A method and system for detecting prestressed steel cylinder concrete pipe prestressed reinforcement fracture impact elastic wave

Abstract

The present invention discloses a method and system for detecting prestressed tendon fractures in prestressed composite steel cylinder concrete pipes using impact elastic waves. The method comprises the following steps: S1: obtaining the surface wave velocity of the pipe core concrete; S2: obtaining the signal's dominant frequency, P-wave arrival time, and P-wave velocity; S3: calculating the component thickness at the measuring point; S4: executing S1-S3 for all measuring points within the measuring area to obtain the P-wave velocity, dominant frequency, P-wave arrival time, and component thickness at all measuring points, forming a data set; and S5: reconstructing the P-wave velocity, dominant frequency, P-wave time, component thickness, and whether the label is broken into a time series, designing a two-layer LSTM network, and training the LSTM model. The present invention utilizes machine learning technology to comprehensively consider the changes in the P-wave velocity of the pipe core concrete caused by prestressed tendon fracture and the effects of the spalling of the external mortar protective layer on the component thickness, thereby detecting the location of prestressed tendon fractures in prestressed composite steel cylinder concrete pipes.

Description

Technical Field

[0001] The invention relates to a method and system for detecting the breakage of prestressed reinforcement bars in a prestressed composite steel cylinder concrete pipe using impact elastic waves, and belongs to the field of detection. Background Art

[0002] Prestressed composite concrete cylinder pipe (PCCP) is a composite structural pipe consisting of a concrete core, a steel cylinder, metallic or non-metallic high-strength prestressed reinforcement, and an outer mortar protective layer. Currently, prestressed concrete cylinder pipe (PCCP) is widely used in engineering projects. Due to its high strength, high impermeability, excellent sealing, strong seismic resistance, excellent durability, and low maintenance costs, over 26,000 km of PCCP has been installed in China. Technological advances have led to the development of PCCP made from non-metallic composite materials, such as basalt fiber reinforcement, to replace prestressing steel wire.

[0003] Prestressed tendon failure detection technology involves draining the water from the pipeline during maintenance, allowing inspectors to enter the pipe and perform the inspection. This eliminates the need for pre-installed monitoring equipment. Key methods include acoustic testing and electromagnetic testing.

[0004] The acoustic method is one of the earliest traditional methods used to detect prestressed tendon fractures in prestressed steel cylinder concrete pipes. Its basic principle is that technicians tap on the pipe and determine whether the prestressed tendon is broken based on the echo characteristics of the pipe. This method relies on the subjective experience of the inspector and has the advantages of simple operation and low cost, but it also has obvious limitations: the accuracy of the test results is highly dependent on the technician's experience and is easily affected by human factors; the pipe must be tapped section by section, which is a huge workload and difficult to apply to rapid inspection of long-distance pipelines; the location of prestressed tendon fractures cannot be accurately located, and it is difficult to identify prestressed tendon fractures in their early stages. Although the acoustic method played a certain role in early applications, with the development of detection technology, it has gradually been replaced by more advanced detection methods.

[0005] There are two main electromagnetic testing methods: far-field eddy current testing and orthogonal electromagnetic testing. Far-field eddy current testing utilizes eddy current signals induced in the pipe wall by releasing a low-frequency alternating magnetic field within the pipe. When the eddy current signal encounters defects such as prestressed tendon fractures, its amplitude and phase are distorted. By analyzing the received signal, the location and number of prestressed tendon fractures can be determined. Remote-field eddy current testing was originally developed by Pure and PICC in Canada in the late 20th century and successfully applied to the detection of prestressed tendon fractures in prestressed concrete cylinder pipes. This method, based on eddy current principles, detects prestressed tendon fractures by analyzing impedance fluctuations based on magnetic flux changes, enabling the prediction of prestressed steel wire damage. However, the remote-field eddy current method has certain limitations: When the distance between the exciter and the detection probe is large, the signal attenuates significantly, making long-distance detection difficult; and it is susceptible to environmental noise and uneven pipe metal material. Orthogonal electromagnetic testing is an improved version of remote-field eddy current testing, offering a wider detection range and higher signal recognition capabilities. The basic principle is to improve detection accuracy by reducing direct magnetic flux coupling and noise interference through orthogonal arrangements of emitters and detectors. Field tests verified the effectiveness of orthogonal electromagnetic detection technology by analyzing the signal characteristics of prestressed tendon failure at different excitation frequencies. Finite element simulations were used to investigate the effect of tendon failure location on the detection signal, and the accuracy of the simulation results was verified through experiments.

[0006] Acoustic emission monitoring is a real-time, dynamic, nondestructive monitoring technology that detects the occurrence and location of prestressed tendon failure by capturing elastic wave signals released during the fracture process. Research on this technology began in the 1950s, with the first classification of acoustic emission signals into burst and continuous types, laying the foundation for its application. In the field of prestressed steel cylinder concrete pipe (CCC) fracture detection, Pure, a Canadian company, has developed a sophisticated acoustic emission detection system. By deploying probes within the pipe wall to collect acoustic signals, it can determine the location and number of prestressed tendon failures. However, the application of this technology in practical engineering projects requires further verification. While acoustic emission detection offers the advantages of dynamically monitoring the onset of prestressed tendon failure and detecting early signs of prestressed tendon failure, it also has the disadvantages of significant influence on detection accuracy due to the placement and number of probes, and potential interference from pipe material and ambient noise during signal propagation.

[0007] The hydroacoustic detection method uses hydrophones or acoustic sensors to capture the hydroacoustic signals generated during the prestressed tendon breakage process, thereby locating the prestressed tendon breakage. Depending on the different sensor deployment methods, the hydroacoustic detection method can be divided into three forms: hydrophone base station, hydrophone array, and pipe surface sensor. By deploying fiber optic Bragg grating sensors on the inner wall of prestressed concrete cylinder pipes and combining them with a cantilever beam device with micro-strain sensitivity, the accuracy of prestressed tendon breakage monitoring is improved. Based on fiber optic Bragg grating sensing technology, a system for monitoring prestressed tendon breakage and leakage in prestressed concrete cylinder pipes has been developed and successfully applied in the South-to-North Water Diversion Project. The hydroacoustic detection method has demonstrated high accuracy in locating prestressed tendon breakage, but its application is still limited by issues such as signal attenuation and deployment costs.

[0008] "Bridge cable broken wire signal identification method and system based on long short-term memory network" CN114595733B and "A new prestressed steel tube concrete pipe with embedded acoustic emission sensor and distributed optical fiber and its manufacturing and broken wire monitoring method" CN118465078B both disclose a technical solution for monitoring broken wires. However, this technical solution requires the deployment of monitoring equipment in advance. It is difficult to implement the detailed information of whether there are broken wires in existing pipelines. Summary of the Invention

[0009] Purpose of the invention: In order to overcome the shortcomings of the prior art, the present invention provides a method for detecting prestressed tendon breakage in prestressed composite steel cylinder concrete pipes using impact elastic waves. The method uses machine learning technology to comprehensively consider the change in P-wave velocity of the pipe core concrete caused by prestressed tendon breakage and the influence of the peeling of the external mortar protective layer on the thickness of the component, and can detect the location of prestressed tendon breakage in prestressed composite steel cylinder concrete pipes.

[0010] Technical solution: To solve the above technical problems, the present invention provides a method for detecting prestressed tendon fracture in prestressed composite steel cylinder concrete pipes using impact elastic waves, comprising the following steps:

[0011] S1: Delineate the measurement area and measurement line in the experimental prestressed steel cylinder concrete pipe (the experimental section must be able to meet the following requirements: artificial setting of prestressed tendon breakage, or allow excavation to verify prestressed tendon breakage, that is, to determine whether the measuring point is broken or not), carry out the impact elastic wave method surface wave velocity test, and obtain the surface wave velocity of the pipe core concrete ;

[0012] S2: Assemble the excitation device and probe of the shock elastic wave instrument into a shock elastic wave thickness measuring device, carry out shock elastic wave testing in the prestressed steel cylinder concrete pipe, obtain the time domain signal of the measuring point, and perform preprocessing to obtain the excellent frequency of the signal , P wave arrival time ;

[0013] S3: The P wave velocity of the core concrete obtained in steps S1 and S2 , excellent frequency of measuring points Convert the thickness of the component at the measuring point into the following formula: , It is an empirical parameter and is taken as 0.96 in plate-like components;

[0014] S4: Execute S1-S3 for all measuring points in the survey area to obtain the P-wave velocity, predominant frequency, P-wave arrival time, and component thickness of all measuring points to form a data set. Preprocess the data set to remove outliers (such as records with a velocity of 0 or outside a reasonable range). Use the original features (P-wave velocity, predominant frequency, P-wave arrival time, and component thickness) directly as the feature parameters of the data set. Split the training set, validation set, and test set into a ratio of 7:2:1 to ensure that the test set represents the actual distribution.

[0015] S5: Reconstruct the four characteristic features (P-wave velocity, dominant frequency, P-wave time, component thickness, and one label (wire breakage)) into a time series using a sliding window to adapt it to the input of the LSTM model. A two-layer LSTM network was designed. The first layer extracted local waveform features, and the second layer integrated global time series patterns to output the probability of prestressed tendon breakage. The test set was then fed into the LSTM model for training.

[0016] In the present invention, the label The raw data in does not need to be standardized, and the component label sample input matrix When the prestressed tendon at the measuring point is broken Take 1, when the prestressed steel bar at the measuring point is intact Take 0, which is the final feature input to the LSTM model for ; Input label for The change in component thickness caused by prestressed tendon breakage will significantly affect the dominant frequency, P-wave arrival time, and estimated component thickness. The change in stress in the pipe core concrete caused by prestressed tendon breakage will affect the P-wave velocity. Therefore, the present invention uses P-wave velocity, dominant frequency, P-wave arrival time, and component thickness as characteristic parameters of the LSTM model, comprehensively considering the impact of prestressed tendon breakage on the pipeline structure and material.

[0017] As an advantage, the surface wave velocity in step S1 is The test method is as follows: Measure the length of a measuring line on the measuring line The measuring line segment is arranged with a shock elastic wave exciter at one end point and a signal receiver at the other end point. A signal is sent on the host to make the exciter send a shock elastic wave signal, and the signal receiver receives the surface wave signal. The surface wave signal received by the signal receiver is analyzed on the host to obtain the surface wave propagation time. Calculate the surface wave velocity using the formula , then the calculation formula for the surface wave velocity is: , the P wave velocity can be calculated by the formula : ,in is Poisson's ratio.

[0018] As a preferred method, the thickness measurement method using the impact elastic wave method is as follows: draw a measuring point every 5 cm on the measuring line, align the marking point of the impact elastic wave device with the measuring point, press the device to make the vibrator and the signal receiver in good contact with the concrete surface of the core pipe, press the switch to make the host send an electrical signal, the vibrator hits the concrete surface of the core pipe after receiving the electrical signal, the signal receiver receives the echo time domain signal, analyze the time domain signal of the measuring point on the host, find the first peak of the time domain signal, and calibrate its time as the P wave arrival time. , perform fast Fourier transform on the time domain signal to obtain the frequency domain diagram of the signal, and find the dominant frequency in the frequency domain diagram .

[0019] Preferably, step S5 specifically includes the following steps:

[0020] S51. Input feature preprocessing

[0021] The LSTM model inputs four key features and one label of the prestressed steel cylinder concrete pipe measurement points: P-wave velocity, dominant frequency, P-wave time, component thickness, and wire breakage. The raw data is first cleaned to remove outliers caused by sensor failure or environmental interference (such as data with a velocity of 0 or outside the reasonable range of 3000-5000 m / s). Missing values ​​are filled using linear interpolation of adjacent measurement points. Subsequently, each feature is normalized using Z-score standardization to eliminate the impact of dimensional differences on the model. The standardization formula is:

[0022]

[0023] in is the characteristic mean, is the standard deviation, and the shape of the standardized data matrix is ,in is the number of measuring points.

[0024] S52. Time series construction

[0025] Since LSTM needs to process time series data, and the single point feature has no time dimension, the sliding window method is used to convert adjacent The data of the measurement points are spliced ​​into a time series sample to generate an input matrix with the shape of (N is any number between 10 and 100);

[0026] S53. Model Architecture Design

[0027] The LSTM model adopts a two-layer network structure, and the specific configuration is as follows:

[0028] First LSTM layer: 64 neurons, input shape is , set return_sequences=True to output the complete timing results for processing by the next layer. This layer captures local timing dependencies through the gating mechanism (forget gate, input gate, output gate), such as the short-term fluctuation relationship between wave speed and stress;

[0029] Dropout layer: dropout rate is 0.2, randomly blocking some neurons to prevent overfitting;

[0030] Second LSTM layer: 32 neurons, does not return sequences (return_sequences=False), extracts global temporal features and compresses them into a single vector;

[0031] Fully connected layer: 16 ReLU activated neurons to further integrate high-order features;

[0032] Output layer: 1 Sigmoid-activated neuron, outputting the probability value of prestressed tendon breakage (0-1);

[0033] S54. Model Training and Optimization

[0034] Loss function and optimizer:

[0035] The binary cross entropy loss (binary_crossentropy) is used for the binary classification task, and the mean square error (MSE) is used for the regression task;

[0036] The optimizer is Adam, the initial learning rate is set to 0.001, and the dynamic adjustment strategy is implemented through the learning rate scheduler ReduceLROnPlateau (monitoring validation loss, patience = 5);

[0037] Training configuration:

[0038] The batch size (batch_size) is set to 32, the number of iterations (epochs) is 100, and the early stopping mechanism (EarlyStopping) is used to terminate the training when the validation loss does not decrease for 10 consecutive times;

[0039] Regularization and evaluation:

[0040] In addition to Dropout, add L2 weight regularization (coefficient 0.01) to constrain the parameter range;

[0041] Evaluation indicators include accuracy, AUC-ROC curve and confusion matrix, focusing on the recall rate of the minority class (prestressed tendon broken samples);

[0042] S55. Model Validation

[0043] Calculate the accuracy and F1 score on an independent test set, and analyze the feature contribution through SHAP value;

[0044] The trained model is lightweight (converted to TensorFlow Lite) and then embedded in the testing equipment. For practical application, simply repeat steps S1 through S5 (without recording and inputting broken wire labels). Input data into the trained LSTM model, which then outputs the broken wire status at each measurement point, allowing rapid identification of the location and condition of prestressed tendon damage.

[0045] A detection system for a method for detecting damage to prestressed tendons in a prestressed composite steel cylinder concrete pipe using impact elastic waves comprises an exciter and a receiver. The exciter is mounted on an exciter handle via a first connecting rod, and the receiver is mounted on the receiver handle via a second connecting rod. The exciter handle and the receiver handle are connected via a detachable pin, and a trigger button is provided on the exciter handle.

[0046] This technical solution is a detection technique that does not require any pre-installed equipment. Inspection personnel must carry equipment into the pipeline during water outages and maintenance periods. Detection techniques primarily based on electromagnetic waves are not suitable for detecting prestressed composite steel cylinder concrete pipes made with non-metallic prestressed tendons, such as basalt fiber tendons. This technology, based on the impact elastic wave method (principle), integrates the changes in wave velocity in the concrete core and thickness caused by prestressed tendon failure to develop a highly accurate detection technique for prestressed tendon failure (defect type) in prestressed composite steel cylinder concrete pipes (structural form).

[0047] Beneficial effect: The impact elastic wave detection method for prestressed tendon breakage in prestressed composite steel cylinder concrete pipes of the present invention uses machine learning technology to comprehensively consider the change in P-wave velocity of the pipe core concrete caused by the breakage of the prestressed tendon and the influence of the peeling of the external mortar protective layer on the thickness of the component, and proposes a impact elastic wave detection method for prestressed tendon breakage in prestressed composite steel cylinder concrete pipes; this method can detect the prestressed composite, the material of the prestressed tendon is arbitrary, whether it is metal or non-metal, and the position of the prestressed tendon breakage in the steel cylinder concrete pipe. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a schematic diagram of the process of the present invention.

[0049] Figure 2 It is a top view of the device in the present invention.

[0050] Figure 3 It is a right side view of the device of the present invention. DETAILED DESCRIPTION

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

[0052] like Figure 2 and Figure 3 As shown, the structure of the test system in the present invention includes: a trigger button 1 for sending instructions to the host; an exciter handle 2, on which a trigger button is attached; a connecting rod 3 for connecting the handle and the exciter / receiver; an exciter 4 for generating an impact elastic wave signal; a connecting piece 5 for fixing the exciter / receiver on the connecting rod, allowing the exciter / receiver to rotate freely to fit the test surface; a receiver handle 6; a receiver 7 for receiving echo signals and surface wave signals; the connection 8 between the exciter handle and the receiver handle is a pin with a quick-release function, which allows the exciter handle and the receiver handle to be separated. The device should also include a matching connecting cable and a host.

[0053] like Figure 1 As shown, a method for detecting prestressed tendon damage in a prestressed composite steel cylinder concrete pipe by impact elastic waves of the present invention comprises the following steps:

[0054] S1: Delineate the measurement area and measurement line in the experimental prestressed steel cylinder concrete pipe (in this embodiment, an indoor test section with artificial broken wires is used as an illustration), conduct the surface wave velocity test using the impact elastic wave method, and obtain the surface wave velocity V of the concrete in the pipe core. R .

[0055] The apparent wave velocity test method is as follows: measure the length of a measuring line on the measuring line. l = 40cm measuring line segment, place a shock elastic wave exciter at one end and a signal receiver at the other end. Send a signal on the host to make the exciter emit a shock elastic wave signal, and the signal receiver receives the surface wave signal. Analyze the surface wave signal received by the signal receiver on the host to obtain the surface wave propagation time t R Calculate the surface wave velocity V using the formula R The calculation formula of surface wave velocity is: V R = l / t R . Calculate c using the formula: ,in is Poisson's ratio.

[0056] S2: Assemble the excitation device and probe of the shock elastic wave instrument into a shock elastic wave thickness measuring device, carry out shock elastic wave testing in the prestressed steel cylinder concrete pipe, obtain the time domain signal of the measuring point, and perform preprocessing to obtain the signal's superior frequency f and P wave arrival time t P .

[0057] The shock wave method for thickness measurement is as follows: Draw a measuring point every 5 cm along the survey line. Align the marking on the shock wave device with the measuring point. Press the device to ensure good contact between the vibrator and the signal receiver and the concrete surface. Press the switch to cause the host to emit an electrical signal. The vibrator, upon receiving the signal, strikes the concrete surface, and the signal receiver receives the echo in the time domain. The time domain signal at that measuring point is analyzed on the host.

[0058] The preprocessing method is to find the first peak of the time domain signal and mark its time as the P wave arrival time t P , perform fast Fourier transform on the time domain signal to obtain the frequency domain diagram of the signal, and find the dominant frequency f in the frequency domain diagram.

[0059] S3: The core concrete P wave velocity V obtained by S1 and S2 P , the outstanding frequency f of the measuring point, and the P wave arrival time t are converted into the component thickness of the measuring point through the formula. Component thickness: .

[0060] S4: Execute S1 to S3 for all measuring points in the measurement area to obtain the P-wave velocity, dominant frequency, P-wave arrival time, and component thickness of all measuring points to form a data set.

[0061] S5: Preprocess the data set, remove outliers (such as records with a wave velocity of 0 or outside a reasonable range), and directly use the original features (P-wave velocity, dominant frequency, P-wave arrival time, component thickness) as the characteristic parameters of the data set.

[0062] The training set, validation set, and test set are divided into 7:2:1 ratios to ensure that the test set represents the actual distribution.

[0063] In this example, since wire breakage is manually determined, the wire breakage status at each measurement point is known. Four features (P-wave velocity, dominant frequency, P-wave duration, and component thickness) and one label (whether the wire is broken or not) are reconstructed into a time series using a sliding window to adapt the input to the LSTM model. A two-layer LSTM network is designed. The first layer extracts local waveform features, and the second layer integrates global time series patterns to output the probability of prestressed tendon breakage. The test set is fed into the LSTM model to train the model.

[0064] The steps to train the LSTM model are as follows:

[0065] S51. Input feature preprocessing

[0066] The LSTM model input of the present invention is four key features of the prestressed steel cylinder concrete pipe measurement point (P wave velocity, dominant frequency, P wave time, component thickness) and one label (whether the wire is broken or not). The raw data was first cleaned to remove outliers caused by sensor failure or environmental interference (such as data with a wave velocity of 0 or outside the reasonable range of 3000-5000 m / s). Missing values ​​were filled using linear interpolation of adjacent measurement points. Subsequently, Z-score normalization was used to normalize each feature to eliminate the impact of dimensional differences on the model. The normalization formula is:

[0067]

[0068] in is the characteristic mean, is the standard deviation. The shape of the standardized data matrix is ,in is the number of measuring points.

[0069] S52. Time series construction

[0070] Since LSTM needs to process time series data, and the single point feature has no time dimension, the sliding window method is used to convert adjacent The data of the measurement points are spliced ​​into a time series sample to generate an input matrix with the shape of Label The raw data in does not need to be normalized. The component label sample input matrix (n_samples, Y) is 1 when the prestressed tendon at the measurement point is damaged, and 0 when the prestressed tendon at the measurement point is intact. Therefore, the final feature X input to the LSTM model is (n_samples-N+1,N,4); the input label Y is (n_samples,Y).

[0071] S53. Model Architecture Design

[0072] The LSTM model adopts a two-layer network structure, and the specific configuration is as follows:

[0073] First LSTM layer: 64 neurons, input shape is , set return_sequences=True to output the complete time series results for processing by the next layer. This layer captures local time series dependencies, such as the short-term fluctuation relationship between wave speed and stress, through gating mechanisms (forget gate, input gate, output gate).

[0074] Dropout layer: The dropout rate is 0.2, and some neurons are randomly blocked to prevent overfitting.

[0075] Second LSTM layer: 32 neurons, does not return sequences (return_sequences=False), extracts global temporal features and compresses them into a single vector.

[0076] Fully connected layer: 16 ReLU activated neurons to further integrate high-order features.

[0077] Output layer: 1 Sigmoid-activated neuron, outputting the probability value of prestressed tendon breakage (0-1).

[0078] S54. Model Training and Optimization

[0079] Loss function and optimizer:

[0080] The binary cross entropy loss (binary_crossentropy) is used for the binary classification task, and the mean square error (MSE) is used for the regression task.

[0081] The optimizer selected is Adam, the initial learning rate is set to 0.001, and the dynamic adjustment strategy is implemented through ReduceLROnPlateau (monitoring the validation loss, patience = 5).

[0082] Training configuration:

[0083] The batch size (batch_size) is set to 32, the number of iterations (epochs) is 100, and the early stopping mechanism (EarlyStopping) is used to terminate the training when the validation loss does not decrease for 10 consecutive times.

[0084] Regularization and evaluation:

[0085] In addition to Dropout, L2 weight regularization (coefficient 0.01) is added to constrain the parameter range.

[0086] The evaluation indicators include accuracy, AUC-ROC curve and confusion matrix, focusing on the recall rate of the minority class (prestressed tendon broken samples).

[0087] S55. Model Validation

[0088] The accuracy and F1 score are calculated on an independent test set, and the feature contribution is analyzed by SHAP value.

[0089] Step 6: Perform lightweight processing (TensorFlow Lite conversion) on the trained model and then embed it into the test equipment. When applying it in actual projects, repeat steps 1 to 5 for the project to be tested (no need to mark whether the wire is broken or not). ), input data into the trained LSTM model, and the LSTM model outputs the labels of each measurement point , in order to quickly determine the location and condition of prestressed tendon damage.

[0090] After the trained model is lightweighted (converted to TensorFlow Lite), it is embedded in the test equipment. When it is applied in actual projects, repeat S1 to S5 for the project to be tested (no broken wire label is required). ), input data into the trained LSTM model, and the LSTM model outputs the labels of each measurement point , in order to quickly determine the location and condition of prestressed tendon damage.

[0091] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A method for detecting prestressed steel cylinder concrete pipe prestressed tendon damage by impact elastic wave, characterized in that: The following steps are involved: S1: Delineate the measurement area and measurement line in the experimental prestressed steel cylinder concrete pipe, carry out the surface wave velocity test of the impact elastic wave method, and obtain the surface wave velocity of the concrete in the pipe core. ; S2: Assemble the excitation device and probe of the shock elastic wave instrument into a shock elastic wave thickness measuring device, carry out shock elastic wave testing in the prestressed steel cylinder concrete pipe, obtain the time domain signal of the measuring point, and perform preprocessing to obtain the excellent frequency of the signal , P wave arrival time 、P wave velocity V P ; S3: The P wave velocity of the core concrete obtained in steps S1 and S2 , excellent frequency of measuring points Convert the thickness of the component at the measuring point into the following formula: , is an empirical parameter; S4: Execute S1 to S3 for all measuring points in the measurement area to obtain the P-wave velocity, predominant frequency, P-wave arrival time, and component thickness of all measuring points to form a data set. Preprocess the data set to remove outliers and directly use the original features, namely P-wave velocity, predominant frequency, P-wave arrival time, and component thickness, as the feature parameters of the data set. The training set, validation set, and test set are divided into two parts in a ratio of 7:2:

1. S5: Reconstruct the four characteristic features (P-wave velocity, dominant frequency, P-wave time, component thickness, and one label (whether the wire is broken) into a time series using a sliding window. A two-layer LSTM network is designed. The first layer extracts local waveform features, and the second layer integrates global time series patterns to output the probability of prestressed tendon breakage. The test set is then fed into the LSTM model for training. The surface wave velocity in step S1 The test method is as follows: Measure the length of a measuring line on the measuring line The measuring line segment is arranged with a shock elastic wave exciter at one end point and a signal receiver at the other end point. A signal is sent on the host to make the exciter send a shock elastic wave signal, and the signal receiver receives the surface wave signal. The surface wave signal received by the signal receiver is analyzed on the host to obtain the surface wave propagation time. , calculate the surface wave velocity using the formula , then the calculation formula for the surface wave velocity is: , l The length of the measuring line segment can be calculated by the formula: : ,in is Poisson's ratio.

2. The method for detecting prestressed tendon damage in prestressed composite steel cylinder concrete pipes by impact elastic waves according to claim 1, characterized in that: The thickness measurement method using the impact elastic wave method is as follows: draw a measuring point every 5 cm on the measuring line, align the marking point of the impact elastic wave device with the measuring point, press the device to make the vibrator and the signal receiver in good contact with the concrete surface of the pipe core, press the switch to make the host send out an electrical signal, the vibrator hits the concrete surface of the pipe core after receiving the electrical signal, the signal receiver receives the echo time domain signal, analyze the time domain signal of the measuring point on the host, find the first peak of the time domain signal, and calibrate its time as the P wave arrival time. , perform fast Fourier transform on the time domain signal to obtain the frequency domain diagram of the signal, and obtain the outstanding frequency in the frequency domain diagram .

3. The method for detecting prestressed steel cylinder concrete pipe prestressed tendon damage by impact elastic wave according to claim 1, characterized in that: In step S5, the LSTM model input is the four key features X and one label of the prestressed steel cylinder concrete pipe measurement point: P wave velocity, dominant frequency, P wave time, component thickness and whether the wire is broken. The original data is first cleaned to remove outliers, and missing values ​​are filled by linear interpolation of adjacent measurement points. Z-score standardization is used to normalize each feature. The standardization formula is: ,in is the characteristic mean, is the standard deviation, and the shape of the standardized data matrix is ,in is the number of measurement points, and the sliding window method is used to divide adjacent The data of the measurement points are spliced ​​into a time series sample to generate an input matrix with the shape of , N takes any value from 10 to 100.

Images