Bridge deck pavement layering test method and system based on multi-point knocking and deep learning

Abstract

The invention discloses a bridge deck pavement layering test method and system based on multi-point knocking and deep learning, belongs to the technical field of bridge engineering, and aims to solve the technical problem of how to efficiently and non-destructively realize bridge deck pavement layering detection and ensure detection precision and construction quality. Comprising the following steps: knocking bridge floor sections through detection equipment provided with a plurality of electromagnetic knocking hammers to obtain sound wave signals in an analog signal form; performing signal amplification, filtering and analog-to-digital conversion processing on the acquired sound wave signal; time domain features and frequency domain features of the sound wave signals are calculated respectively, and a feature matrix is constructed based on the time domain features and the frequency domain features; analyzing the health state of the bridge deck segments through the trained bridge deck pavement layered disease detection model, and predicting and outputting an analysis result; and determining the damaged area of the bridge floor based on the analysis result of each bridge floor segment, constructing a repair suggestion, and constructing a pseudo-color thermodynamic diagram of the bridge floor based on the analysis result of each bridge floor segment.

Description

Technical Field

[0001] The present invention relates to the field of bridge engineering technology, and more specifically to a bridge deck pavement layering testing method and system based on multi-point tapping and deep learning. Background Art

[0002] The durability of bridge deck pavement directly impacts the service life of the bridge. Delamination within the bridge deck can lead to structural damage and reduced durability, posing safety risks. Currently, commonly used testing methods include core drilling, impact echo, and ultrasonic testing. However, these methods suffer from high destructiveness, low accuracy, and limited applicability.

[0003] How to efficiently and non-destructively detect the stratified condition of bridge deck pavement and ensure detection accuracy and construction quality is a technical problem that needs to be solved. Summary of the Invention

[0004] The technical task of the present invention is to address the above shortcomings and provide a bridge deck pavement stratification testing method and system based on multi-point tapping and deep learning to solve the technical problem of how to efficiently and non-destructively realize the stratification condition detection of bridge deck pavement and ensure the detection accuracy and construction quality.

[0005] In a first aspect, the present invention provides a bridge deck pavement layer testing method based on multi-point tapping and deep learning, comprising the following steps:

[0006] Acoustic wave acquisition: The bridge deck to be inspected is divided into multiple segments. For each segment, the bridge deck is struck at equal intervals or equal times using a testing device equipped with multiple electromagnetic hammers to form a multi-point excitation source and acquire an analog acoustic wave signal.

[0007] Signal processing: performing signal amplification, filtering, and analog-to-digital conversion on the collected sound wave signal to obtain a processed sound wave signal, which is a digital signal;

[0008] Feature extraction: Calculate the time domain features of the acoustic signal in the form of an analog signal, perform discrete wavelet transform on the acoustic signal in the form of a digital signal, extract the wavelength, amplitude, and spectral features of the acoustic signal, calculate the frequency domain features of the acoustic signal, and construct a feature matrix based on the time domain features and frequency domain features;

[0009] Bridge Deck Analysis: For each bridge deck segment, the health status of the bridge deck segment is analyzed using the corresponding feature matrix as input through the trained bridge deck pavement delamination disease detection model, and the analysis results are predicted and output. The analysis results include the bridge deck disease severity type and the disease severity type probability value, where the disease severity type includes no disease, slight disease, moderate disease, and severe disease. The bridge deck pavement delamination disease detection model is a deep learning network model built based on convolutional neural networks and long short-term memory networks;

[0010] Report analysis: Based on the analysis results of each bridge deck segment, the damaged area of the bridge deck is determined and repair recommendations are formulated. Based on the analysis results of each bridge deck segment, a pseudo-color heat map of the bridge deck is constructed to display the damage status of the bridge deck.

[0011] Preferably, the signal processing comprises the following steps:

[0012] The acoustic wave signal is amplified by an operational amplifier to obtain an amplified acoustic wave signal;

[0013] Filtering the amplified sound wave signal through a bandpass filter to extract the sound wave signal within a predetermined frequency range to obtain a filtered sound wave signal;

[0014] The filtered sound wave signal is converted into a digital sound wave signal through an analog-to-digital converter to obtain a digital sound wave signal.

[0015] Preferably, the time domain features of the acoustic signal include signal mean, root mean square, peak, kurtosis, skewness and instantaneous energy, and the frequency domain features include low frequency energy, high frequency energy, spectrum entropy, main frequency, energy ratio and wavelet packet entropy;

[0016] Signal mean μ x The calculation formula is:

[0017]

[0018] Among them, x i represents the signal amplitude of the i-th sampling point, and N represents the total number of sampling points of the acoustic signal, that is, the total number of points at which a knocking signal is sampled when collecting the acoustic signal;

[0019] The calculation formula for the root mean square RMS is:

[0020]

[0021] The formula for calculating peak PV is:

[0022] PV = x max -x min ,

[0023] Among them, x max represents the maximum amplitude of the knock signal within the target time window, x min Indicates the minimum amplitude of the tapping signal within the target time window;

[0024] The calculation formula of kurtosis K is:

[0025]

[0026] The calculation formula of skewness S is:

[0027]

[0028] The calculation formula of instantaneous energy E is:

[0029]

[0030] Low frequency energy E A The calculation formula is:

[0031]

[0032] Among them, A i Represents the coefficient value in the i-th low-frequency subband after wavelet decomposition, represents the energy contribution of the coefficient;

[0033] High frequency energy E D The calculation formula is:

[0034]

[0035] Among them, D i represents the detail coefficient in the i-th high frequency subband after discrete wavelet transform, Represents the energy contribution of each high-frequency coefficient;

[0036] The calculation formula of spectral entropy SE is:

[0037] SE=-∑P i logP i ,

[0038] in, X i Represents the spectral coefficients after discrete wavelet transform;

[0039] The calculation formula of energy ratio ER is:

[0040]

[0041] Main frequency E maxIt is the maximum energy coefficient obtained based on the wavelet coefficient, and the wavelet packet entropy WE represents the entropy calculated for the wavelet packet decomposition signal.

[0042] Preferably, the bridge deck pavement delamination disease detection model includes a CNN feature extraction layer, an LSTM feature extraction layer, a fully connected layer and an output layer;

[0043] The CNN feature extraction layer includes a convolution layer, an activation layer, and a pooling layer, which is used to extract features from the input feature matrix and output a feature sequence;

[0044] The LSTM feature extraction layer is used to capture the long-term dependency of the time series by taking the feature sequence output by the CNN feature extraction layer as input;

[0045] The fully connected layer is used to flatten the output of the LSTM feature extraction layer into a one-dimensional vector and integrate the extracted features through one or more fully connected layers;

[0046] The output layer is used to diagnose bridge deck defects using the features output by the fully connected layer as input, and output the defect severity type and the probability value of the defect severity type.

[0047] Preferably, the method further comprises the following steps:

[0048] Bridge deck trend analysis: Periodically perform tap inspections on the target bridge deck to obtain the feature matrix, location coordinates, inspection time, and analysis results corresponding to each inspection point on the target bridge deck as historical data. For each bridge deck segment, the probability values of the disease severity type of each relevant inspection point are weighted and summed to generate the health index corresponding to the bridge deck segment. The feature matrix and health index corresponding to the bridge deck segment are then used to construct a time series sample based on time. The time series sample is used as input to perform trend analysis on the bridge deck segment using the trained bridge deck trend prediction model. The predicted output is the disease severity type, disease severity type probability value, and health index corresponding to the bridge deck segment. For each bridge deck segment, a trend analysis curve is constructed based on the historical health index and the predicted health index. A multi-point trend heat map is constructed based on the change rate of the health index of all inspection points. The multi-point trend heat map displays the degree and trend of disease changes in different areas of the target bridge deck. The bridge deck trend prediction model is a deep learning model based on an LSTM network or a combination of an LSTM network and a CNN network.

[0049] Preferably, the detection device is a hand-push type delamination detection device, comprising a trolley body and a battery, a distance measuring wheel, an encoder, an electromagnetic coil, and an electromagnetic percussion hammer equipped with an acoustic wave sensor arranged on the trolley body;

[0050] There are multiple sets of measuring wheels, which are distributed under the trolley body and drive the trolley body to move;

[0051] The encoder is used to collect the rotation angle of the distance measuring wheel and convert the rotation angle of the distance measuring wheel into a pulse signal, and calculate the actual moving distance based on the circumference of the distance measuring wheel and the pulse signal;

[0052] The electromagnetic coil is used to drive the electromagnetic hammer to strike the bridge deck segments;

[0053] The battery is used to provide power.

[0054] In a second aspect, the present invention provides a bridge deck pavement delamination testing system based on multi-point tapping and deep learning, which is used to implement bridge deck pavement delamination testing using a bridge deck pavement delamination testing method based on multi-point tapping and deep learning as described in any one of the first aspects. The system includes an acoustic wave acquisition module, a signal processing module, a feature extraction module, a bridge deck analysis module, a report analysis module, and a bridge deck trend analysis module.

[0055] The acoustic wave acquisition module is configured to perform the following operations: divide the bridge deck to be inspected into multiple bridge deck segments, and for each bridge deck segment, strike the bridge deck segment using an inspection device equipped with multiple electromagnetic hammers in an equidistant or equidistant manner to form a multi-point excitation source, thereby acquiring an acoustic wave signal in the form of an analog signal;

[0056] The signal processing module is used to perform the following: signal amplification, filtering and analog-to-digital conversion processing on the collected sound wave signal to obtain a processed sound wave signal, wherein the processed sound wave signal is a digital signal;

[0057] The feature extraction module is used to perform the following: calculate the time domain features based on the acoustic signal in the form of an analog signal, perform discrete wavelet transform on the acoustic signal in the form of a digital signal, extract the wavelength, amplitude and spectral features of the acoustic signal, calculate the frequency domain features of the acoustic signal, and construct a feature matrix based on the time domain features and frequency domain features;

[0058] The bridge deck analysis module is configured to perform the following operations: for each bridge deck segment, using the corresponding feature matrix as input, analyze the health status of the bridge deck segment using a trained bridge deck pavement delamination disease detection model, and predict and output analysis results. The analysis results include the bridge deck disease type and disease type probability value, where the disease severity types include no disease, minor disease, moderate disease, and severe disease. The bridge deck pavement delamination disease detection model is a network model constructed based on a convolutional neural network and a long short-term memory network.

[0059] The report analysis module is used to perform the following: determine the damaged area of the bridge deck based on the analysis results of each bridge deck segment and formulate repair recommendations; construct a pseudo-color heat map of the bridge deck based on the analysis results of each bridge deck segment, and display the damage status of the bridge deck through the pseudo-color heat map;

[0060] The bridge deck trend analysis module is used to perform the following: periodically perform tap inspections on the target bridge deck, obtain the feature matrix, location coordinates, inspection time and analysis results corresponding to each inspection point on the target bridge deck as historical data, and for each bridge deck segment, perform weighted summation on the probability values of the disease degree type of each relevant inspection point to generate the health index corresponding to the bridge deck segment. The feature matrix and health index corresponding to the bridge deck segment are constructed into a time series sample according to time. The time series sample is used as input and the trained bridge deck trend prediction model is used to perform trend analysis on the bridge deck segment. The prediction output is the disease degree type, disease degree type probability value and health index corresponding to the bridge deck segment. For each bridge deck segment, a trend analysis curve is constructed based on the historical health index and the predicted health index, and a multi-point trend heat map is constructed based on the change rate of the health index of all inspection points. The multi-point trend heat map can be used to display the degree and trend of disease changes in different areas of the target bridge deck.

[0061] Preferably, the signal processing module is configured to perform the following operations:

[0062] The acoustic wave signal is amplified by an operational amplifier to obtain an amplified acoustic wave signal;

[0063] Filtering the amplified sound wave signal through a bandpass filter to extract the sound wave signal within a predetermined frequency range to obtain a filtered sound wave signal;

[0064] The filtered sound wave signal is converted into a digital sound wave signal through an analog-to-digital converter to obtain a digital sound wave signal.

[0065] Preferably, the time domain features of the acoustic signal include signal mean, root mean square, peak, kurtosis, skewness and instantaneous energy, and the frequency domain features include low frequency energy, high frequency energy, spectrum entropy, main frequency, energy ratio and wavelet packet entropy;

[0066] Signal mean μ x The calculation formula is:

[0067]

[0068] Among them, x i represents the signal amplitude of the i-th sampling point, and N represents the total number of sampling points of the acoustic signal, that is, the total number of points at which a knocking signal is sampled when collecting the acoustic signal;

[0069] The calculation formula for the root mean square RMS is:

[0070]

[0071] The formula for calculating peak PV is:

[0072] PV = x max -x min ,

[0073] Among them, x max represents the maximum amplitude of the knock signal within the target time window, x min Indicates the minimum amplitude of the tapping signal within the target time window;

[0074] The calculation formula of kurtosis K is:

[0075]

[0076] The calculation formula of skewness S is:

[0077]

[0078] The calculation formula of instantaneous energy E is:

[0079]

[0080] Low frequency energy E A The calculation formula is:

[0081]

[0082] Among them, A i Represents the coefficient value in the i-th low-frequency subband after wavelet decomposition, represents the energy contribution of the coefficient;

[0083] High frequency energy E D The calculation formula is:

[0084]

[0085] Among them, D i represents the detail coefficient in the i-th high frequency subband after discrete wavelet transform, Represents the energy contribution of each high-frequency coefficient;

[0086] The calculation formula of spectral entropy SE is:

[0087] SE=-∑P i logP i ,

[0088] in, X i Represents the spectral coefficients after discrete wavelet transform;

[0089] The calculation formula of energy ratio ER is:

[0090]

[0091] Main frequency E max It is the maximum energy coefficient obtained based on the wavelet coefficient, and the wavelet packet entropy WE represents the entropy calculated for the wavelet packet decomposition signal.

[0092] Preferably, the bridge deck pavement delamination disease detection model includes a CNN feature extraction layer, an LSTM feature extraction layer, a fully connected layer and an output layer;

[0093] The CNN feature extraction layer includes a convolution layer, an activation layer, and a pooling layer, which is used to extract features from the input feature matrix and output a feature sequence;

[0094] The LSTM feature extraction layer is used to capture the long-term dependency of the time series by taking the feature sequence output by the CNN feature extraction layer as input;

[0095] The fully connected layer is used to flatten the output of the LSTM feature extraction layer into a one-dimensional vector and integrate the extracted features through one or more fully connected layers;

[0096] The output layer is used to diagnose bridge deck defects using the features output by the fully connected layer as input, and output the defect severity type and the probability value of the defect severity type.

[0097] The bridge deck pavement delamination testing method and system based on multi-point tapping and deep learning of the present invention have the following advantages: acoustic wave signals are collected through multi-point tapping, a feature matrix including time domain features and frequency domain features is constructed based on the acoustic wave signals, and bridge deck defects are analyzed using a network model constructed based on a convolutional neural network and a long short-term memory network with the feature matrix as input, thereby achieving rapid and accurate detection of bridge deck pavement layer delamination. BRIEF DESCRIPTION OF THE DRAWINGS

[0098] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments or descriptions of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0099] The present invention will be further described below with reference to the accompanying drawings.

[0100] Figure 1 This is a flowchart of a bridge deck pavement layer testing method based on multi-point tapping and deep learning in Example 1;

[0101] Figure 2 This is a schematic structural diagram of a detection device in a bridge deck pavement layer testing method based on multi-point tapping and deep learning in Example 1;

[0102] Among them, 1. signal processing module, 2. handle, 3. distance measuring wheel, 4. electromagnetic hammer, 5. acoustic wave sensor, 6. encoder, 7. electromagnetic coil, 8. battery. DETAILED DESCRIPTION

[0103] The present invention will be further described below with reference to the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it. However, the embodiments given are not intended to limit the present invention. Unless there is a conflict, the embodiments of the present invention and the technical features in the embodiments may be combined with each other.

[0104] The embodiments of the present invention provide a bridge deck pavement delamination testing method and system based on multi-point tapping and deep learning, which are used to solve the technical problem of how to efficiently and non-destructively realize the delamination condition detection of bridge deck pavement and ensure the detection accuracy and construction quality.

[0105] Example 1:

[0106] The present invention provides a bridge deck pavement layer testing method based on multi-point tapping and deep learning, such as Figure 1 The diagram shows five steps including acoustic wave acquisition, signal processing, feature extraction, bridge deck analysis and report analysis.

[0107] Step S100: Acquisition of acoustic waves: Divide the bridge deck to be inspected into multiple bridge deck segments. For each bridge deck segment, the bridge deck segment is struck by an inspection device equipped with multiple electromagnetic hammers in an equidistant or equitemporal manner to form a multi-point excitation source and obtain an acoustic wave signal in the form of an analog signal.

[0108] like Figure 2 As shown, the detection device of this embodiment is a hand-pushed delamination detection device, comprising a trolley body, a battery 8 mounted on the trolley body, distance measuring wheels 3, an encoder 6, an electromagnetic coil 7, and an electromagnetic hammer 4 equipped with an acoustic wave sensor. Multiple sets of distance measuring wheels 3 are distributed beneath the trolley body and drive the trolley body. The encoder 6 is used to collect the rotation angle of the distance measuring wheels and convert the rotation angle into a pulse signal, calculating the actual travel distance based on the circumference of the distance measuring wheels and the pulse signal. The electromagnetic coil 7 is used to drive the electromagnetic hammer 4 to strike the bridge deck segments. The battery 8 is used to provide power. In this embodiment, a handle 2 is provided on the trolley body for ease of operation, and a signal processing module 1 for signal processing is also mounted on the trolley body.

[0109] In this embodiment, the hand-push delamination detection device includes five electromagnetic hammers with acoustic wave sensors inside, which perform independent or synchronous striking at set times or distances to form a multi-point excitation source and generate a stable acoustic wave signal.

[0110] Equidistant detection uses an encoder to convert the rotation angle of a measuring wheel into a pulse signal. Given a known wheel circumference, the actual travel distance is calculated as the number of pulses multiplied by the wheel circumference / number of pulses per revolution. Isochronous detection uses an electromagnetic coil to drive a hammer, which strikes the road surface at set intervals.

[0111] In specific implementation, the acoustic wave sensor array is arranged inside the electromagnetic percussion hammer to receive the acoustic wave signal generated by the percussion. Synchronous acquisition technology is used to ensure that each sensor records the acoustic wave signal at the same time point to avoid time errors. Synchronous acquisition technology uses external clock synchronization, using a master clock (such as FPGA / microcontroller clock) to provide the same synchronous clock signal to all sensors. The sensor only collects data when it receives a clock pulse, ensuring that the trigger time of all sensors is the same. Synchronous acquisition technology uses external clock synchronization, using a master clock (such as FPGA / microcontroller clock) to provide the same synchronous clock signal to all sensors.

[0112] When collecting acoustic wave signals by striking the bridge deck with five electromagnetic hammers, multi-channel parallel collection is adopted, which can be done individually or with five at the same time.

[0113] Step S200: Signal processing: performing signal amplification, filtering, and analog-to-digital conversion on the collected sound wave signal to obtain a processed sound wave signal, which is a digital signal.

[0114] As a specific implementation of signal processing, the following steps are included:

[0115] (1) amplifying the acoustic wave signal through an operational amplifier to obtain an amplified acoustic wave signal;

[0116] (2) filtering the amplified sound wave signal through a bandpass filter to extract the sound wave signal within a predetermined frequency range to obtain a filtered sound wave signal;

[0117] (3) For the filtered sound wave signal, the sound wave signal in analog form is converted into a sound wave signal in digital form by an analog-to-digital converter to obtain a sound wave signal in digital form.

[0118] In this embodiment, the operational amplifier boosts weak signals to a suitable voltage range. This is achieved through negative feedback and high-gain amplification, allowing for subsequent acquisition and processing by the analog-to-digital converter (ADC). The bandpass filter extracts signals within a specific frequency range, suppressing low-frequency interference (ground vibrations from adjacent electromagnetic wheels) and high-frequency noise (sound waves from adjacent electromagnetic hammers).

[0119] The basic equation for an operational amplifier is:

[0120] V out =A open-loop ×(V + -V - ),

[0121] Among them, V out Indicates the output voltage, A open-loop represents the open-loop gain (typically up to 10 5 to 10 6 ), V + Indicates the voltage at the non-inverting input terminal, V - Indicates the voltage at the non-inverting output terminal.

[0122] Step S300 Feature Extraction: Calculate the time domain features based on the sound wave signal in the form of an analog signal, perform discrete wavelet transform on the sound wave signal in the form of a digital signal, extract the wavelength, amplitude and spectral features of the sound wave signal, calculate the frequency domain features of the sound wave signal, and construct a feature matrix based on the time domain features and frequency domain features.

[0123] In this embodiment, the sound wave signal is subjected to discrete wavelet transform to extract the low-frequency part and the high-frequency part, wherein the low-frequency part (approximate coefficient A n ) represents the large-scale characteristics of the sound wave signal, such as the main wavelength information, the high-frequency part (detail coefficient D n ) represents the detailed characteristics of the signal, such as local amplitude and spectrum changes.

[0124] The signal characteristics of bridge deck pavement stratification are complex and require multi-dimensional information support. The core goal of the high-dimensional feature matrix is to improve the accuracy and robustness of stratification detection. In this embodiment, the time domain characteristics of the acoustic signal include signal mean, root mean square, peak, kurtosis, skewness and instantaneous energy. The frequency domain characteristics include low-frequency energy, high-frequency energy, spectral entropy, main frequency, energy ratio and wavelet packet entropy.

[0125] The signal mean reflects the overall trend of the signal, and the signal mean μ x The calculation formula is:

[0126]

[0127] Among them, x i It represents the signal amplitude of the i-th sampling point (usually the displacement, velocity or acceleration of the vibration signal), and N represents the total number of sampling points of the sound wave signal, that is, the total number of points at which a knock signal is sampled when collecting the sound wave signal.

[0128] The root mean square measures the energy of the signal. The calculation formula for the root mean square RMS is:

[0129]

[0130] The peak value represents the maximum amplitude range of the signal. The calculation formula of the peak value PV is:

[0131] PV = x max -x min ,

[0132] Among them, x max represents the maximum amplitude of the knock signal within the target time window, x min Indicates the minimum amplitude of the tap signal within the target time window.

[0133] Kurtosis reflects the sharpness of the signal. The calculation formula of Kurtosis K is:

[0134]

[0135] Skewness reflects the symmetry of the signal. The calculation formula of skewness S is:

[0136]

[0137] Instantaneous energy reflects the total energy of the signal. The calculation formula of instantaneous energy E is:

[0138]

[0139] Low-frequency energy is the energy of the low-frequency component in the sound wave signal, reflecting the stability of the signal. Low-frequency energy E A The calculation formula is:

[0140]

[0141] Among them, A i Represents the coefficient value in the i-th low-frequency subband after wavelet decomposition, represents the energy contribution of the coefficient.

[0142] High-frequency energy is the high-frequency component energy in the sound wave signal, reflecting the delamination defect. High-frequency energy E D The calculation formula is:

[0143]

[0144] Among them, D i represents the detail coefficient in the i-th high frequency subband after discrete wavelet transform, Represents the energy contribution of each high-frequency coefficient.

[0145] The spectrum entropy reflects the complexity of the signal. The calculation formula of spectrum entropy SE is:

[0146] SE=-∑P i logP i ,

[0147] in, X i Represents the spectral coefficient after discrete wavelet transform. i It represents the normalized energy of the ith frequency component, that is, the ratio of the energy of this frequency component to the total energy of the entire signal, and is used to calculate the spectral entropy (SE).

[0148] The energy ratio reflects the relative proportion of high frequency and low frequency. The calculation formula of energy ratio ER is:

[0149]

[0150] The main frequency is the main vibration frequency, expressed as E max , is the maximum energy coefficient obtained based on the wavelet coefficients.

[0151] Wavelet packet entropy measures the non-stationarity of the signal and is expressed as WE. The entropy is calculated for the wavelet packet decomposition signal.

[0152] When constructing a feature matrix based on the above time domain features and frequency domain features, each row corresponds to a detection point, and each row has 12 elements, and the 12 elements correspond to the time domain features and frequency domain corresponding to the detection point.

[0153] Step S400: Bridge deck analysis: For each bridge deck segment, the health status of the bridge deck segment is analyzed using the trained bridge deck pavement layer disease detection model with the corresponding feature matrix as input, and the analysis results are predicted and output. The analysis results include the bridge deck disease severity type and the disease severity type probability value, where the disease severity type includes no disease, minor disease, moderate disease and severe disease. The bridge deck pavement layer disease detection model is a network model constructed based on convolutional neural network and long short-term memory network.

[0154] In this embodiment, the bridge deck pavement delamination disease detection model includes a CNN feature extraction layer, an LSTM feature extraction layer, a fully connected layer and an output layer.

[0155] The CNN feature extraction layer includes convolutional layer, activation layer and pooling layer, which is used to extract features from the input feature matrix and output feature sequence.

[0156] The LSTM feature extraction layer is used to capture the long-term dependencies of the time series using the feature sequence output by the CNN feature extraction layer as input.

[0157] The fully connected layer is used to flatten the output of the LSTM feature extraction layer into a one-dimensional vector and integrate the extracted features through one or more fully connected layers.

[0158] The output layer is used to diagnose bridge deck defects using the features output by the fully connected layer as input, and output the defect severity type and the probability value of the defect severity type.

[0159] Step S500 report analysis: based on the analysis results of each bridge deck segment, determine the damaged area of the bridge deck and construct a repair suggestion; based on the analysis results of each bridge deck segment, construct a pseudo-color heat map of the bridge deck, and display the damage status of the bridge deck through the pseudo-color heat map.

[0160] In this embodiment, the analysis results corresponding to each detection point are mapped to the bridge deck topology structure through Matplotlib software according to the analysis results to generate a pseudo-color heat map. In the pseudo-color heat map, the redder the color, the more serious the disease.

[0161] As an improvement to this embodiment, the method further includes the following steps:

[0162] Step S600 bridge deck trend analysis: periodically perform tapping inspections on the target bridge deck, obtain the feature matrix, position coordinates, inspection time and analysis results corresponding to each inspection point on the target bridge deck as historical data, and for each bridge deck segment, perform weighted summation on the probability values of the disease degree type of each relevant inspection point to generate a health index corresponding to the bridge deck segment, and construct a time series sample of the feature matrix and health index corresponding to the bridge deck segment according to time. Use the time series sample as input and the trained bridge deck trend prediction model to perform trend analysis on the bridge deck segment, and predict the output as the disease degree type, disease degree type probability value and health index corresponding to the bridge deck segment; for each bridge deck segment, construct a trend analysis curve based on the historical health index and the predicted health index, and construct a multi-point trend heat map based on the change rate of the health index of all inspection points, and display the degree and trend of disease changes in different areas of the target bridge deck through the multi-point trend heat map, wherein the bridge deck trend prediction model is a deep learning model based on the LSTM network or a combination of the LSTM network and the CNN network.

[0163] The calculation formula of the Bridge Health Index (BHI) in this embodiment is as follows:

[0164]

[0165] Among them, N is the total number of detection points, ω i The importance weight of each point (can be equal weight), is the probability of the disease severity type at the i-th detection point at time t.

[0166] When training the bridge deck trend prediction model, the supervised learning + historical sample set + cross-validation method commonly used in existing technologies is used for model training and verification.

[0167] As a concrete implementation of bridge deck trend analysis, a regular inspection mechanism (e.g., once every six months / year) is established, storing historical inspection data. During inspections, acoustic wave acquisition, signal processing, and bridge deck analysis are periodically performed. The characteristic matrix, location coordinates, inspection time, and analysis results corresponding to each inspection point on the target bridge deck are used as historical data. A health index is introduced, and a weighted calculation is performed on the probability value corresponding to the disease severity type at each inspection point in the bridge deck segment during each inspection to obtain the health index of the bridge deck segment. The characteristic matrix and health index corresponding to the bridge deck segment are then used to construct time series samples. Using these time series samples as input, the trained bridge deck trend prediction model predicts the disease severity type, disease severity type probability, and health index change trend of the bridge deck segment at several future moments, achieving early warning. The results are visualized. For each inspection point and each bridge deck segment, a trend analysis curve is constructed based on the historical and predicted health indices. A multi-point trend heat map is constructed based on the rate of change of the health indices at all inspection points. This multi-point trend heat map displays the extent and trend of disease changes in different areas of the target bridge deck. The color gradient in the multi-point trend heat map indicates the rate of increase or decrease in the health index. If the BHI is predicted to continue to decline in the future and cross a preset threshold, an early warning report is automatically generated.

[0168] Example 2:

[0169] The present invention provides a bridge deck pavement layering testing system based on multi-point tapping and deep learning, which includes an acoustic wave acquisition module, a signal processing module, a feature extraction module, a bridge deck analysis module, a report analysis module and a bridge deck trend analysis module.

[0170] The acoustic wave acquisition module is used to perform the following: divide the bridge deck to be inspected into multiple bridge deck segments. For each bridge deck segment, the bridge deck segment is struck by a detection device equipped with multiple electromagnetic hammers in an equidistant or equitemporal manner to form a multi-point excitation source and obtain an acoustic wave signal in the form of an analog signal.

[0171] The detection device in this embodiment is a hand-pushed delamination detection device, comprising a trolley body, a battery 8 mounted on the trolley body, distance measuring wheels 3, an encoder 6, an electromagnetic coil 7, and an electromagnetic hammer 4 equipped with an acoustic wave sensor. Multiple sets of distance measuring wheels 3 are distributed beneath the trolley body and drive the trolley body. The encoder 6 is used to collect the rotation angle of the distance measuring wheels and convert the rotation angle into a pulse signal, calculating the actual travel distance based on the circumference of the distance measuring wheels and the pulse signal. The electromagnetic coil 7 is used to drive the electromagnetic hammer 4 to strike the bridge deck segments. The battery 8 is used to provide power. In this embodiment, a handle 2 is provided on the trolley body for ease of operation, and a signal processing module 1 for signal processing is also mounted on the trolley body.

[0172] In this embodiment, the hand-push delamination detection device includes five electromagnetic hammers with acoustic wave sensors inside, which perform independent or synchronous striking at set times or distances to form a multi-point excitation source and generate a stable acoustic wave signal.

[0173] Equidistant detection uses an encoder to convert the rotation angle of a measuring wheel into a pulse signal. Given a known wheel circumference, the actual travel distance is calculated as the number of pulses multiplied by the wheel circumference / number of pulses per revolution. Isochronous detection uses an electromagnetic coil to drive a hammer, which strikes the road surface at set intervals.

[0174] In specific implementation, the acoustic wave sensor array is arranged inside the electromagnetic percussion hammer to receive the acoustic wave signal generated by the percussion. Synchronous acquisition technology is used to ensure that each sensor records the acoustic wave signal at the same time point to avoid time errors. Synchronous acquisition technology uses external clock synchronization, using a master clock (such as FPGA / microcontroller clock) to provide the same synchronous clock signal to all sensors. The sensor only collects data when it receives a clock pulse, ensuring that the trigger time of all sensors is the same. Synchronous acquisition technology uses external clock synchronization, using a master clock (such as FPGA / microcontroller clock) to provide the same synchronous clock signal to all sensors.

[0175] When collecting acoustic wave signals by striking the bridge deck with five electromagnetic hammers, multi-channel parallel collection is adopted, which can be done individually or with five at the same time.

[0176] The signal processing module is used to perform the following: signal amplification, filtering and analog-to-digital conversion processing on the collected sound wave signal to obtain a processed sound wave signal, and the processed sound wave signal is a digital signal.

[0177] As a specific implementation of the signal processing module, this module is used to perform the following operations:

[0178] (1) amplifying the acoustic wave signal through an operational amplifier to obtain an amplified acoustic wave signal;

[0179] (2) filtering the amplified sound wave signal through a bandpass filter to extract the sound wave signal within a predetermined frequency range to obtain a filtered sound wave signal;

[0180] (3) For the filtered sound wave signal, the sound wave signal in analog form is converted into a sound wave signal in digital form by an analog-to-digital converter to obtain a sound wave signal in digital form.

[0181] In this embodiment, the operational amplifier boosts weak signals to a suitable voltage range. This is achieved through negative feedback and high-gain amplification, allowing for subsequent acquisition and processing by the analog-to-digital converter (ADC). The bandpass filter extracts signals within a specific frequency range, suppressing low-frequency interference (ground vibrations from adjacent electromagnetic wheels) and high-frequency noise (sound waves from adjacent electromagnetic hammers).

[0182] The basic equation for an operational amplifier is:

[0183] V out =A open-loop ×(V + -V - ),

[0184] Among them, V out Indicates the output voltage, A open-loop represents the open-loop gain (typically up to 10 5 to 10 6 ), V + Indicates the voltage at the non-inverting input terminal, V - Indicates the voltage at the non-inverting output terminal.

[0185] The feature extraction module is used to perform the following: calculate the time domain features based on the sound wave signal in the form of an analog signal, perform discrete wavelet transform on the sound wave signal in the form of a digital signal, extract the wavelength, amplitude and spectral features of the sound wave signal, calculate the frequency domain features of the sound wave signal, and construct a feature matrix based on the time domain features and frequency domain features.

[0186] In this embodiment, the sound wave signal is subjected to discrete wavelet transform to extract the low-frequency part and the high-frequency part, wherein the low-frequency part (approximate coefficient A n ) represents the large-scale characteristics of the sound wave signal, such as the main wavelength information, the high-frequency part (detail coefficient D n ) represents the detailed characteristics of the signal, such as local amplitude and spectrum changes.

[0187] The signal characteristics of bridge deck pavement stratification are complex and require multi-dimensional information support. The core goal of the high-dimensional feature matrix is to improve the accuracy and robustness of stratification detection. In this embodiment, the time domain characteristics of the acoustic signal include signal mean, root mean square, peak, kurtosis, skewness and instantaneous energy. The frequency domain characteristics include low-frequency energy, high-frequency energy, spectral entropy, main frequency, energy ratio and wavelet packet entropy.

[0188] The signal mean reflects the overall trend of the signal, and the signal mean μ x The calculation formula is:

[0189]

[0190] Among them, x i It represents the signal amplitude of the i-th sampling point (usually the displacement, velocity or acceleration of the vibration signal), and N represents the total number of sampling points of the sound wave signal, that is, the total number of points at which a knock signal is sampled when collecting the sound wave signal.

[0191] The root mean square measures the energy of the signal. The calculation formula for the root mean square RMS is:

[0192]

[0193] The peak value represents the maximum amplitude range of the signal. The calculation formula of the peak value PV is:

[0194] PV = x max -x min ,

[0195] Among them, x max represents the maximum amplitude of the knock signal within the target time window, x min Indicates the minimum amplitude of the tap signal within the target time window.

[0196] Kurtosis reflects the sharpness of the signal. The calculation formula of Kurtosis K is:

[0197]

[0198] Skewness reflects the symmetry of the signal. The calculation formula of skewness S is:

[0199]

[0200] Instantaneous energy reflects the total energy of the signal. The calculation formula of instantaneous energy E is:

[0201]

[0202] Low-frequency energy is the energy of the low-frequency component in the sound wave signal, reflecting the stability of the signal. Low-frequency energy E A The calculation formula is:

[0203]

[0204] Among them, A i Represents the coefficient value in the i-th low-frequency subband after wavelet decomposition, represents the energy contribution of the coefficient.

[0205] High-frequency energy is the high-frequency component energy in the sound wave signal, reflecting the delamination defect. High-frequency energy E D The calculation formula is:

[0206]

[0207] Among them, D i represents the detail coefficient in the i-th high frequency subband after discrete wavelet transform, Represents the energy contribution of each high-frequency coefficient.

[0208] The spectrum entropy reflects the complexity of the signal. The calculation formula of spectrum entropy SE is:

[0209] SE=-∑P i logP i ,

[0210] in, X i Represents the spectral coefficient after discrete wavelet transform. i It represents the normalized energy of the ith frequency component, that is, the ratio of the energy of this frequency component to the total energy of the entire signal, and is used to calculate the spectral entropy (SE).

[0211] The energy ratio reflects the relative proportion of high frequency and low frequency. The calculation formula of energy ratio ER is:

[0212]

[0213] The main frequency is the main vibration frequency, expressed as E max , is the maximum energy coefficient obtained based on the wavelet coefficients.

[0214] Wavelet packet entropy measures the non-stationarity of the signal and is expressed as WE. The entropy is calculated for the wavelet packet decomposition signal.

[0215] When constructing a feature matrix based on the above time domain features and frequency domain features, each row corresponds to a detection point, and each row has 12 elements, and the 12 elements correspond to the time domain features and frequency domain corresponding to the detection point.

[0216] The bridge deck analysis module is used to perform the following: for each bridge deck segment, the corresponding feature matrix is used as input, and the health status of the bridge deck segment is analyzed through the trained bridge deck pavement delamination disease detection model, and the analysis results are predicted and output. The analysis results include the bridge deck disease severity type and the disease severity type probability value, where the disease severity type includes no disease, minor disease, moderate disease and severe disease. The bridge deck pavement delamination disease detection model is a network model constructed based on convolutional neural network and long short-term memory network.

[0217] In this embodiment, the bridge deck pavement delamination disease detection model includes a CNN feature extraction layer, an LSTM feature extraction layer, a fully connected layer and an output layer.

[0218] The CNN feature extraction layer includes convolutional layer, activation layer and pooling layer, which is used to extract features from the input feature matrix and output feature sequence.

[0219] The LSTM feature extraction layer is used to capture the long-term dependencies of the time series using the feature sequence output by the CNN feature extraction layer as input.

[0220] The fully connected layer is used to flatten the output of the LSTM feature extraction layer into a one-dimensional vector and integrate the extracted features through one or more fully connected layers.

[0221] The output layer is used to diagnose bridge deck defects using the features output by the fully connected layer as input, and output the defect severity type and the probability value of the defect severity type.

[0222] The report analysis module is used to perform the following: determine the damaged area of the bridge deck based on the analysis results of each bridge deck segment and formulate repair recommendations; construct a pseudo-color heat map of the bridge deck based on the analysis results of each bridge deck segment, and display the damage status of the bridge deck through the pseudo-color heat map.

[0223] In this embodiment, the report analysis module is used to map the analysis results corresponding to each detection point to the bridge deck topology structure through Matplotlib software according to the analysis results, and generate a pseudo-color heat map. In the pseudo-color heat map, the redder the color, the more serious the disease.

[0224] The bridge deck trend analysis module is used to perform the following: periodically perform tapping inspections on the target bridge deck, obtain the feature matrix, location coordinates, inspection time, and analysis results corresponding to each inspection point on the target bridge deck as historical data, perform weighted summation on the probability values of the disease degree type of each relevant inspection point for each bridge deck segment, generate the health index corresponding to the bridge deck segment, and construct a time series sample based on the feature matrix and health index corresponding to the bridge deck segment according to time. The time series sample is used as input and the trained bridge deck trend prediction model is used to perform trend analysis on the bridge deck segment. The prediction output is the disease degree type, disease degree type probability value, and health index corresponding to the bridge deck segment. For each bridge deck segment, a trend analysis curve is constructed based on the historical health index and the predicted health index, and a multi-point trend heat map is constructed based on the change rate of the health index of all inspection points. The multi-point trend heat map displays the degree and trend of disease changes in different areas of the target bridge deck. The bridge deck trend prediction model is a deep learning model based on an LSTM network or a combination of an LSTM network and a CNN network.

[0225] The calculation formula of the Bridge Health Index (BHI) in this embodiment is as follows:

[0226]

[0227] Among them, N is the total number of detection points, ω i The importance weight of each point (can be equal weight), is the probability of the disease severity type at the i-th detection point at time t.

[0228] When training the bridge deck trend prediction model, the supervised learning + historical sample set + cross-validation method commonly used in existing technologies is used for model training and verification.

[0229] As a concrete implementation of bridge deck trend analysis, a regular inspection mechanism (e.g., once every six months / year) is established, storing historical inspection data. During inspections, acoustic wave acquisition, signal processing, and bridge deck analysis are periodically performed. The characteristic matrix, location coordinates, inspection time, and analysis results corresponding to each inspection point on the target bridge deck are used as historical data. A health index is introduced, and a weighted calculation is performed on the probability value corresponding to the disease severity type at each inspection point in the bridge deck segment during each inspection to obtain the health index of the bridge deck segment. The characteristic matrix and health index corresponding to the bridge deck segment are then used to construct time series samples. Using these time series samples as input, the trained bridge deck trend prediction model predicts the disease severity type, disease severity type probability, and health index change trend of the bridge deck segment at several future moments, achieving early warning. The results are visualized. For each inspection point and each bridge deck segment, a trend analysis curve is constructed based on the historical and predicted health indices. A multi-point trend heat map is constructed based on the rate of change of the health indices at all inspection points. This multi-point trend heat map displays the extent and trend of disease changes in different areas of the target bridge deck. The color gradient in the multi-point trend heat map indicates the rate of increase or decrease in the health index. If the BHI is predicted to continue to decline in the future and cross a preset threshold, an early warning report is automatically generated.

[0230] The system of this embodiment can execute the method disclosed in Example 1 to quickly and accurately detect the delamination of the bridge deck pavement layer.

[0231] The above is a detailed introduction to the bridge deck pavement layer testing method and system based on multi-point tapping and deep learning provided by the present invention. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for general technical personnel in this field, according to the ideas of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A bridge deck pavement layer testing method based on multi-point tapping and deep learning, characterized in that: The steps include: Acoustic wave acquisition: The bridge deck to be inspected is divided into multiple segments. For each segment, the bridge deck is struck at equal intervals or equal times using a testing device equipped with multiple electromagnetic hammers to form a multi-point excitation source and acquire an analog acoustic wave signal. Signal processing: performing signal amplification, filtering, and analog-to-digital conversion on the collected sound wave signal to obtain a processed sound wave signal, which is a digital signal; Feature extraction: Calculate the time domain features of the acoustic signal in the form of an analog signal, perform discrete wavelet transform on the acoustic signal in the form of a digital signal, extract the wavelength, amplitude, and spectral features of the acoustic signal, calculate the frequency domain features of the acoustic signal, and construct a feature matrix based on the time domain features and frequency domain features; Bridge Deck Analysis: For each bridge deck segment, the health status of the bridge deck segment is analyzed using the corresponding feature matrix as input through the trained bridge deck pavement delamination disease detection model, and the analysis results are predicted and output. The analysis results include the bridge deck disease severity type and the disease severity type probability value, where the disease severity type includes no disease, slight disease, moderate disease, and severe disease. The bridge deck pavement delamination disease detection model is a deep learning network model built based on convolutional neural networks and long short-term memory networks; Report analysis: Based on the analysis results of each bridge deck segment, the damaged area of the bridge deck is determined and repair recommendations are formulated. Based on the analysis results of each bridge deck segment, a pseudo-color heat map of the bridge deck is constructed to display the damage status of the bridge deck.

2. The bridge deck pavement layer testing method based on multi-point tapping and deep learning according to claim 1 is characterized in that: Signal processing includes the following steps: The acoustic wave signal is amplified by an operational amplifier to obtain an amplified acoustic wave signal; Filtering the amplified sound wave signal through a bandpass filter to extract the sound wave signal within a predetermined frequency range to obtain a filtered sound wave signal; The filtered sound wave signal is converted into a digital sound wave signal through an analog-to-digital converter to obtain a digital sound wave signal.

3. The bridge deck pavement layer testing method based on multi-point tapping and deep learning according to claim 1 is characterized in that: The time domain characteristics of acoustic signals include signal mean, root mean square, peak, kurtosis, skewness and instantaneous energy; the frequency domain characteristics include low-frequency energy, high-frequency energy, spectrum entropy, main frequency, energy ratio and wavelet packet entropy; Signal mean μ x The calculation formula is: Among them, x i represents the signal amplitude of the i-th sampling point, and N represents the total number of sampling points of the acoustic signal, that is, the total number of points at which a knocking signal is sampled when collecting the acoustic signal; The calculation formula for the root mean square RMS is: The formula for calculating peak PV is: PV=x max -x min , Among them, x max represents the maximum amplitude of the knock signal within the target time window, x min Indicates the minimum amplitude of the tapping signal within the target time window; The calculation formula of kurtosis K is: The calculation formula of skewness S is: The calculation formula of instantaneous energy E is: Low frequency energy E A The calculation formula is: Among them, A i Represents the coefficient value in the i-th low-frequency subband after wavelet decomposition, represents the energy contribution of the coefficient; High frequency energy E D The calculation formula is: Among them, D i represents the detail coefficient in the i-th high frequency subband after discrete wavelet transform, Represents the energy contribution of each high-frequency coefficient; The calculation formula of spectral entropy SE is: SE=-∑P i logP i , in, X i Represents the spectral coefficients after discrete wavelet transform; The calculation formula of energy ratio ER is: Main frequency E max is the maximum energy coefficient obtained based on the wavelet coefficient, and the wavelet packet entropy WE represents the entropy calculated for the wavelet packet decomposition signal.

4. The bridge deck pavement layer testing method based on multi-point tapping and deep learning according to claim 1 is characterized in that: The bridge deck pavement delamination disease detection model includes a CNN feature extraction layer, an LSTM feature extraction layer, a fully connected layer and an output layer; The CNN feature extraction layer includes a convolution layer, an activation layer, and a pooling layer, which is used to extract features from the input feature matrix and output a feature sequence; The LSTM feature extraction layer is used to capture the long-term dependency of the time series by taking the feature sequence output by the CNN feature extraction layer as input; The fully connected layer is used to flatten the output of the LSTM feature extraction layer into a one-dimensional vector and integrate the extracted features through one or more fully connected layers; The output layer is used to diagnose bridge deck defects using the features output by the fully connected layer as input, and output the defect severity type and the probability value of the defect severity type.

5. The bridge deck pavement layer testing method based on multi-point tapping and deep learning according to any one of claims 1 to 4, characterized in that: The method further comprises the steps of: Bridge deck trend analysis: Periodically perform tap inspections on the target bridge deck to obtain the feature matrix, location coordinates, inspection time, and analysis results corresponding to each inspection point on the target bridge deck as historical data. For each bridge deck segment, the probability values of the disease severity type of each relevant inspection point are weighted and summed to generate the health index corresponding to the bridge deck segment. The feature matrix and health index corresponding to the bridge deck segment are then used to construct a time series sample based on time. The time series sample is used as input to perform trend analysis on the bridge deck segment using the trained bridge deck trend prediction model. The predicted output is the disease severity type, disease severity type probability value, and health index corresponding to the bridge deck segment. For each bridge deck segment, a trend analysis curve is constructed based on the historical health index and the predicted health index. A multi-point trend heat map is constructed based on the change rate of the health index of all inspection points. The multi-point trend heat map displays the degree and trend of disease changes in different areas of the target bridge deck. The bridge deck trend prediction model is a deep learning model based on an LSTM network or a combination of an LSTM network and a CNN network.

6. The bridge deck pavement layer testing method based on multi-point tapping and deep learning according to claim 1 is characterized in that: The detection device is a hand-pushed delamination detection device, comprising a trolley body and a battery, a distance measuring wheel, an encoder, an electromagnetic coil, and an electromagnetic percussion hammer equipped with an acoustic wave sensor arranged on the trolley body; There are multiple sets of measuring wheels, which are distributed under the trolley body and drive the trolley body to move; The encoder is used to collect the rotation angle of the distance measuring wheel and convert the rotation angle of the distance measuring wheel into a pulse signal, and calculate the actual moving distance based on the circumference of the distance measuring wheel and the pulse signal; The electromagnetic coil is used to drive the electromagnetic hammer to strike the bridge deck segments; The battery is used to provide power.

7. A bridge deck pavement layer testing system based on multi-point tapping and deep learning, characterized by: A system for implementing a bridge deck pavement delamination test by using a bridge deck pavement delamination test method based on multi-point tapping and deep learning as described in any one of claims 1 to 6, wherein the system comprises an acoustic wave acquisition module, a signal processing module, a feature extraction module, a bridge deck analysis module, a report analysis module, and a bridge deck trend analysis module; The acoustic wave acquisition module is configured to perform the following operations: divide the bridge deck to be inspected into multiple bridge deck segments, and for each bridge deck segment, strike the bridge deck segment using an inspection device equipped with multiple electromagnetic hammers in an equidistant or equidistant manner to form a multi-point excitation source, thereby acquiring an acoustic wave signal in the form of an analog signal; The signal processing module is used to perform the following: signal amplification, filtering and analog-to-digital conversion processing on the collected sound wave signal to obtain a processed sound wave signal, wherein the processed sound wave signal is a digital signal; The feature extraction module is used to perform the following: calculate the time domain features based on the acoustic signal in the form of an analog signal, perform discrete wavelet transform on the acoustic signal in the form of a digital signal, extract the wavelength, amplitude and spectral features of the acoustic signal, calculate the frequency domain features of the acoustic signal, and construct a feature matrix based on the time domain features and frequency domain features; The bridge deck analysis module is configured to perform the following operations: for each bridge deck segment, using the corresponding feature matrix as input, analyze the health status of the bridge deck segment using a trained bridge deck pavement delamination disease detection model, and predict and output analysis results. The analysis results include the bridge deck disease type and disease type probability value, where the disease severity types include no disease, minor disease, moderate disease, and severe disease. The bridge deck pavement delamination disease detection model is a network model constructed based on a convolutional neural network and a long short-term memory network. The report analysis module is used to perform the following: determine the damaged area of the bridge deck based on the analysis results of each bridge deck segment and formulate repair recommendations; construct a pseudo-color heat map of the bridge deck based on the analysis results of each bridge deck segment, and display the damage status of the bridge deck through the pseudo-color heat map; The bridge deck trend analysis module is used to perform the following: periodically perform tap inspections on the target bridge deck, obtain the feature matrix, location coordinates, inspection time and analysis results corresponding to each inspection point on the target bridge deck as historical data, and for each bridge deck segment, perform weighted summation on the probability values of the disease degree type of each relevant inspection point to generate the health index corresponding to the bridge deck segment. The feature matrix and health index corresponding to the bridge deck segment are constructed into a time series sample according to time. The time series sample is used as input and the trained bridge deck trend prediction model is used to perform trend analysis on the bridge deck segment. The prediction output is the disease degree type, disease degree type probability value and health index corresponding to the bridge deck segment. For each bridge deck segment, a trend analysis curve is constructed based on the historical health index and the predicted health index, and a multi-point trend heat map is constructed based on the change rate of the health index of all inspection points. The multi-point trend heat map can be used to display the degree and trend of disease changes in different areas of the target bridge deck.

8. The bridge deck pavement layer testing system based on multi-point tapping and deep learning according to claim 7 is characterized in that: The signal processing module is used to perform the following operations: The acoustic wave signal is amplified by an operational amplifier to obtain an amplified acoustic wave signal; Filtering the amplified sound wave signal through a bandpass filter to extract the sound wave signal within a predetermined frequency range to obtain a filtered sound wave signal; The filtered sound wave signal is converted into a digital sound wave signal through an analog-to-digital converter to obtain a digital sound wave signal.

9. The bridge deck pavement layer testing system based on multi-point tapping and deep learning according to claim 7 is characterized in that: The time domain characteristics of acoustic signals include signal mean, root mean square, peak, kurtosis, skewness and instantaneous energy; the frequency domain characteristics include low-frequency energy, high-frequency energy, spectrum entropy, main frequency, energy ratio and wavelet packet entropy; Signal mean μ x The calculation formula is: Among them, x i represents the signal amplitude of the i-th sampling point, and N represents the total number of sampling points of the acoustic signal, that is, the total number of points at which a knocking signal is sampled when collecting the acoustic signal; The calculation formula for the root mean square RMS is: The formula for calculating peak PV is: PV=x max -x min , Among them, x max represents the maximum amplitude of the knock signal within the target time window, x min Indicates the minimum amplitude of the tapping signal within the target time window; The calculation formula of kurtosis K is: The calculation formula of skewness S is: The calculation formula of instantaneous energy E is: Low frequency energy E A The calculation formula is: Among them, A i Represents the coefficient value in the i-th low-frequency subband after wavelet decomposition, represents the energy contribution of the coefficient; High frequency energy E D The calculation formula is: Among them, D i represents the detail coefficient in the i-th high frequency subband after discrete wavelet transform, Represents the energy contribution of each high-frequency coefficient; The calculation formula of spectral entropy SE is: SE=-∑P i logP i , in, X i Represents the spectral coefficients after discrete wavelet transform; The calculation formula of energy ratio ER is: Main frequency E max It is the maximum energy coefficient obtained based on the wavelet coefficient, and the wavelet packet entropy WE represents the entropy calculated for the wavelet packet decomposition signal.

10. The bridge deck pavement layer testing system based on multi-point tapping and deep learning according to claim 7 is characterized in that: The bridge deck pavement delamination disease detection model includes a CNN feature extraction layer, an LSTM feature extraction layer, a fully connected layer and an output layer; The CNN feature extraction layer includes a convolution layer, an activation layer, and a pooling layer, which is used to extract features from the input feature matrix and output a feature sequence; The LSTM feature extraction layer is used to capture the long-term dependency of the time series by taking the feature sequence output by the CNN feature extraction layer as input; The fully connected layer is used to flatten the output of the LSTM feature extraction layer into a one-dimensional vector and integrate the extracted features through one or more fully connected layers; The output layer is used to diagnose bridge deck defects using the features output by the fully connected layer as input, and output the defect severity type and the probability value of the defect severity type.

Images