Images