Heavy-load highway fatigue crack acoustic emission diagnosis method and system thereof

Abstract

The present application relates to the technical field of highway engineering monitoring, in particular to a heavy-load highway fatigue crack acoustic emission diagnosis method and system, comprising: a multi-depth piezoelectric ceramic sensor array embedded in a pavement structure to collect acoustic emission signals; time-frequency domain topology feature extraction is performed on the collected signals, including wavelet packet entropy value mapping and phase space reconstruction to extract topological invariants such as correlation dimension and Lyapunov exponent; the time-frequency feature vector is input into a transfer learning model to realize multi-scale feature fusion, and a gradual domain is innovatively used to realize smooth transition from a source domain to a target domain; a fractal stress field model is constructed based on the fused features to realize stress diffusion depth estimation and crack evolution prediction; correlation coefficient analysis automatically marks the crack propagation area, generates a crack temperature map and outputs maintenance decision suggestions, realizes early detection of micro-cracks on heavy-load highways, and improves the detection sensitivity by about 35%, and the depth positioning accuracy reaches ±1.5 cm.

Description

Technical Field

[0001] This invention relates to the field of highway engineering monitoring technology, specifically to a method and system for diagnosing fatigue cracks in heavy-duty highways using acoustic emission, and particularly to a method and system for early detection, precise location, and evolution prediction of fatigue cracks in heavy-duty highway pavements by using a multi-depth piezoelectric ceramic sensor array to collect acoustic emission signals and by employing techniques such as time-frequency domain topological feature extraction, transfer learning, and fractal stress field modeling. Background Technology

[0002] With the rapid development of my country's economy, the weight and volume of vehicles carried by highways are constantly increasing, posing a severe durability challenge to heavy-duty highway pavement structures. Under heavy loads, asphalt pavements are prone to fatigue cracks. These cracks often begin at the microscopic level, gradually expanding and affecting the overall performance and service life of the pavement. Traditional pavement inspection methods, such as visual inspection and radar scanning, mainly target macroscopic cracks and are difficult to effectively detect early micro-cracks. Often, cracks are only discovered when they have developed to the point of affecting driving safety, at which point repair costs are high and traffic has already been adversely affected.

[0003] Existing crack detection technologies have several shortcomings: First, traditional surface detection methods, such as image recognition technology, are unable to detect cracks inside the pavement, especially microcracks; second, while ground-penetrating technologies such as ground-penetrating radar can detect the internal structure of the pavement, their resolution is limited, making it difficult to capture micron-level initial cracks; third, existing detection methods are mostly static and lack the ability to monitor the dynamic evolution of cracks; fourth, existing technologies struggle to accurately locate the three-dimensional position of cracks, especially their depth; and finally, there is a lack of effective prediction models to estimate crack propagation trends and the remaining service life of the pavement.

[0004] Acoustic emission (AE) technology, as a non-destructive testing method, can capture transient elastic waves released during the formation and propagation of microcracks within materials, offering advantages such as high real-time performance and high sensitivity. However, existing AE technologies still face challenges when applied to crack detection in heavy-duty highways: on the one hand, the road surface environment is complex, and AE signals are easily interfered with by vehicle vibrations and environmental noise; on the other hand, traditional AE signal processing methods struggle to extract effective crack features from complex backgrounds and lack the ability to detect deep cracks.

[0005] Therefore, there is an urgent need to develop a method and system that can detect early fatigue cracks in heavy-duty highways in real time and accurately, and predict their development trend, so as to provide a scientific basis for highway maintenance decisions. Summary of the Invention

[0006] In view of the shortcomings of the existing technology, the purpose of this invention is to provide an acoustic emission diagnosis method and system for fatigue cracks in heavy-duty highways. The method collects acoustic emission signals through a multi-depth piezoelectric ceramic sensor array and combines advanced technologies such as time-frequency domain topological feature extraction, transfer learning and fractal stress field modeling to achieve early detection, accurate location and evolution prediction of fatigue cracks in heavy-duty highways, thereby providing a scientific basis for pavement maintenance decisions.

[0007] This invention proposes an acoustic emission diagnostic method for fatigue cracks in heavy-duty highways, comprising:

[0008] Acoustic emission signals are collected by a multi-depth piezoelectric ceramic sensor array pre-embedded in the road structure. The multi-depth piezoelectric ceramic sensor array is distributed along the road extension direction and buried at different depths in the vertical direction.

[0009] Extracting time-frequency domain topological features from the acoustic emission signal includes:

[0010] The acoustic emission signal is decomposed into wavelet packets and the entropy value is calculated to obtain the wavelet packet entropy value mapping.

[0011] Construct the phase space of the acoustic emission signal and extract the topological invariants;

[0012] Obtain the time-frequency feature vector of the acoustic emission signal;

[0013] The time-frequency feature vector is input into the transfer learning model for multi-scale feature fusion, including:

[0014] Depth-differentiated features are extracted from signals acquired by sensors at different depths;

[0015] A global feature representation is obtained by fusing features from the different depth sensors through an attention mechanism;

[0016] Gradual domain transfer learning is used to transfer knowledge from the source domain to the target domain.

[0017] Based on the global feature representation, a fractal stress field model is constructed to estimate the stress diffusion depth and predict crack evolution, including:

[0018] The stress is represented as a matrix by equal intervals;

[0019] Calculate the correlation coefficient between adjacent sensor signals;

[0020] When the correlation coefficient is lower than a preset threshold, the crack propagation area is automatically marked;

[0021] Generate crack temperature maps and output maintenance decision recommendations.

[0022] Preferably, the multi-depth piezoelectric ceramic sensor array is arranged as follows:

[0023] Distributed along the road surface extension direction, buried along the road cross section direction, and evenly distributed within the top layer of the subgrade;

[0024] In the vertical direction, a sensor is buried at each adjacent depth, with a height interval of 3 to 5 cm between each sensor in the depth direction;

[0025] The number of piezoelectric ceramic sensors is 10 to 20, the burial depth is 5 to 10 cm, and the burial position is 3 to 5 cm away from the top surface of the road.

[0026] Preferably, the steps of wavelet packet decomposition and entropy calculation specifically include:

[0027] The acoustic emission signal is subjected to multi-level wavelet packet decomposition to obtain multiple frequency band sub-signals;

[0028] Calculate the normalized entropy value Ej(k) for each frequency band sub-signal;

[0029] An acoustic emission event is determined to have occurred when 0.01j(k)≤Ej(k)<0.01j(k)+ε is satisfied, where ε is a tolerance coefficient that is adaptively adjusted according to the signal-to-noise ratio and has a value range of 0.001-0.1.

[0030] Construct an entropy time series and extract statistical characteristics to form an entropy change rate matrix.

[0031] Preferably, the steps of constructing the phase space and extracting topological invariants specifically include:

[0032] The optimal embedding dimension is determined by the pseudo-nearest neighbor method, with a value ranging from 3 to 8 dimensions.

[0033] The optimal time delay is determined using the mutual information method, with a value ranging from 1 to 10 sampling points.

[0034] Mapping a one-dimensional time series to a high-dimensional phase space to construct phase trajectories;

[0035] The correlation dimension, Lyapunov exponent, and recursive graph features were extracted as topological invariants.

[0036] The topological invariants are combined into an eigenvector and then normalized.

[0037] Preferably, the deep differential feature extraction step specifically includes:

[0038] High-frequency components (50-500kHz) are extracted from surface sensors (0-3cm) to reflect the characteristics of surface microcracks;

[0039] Mid-frequency components (10-50kHz) were extracted from the mid-layer sensor (3-6cm) to reflect the characteristics of mid-layer crack development;

[0040] Low-frequency components (0.1-10kHz) are extracted from deep sensors (6-10cm) to reflect the characteristics of deep structural changes;

[0041] The weights of each sensor feature are dynamically adjusted based on signal quality and crack depth.

[0042] Preferably, the steps of the gradient domain transfer learning specifically include:

[0043] Construct the source domain, the target domain, and the gradient domain connecting the two;

[0044] Wherein, the source domain is the acoustic emission characteristics of the reference road surface model, the target domain is the acoustic emission characteristics of the actual detected road surface, and the gradient domain is the continuous transition region from the source domain to the target domain.

[0045] Feature space transformation maps the features of the source and target domains to a shared feature space.

[0046] A multi-layered gradient structure is adopted to achieve a smooth transition from the source domain to the target domain;

[0047] Identify invariant features shared by the source and target domains to enhance transfer performance.

[0048] Preferably, the step of constructing the fractal stress field model specifically includes:

[0049] The stress is represented as an m×n matrix by equal intervals, where m is 10⁻²⁰ and n is 10⁻²⁰.

[0050] Each row represents the sensor sampling points, and each column represents the stress diffusion over time;

[0051] The maximum stress value at each sampling point is taken as the stress value, forming a matrix sequence;

[0052] Define the depth parameter d to represent the maximum depth at which stress diffuses downwards from the surface;

[0053] The diffusion coefficient p is defined as the rate at which stress decays with depth;

[0054] The intensity variation y is defined as the pattern of stress intensity variation with depth.

[0055] Preferably, the step of automatically marking the crack propagation area specifically includes:

[0056] The correlation coefficients between the signals of each piezoelectric ceramic sensor were calculated using the mutual information method.

[0057] Construct a correlation coefficient matrix to reflect the topology of the sensor network;

[0058] Track the changes in correlation coefficients over time to monitor pavement structure stability;

[0059] When the correlation coefficient is lower than the preset threshold of 0.65, it is automatically marked as a crack propagation area;

[0060] The cracked area was divided into nine equidistant rectangles to accurately locate the damage.

[0061] The crack depth is estimated by measuring the changes in the correlation coefficients of sensors at different depths.

[0062] Preferably, the step of generating a crack temperature map and outputting maintenance decision recommendations specifically includes:

[0063] The temperature value is set according to the severity of the crack, and the damage is discretized into a temperature value.

[0064] A crack temperature map is generated based on the crack temperature value and the crack region discretization map;

[0065] A comprehensive damage function is constructed based on crack area, depth, and propagation rate;

[0066] Considering factors such as load class, climate conditions and pavement structure, a mapping relationship between damage degree and remaining service life is established;

[0067] When the correlation coefficient is less than the threshold of 0.65, a heat map is output, and road maintenance tips are generated.

[0068] Based on the stage of crack development, decision-making recommendations are generated for preventive maintenance, restorative maintenance, or reconstructive maintenance.

[0069] The acoustic emission diagnostic system for fatigue cracks in heavy-duty highways implementing the method includes:

[0070] A layered acoustic emission acquisition subsystem, comprising a multi-depth piezoelectric ceramic sensor array and a data acquisition unit, is used to acquire acoustic emission signals;

[0071] The topology feature extraction subsystem is used to extract the time-frequency domain topology features of acoustic emission signals, including a wavelet packet entropy calculation module, a phase space reconstruction module, and a topology invariant extraction module.

[0072] The transfer learning diagnostic subsystem is used to realize multi-scale feature fusion and transfer learning, including a multi-scale feature extraction module, an attention fusion module, and a gradient domain transfer module.

[0073] The crack evolution prediction subsystem is used to estimate the stress diffusion depth and predict crack evolution, including a fractal stress field modeling module, a correlation coefficient analysis module, and a crack location module.

[0074] The intelligent maintenance decision support subsystem is used to generate crack temperature maps and output maintenance decision suggestions, including a thermal map generation module and a maintenance strategy generation module.

[0075] The subsystems are connected via a data bus to form a closed-loop diagnostic system.

[0076] The beneficial effects of this invention include:

[0077] 1. It enables the detection of early microcracks in heavy-duty highways. Compared with traditional methods, the detection sensitivity is improved by about 35%, and it can effectively identify cracks when the crack length is only 1-3mm, discovering potential defects 30-45 days in advance.

[0078] 2. Through the design of a multi-depth sensor array, the precise three-dimensional location of the crack was achieved, especially the depth positioning accuracy reached ±1.5cm, which made up for the shortcomings of traditional methods that mainly focus on surface cracks.

[0079] 3. By adopting the gradual domain transfer learning technique, the adaptability of the system under different road conditions is significantly improved. Compared with traditional transfer learning, the adaptability is improved by about 50%, and it can adapt to complex environments under different climate and load conditions.

[0080] 4. The stress diffusion depth estimation method based on fractal geometry can accurately simulate the nonlinear crack propagation process, improving the prediction accuracy by about 40%, and providing a reliable basis for pavement life assessment.

[0081] 5. Through correlation coefficient analysis and heat map generation technology, the crack area is automatically marked and visualized, which intuitively reflects the road surface damage and makes it easier for maintenance personnel to quickly locate the problem area.

[0082] 6. A maintenance decision support system based on crack development stages was established, realizing the transformation from periodic maintenance to precision maintenance, reducing maintenance costs by about 25% to 30%, and effectively extending the service life of the pavement by 15% to 20%. Attached Figure Description

[0083] Figure 1 This is an overall architecture diagram of the acoustic emission diagnostic system for fatigue cracks in heavy-duty highways according to the present invention.

[0084] Figure 2 This is a flowchart of the time-frequency domain topological feature extraction process for acoustic emission signals according to the present invention.

[0085] Figure 3 This is a diagram of the transfer learning model architecture of the present invention;

[0086] Figure 4 This is a flowchart of the fractal stress field modeling and crack evolution prediction process of this invention.

[0087] Figure 5 This is a flowchart of the maintenance decision support system of the present invention. Detailed Implementation

[0088] Please refer to Figure 1 - Figure 5 The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The scope of protection of the present invention is not limited to the specific embodiments described below.

[0089] Reference Figure 1 The acoustic emission diagnostic system for fatigue cracks in heavy-duty highways provided by this invention includes five subsystems: a hierarchical acoustic emission acquisition subsystem 1, a topological feature extraction subsystem 2, a transfer learning diagnostic subsystem 3, a crack evolution prediction subsystem 4, and an intelligent maintenance decision support subsystem 5. Each subsystem is connected through a data bus to form a closed-loop diagnostic system.

[0090] The hierarchical acoustic emission acquisition subsystem 1 includes a multi-depth piezoelectric ceramic sensor array 11 and a data acquisition unit 12; the topological feature extraction subsystem 2 includes a wavelet packet entropy calculation module 21, a phase space reconstruction module 22, and a topological invariant extraction module 23; the transfer learning diagnosis subsystem 3 includes a multi-scale feature extraction module 31, an attention fusion module 32, and a gradient domain transfer module 33; the crack evolution prediction subsystem 4 includes a fractal stress field modeling module 41, a correlation coefficient analysis module 42, and a crack location module 43; and the intelligent maintenance decision support subsystem 5 includes a heat map generation module 51 and a maintenance strategy generation module 52.

[0091] The acoustic emission diagnostic method for fatigue cracks in heavy-duty highways provided by this invention mainly includes the following steps:

[0092] This invention uses a multi-depth piezoelectric ceramic sensor array to collect acoustic emission signals. The array is distributed along the road surface extension direction and buried at different depths in the vertical direction.

[0093] In a preferred embodiment of the present invention, the multi-depth piezoelectric ceramic sensor array is arranged as follows: distributed along the road surface extension direction, buried along the road cross section direction, and evenly distributed in the top layer of the roadbed; in the vertical direction, one sensor is buried at each adjacent depth, and the height interval of each sensor in the depth direction is 3-5cm; the number of piezoelectric ceramic sensors is 10-20, the burial depth is 5-10cm, and the burial position is 3-5cm away from the top surface of the road surface.

[0094] Preferably, the piezoelectric ceramic sensor uses PZT-5H type piezoelectric ceramic material with a diameter of 15mm, a thickness of 2mm, and a piezoelectric constant d33 of 650pC / N. It exhibits high sensitivity and can effectively capture weak acoustic emission signals. The horizontal spacing between the sensors is preferably 1m, forming a grid-like monitoring network to ensure complete coverage of the monitoring range.

[0095] Furthermore, the connection between the sensor and the data acquisition unit can be wired or wireless. For wired connections, shielded cables are used to reduce electromagnetic interference; for wireless connections, Bluetooth Low Energy or LoRa technology is employed, with a transmission distance of 100–300 meters to meet on-site monitoring needs.

[0096] The data acquisition unit employs 16-channel synchronous acquisition with a sampling rate of 50kHz and a resolution of 16bit, meeting the requirements for high-precision acquisition of acoustic emission signals. The acquisition triggering mechanism includes three modes: time-triggered, event-triggered, and threshold-triggered, flexibly adapting to different monitoring scenarios.

[0097] Reference Figure 2 This invention performs time-frequency domain topological feature extraction on the acquired acoustic emission signal, mainly including two key steps: wavelet packet entropy mapping and phase space reconstruction.

[0098] This invention employs wavelet packet decomposition to perform multi-scale analysis of acoustic emission signals and calculates entropy values ​​to identify acoustic emission events. Specifically, the acoustic emission signal is first subjected to multi-level wavelet packet decomposition to obtain multiple frequency band sub-signals. The preferred wavelet basis function is the db4 wavelet, which has good time-frequency localization characteristics and is suitable for analyzing non-stationary signals. The wavelet packet decomposition level K is adaptively selected based on the signal complexity, typically ranging from 3 to 5 levels.

[0099] Calculate the normalized entropy value for each frequency band sub-signal. The calculation formula is:

[0100] ,

[0101] in: Let be the entropy value of the k-th frequency band at the j-th sampling frequency. Let N be the normalized energy of the i-th sample point, and N be the number of sample points. Normalized energy The calculation method is as follows ,in Let be the amplitude of the signal at the i-th sampling point.

[0102] When satisfied When the acoustic emission event occurs, it is determined that j(k) is the frequency value of the k-th frequency band, in kHz. This is the tolerance coefficient, which is adaptively adjusted based on the signal-to-noise ratio, and its value ranges from 0.001 to 0.1. In practical applications, when the signal-to-noise ratio is high (such as at night or during periods of low traffic), this tolerance factor is less effective. Smaller values, such as 0.001-0.01, can be used when the signal-to-noise ratio is low (e.g., during the day or when traffic is heavy). A larger value, such as 0.05-0.1, can be used to improve noise immunity.

[0103] Furthermore, by constructing an entropy time series and extracting its statistical properties (mean, variance, skewness, kurtosis), an entropy rate of change matrix is ​​formed. Used to characterize crack propagation rate:

[0104] ,

[0105] in: This is the entropy rate of change matrix, with units of 1 / s; Let be the entropy value at time t; For t- The entropy value at time t; The time interval is typically 5 to 20 seconds.

[0106] This invention uses phase space reconstruction technology to map a one-dimensional acoustic emission time series to a high-dimensional phase space to reveal the intrinsic dynamic characteristics of the system. First, the optimal embedding dimension *m* is determined using the pseudo-nearest neighbor method, with a value ranging from 3 to 8. For early-stage microcracks, due to their relatively simple dynamic behavior, *m* is typically taken as 3-4; for complex network cracks, *m* can be taken as 6-8 to fully describe their complexity.

[0107] Meanwhile, the optimal time delay τ is determined using the mutual information method, with a value ranging from 1 to 10 sampling points. In practical applications, for high-frequency signals (50-500kHz), τ is typically taken as 1 to 3 sampling points; for mid-frequency signals (10-50kHz), τ is typically taken as 4 to 6 sampling points; and for low-frequency signals (0.1-10kHz), τ is typically taken as 7 to 10 sampling points.

[0108] Based on the given m and τ, the one-dimensional time series {x(i)} is mapped to the m-dimensional phase space to construct the phase trajectory:

[0109] ,

[0110] in: To reconstruct the i-th point in the phase space, it is an m-dimensional vector; This represents the value of the i-th point in the original time series. The time delay is expressed in the number of sampling points; m is the embedding dimension, representing the dimension of the phase space.

[0111] Extracting topological invariants from the reconstructed phase space, primarily including the correlation dimension. Lyapunov exponent λ and recursive graph features R.

[0112] Correlation dimension The formulas reflecting the system's complexity and degrees of freedom are as follows:

[0113] ,

[0114] in: The correlation dimension is a dimensionless value. The correlation integral represents the proportion of pairs of points in phase space whose distance is less than r; r is the threshold radius, with the same unit as the signal amplitude. The calculation formula is:

[0115] ,

[0116] Where: Θ is the Heaviside step function, which takes the value of 1 when the parameter is greater than 0, and 0 otherwise; Let be the Euclidean distance between the i-th and j-th points in the phase space; N is the total number of points in the phase space.

[0117] For cracked systems Typically non-integer values, reflecting its fractal properties. Early microcracks... The value is relatively small, typically between 1.2 and 1.8; as the crack propagates, The value gradually increases, leading to severe network cracks. The value can reach 2.5-3.0.

[0118] Lyapunov index The formula reflecting the degree of chaos in the system is as follows:

[0119] ,

[0120] in: The Lyapunov exponent is expressed in bits per iteration. Let be the system evolution function, representing the mapping relationship between points in phase space; For the Jacobian matrix at point The value at that point represents the system's sensitivity to initial conditions; This represents the number of iterations.

[0121] During crack propagation A positive value usually indicates that the system has chaotic characteristics. The higher the value, the stronger the instability of crack propagation, and the higher the early warning value. In practice, when... A bit / iteration usually indicates that the crack is expanding rapidly and requires timely intervention.

[0122] Recursive graph features By constructing a recursive graph and extracting diagonal structure features, the periodicity and determinism of the system are reflected. The calculation formula is as follows:

[0123] ,

[0124] in: For the first in the recursive graph Line 1 The elements of the column can take the value 0 or 1; The threshold distance is typically taken as 5% to 15% of the phase space diameter; It is the Heaviside step function; For the first phase space Point and the The Euclidean distance between points.

[0125] Features such as the distribution of diagonal lengths and the number of diagonals in the recurrence plot can be used to distinguish different types of cracks. The recurrence plot of a stable crack usually contains more long diagonals, showing strong determinism; while the recurrence plot of a rapidly propagating crack shows a scattered pattern, with weaker determinism.

[0126] Finally, the topological invariants are combined into an eigenvector. And perform normalization:

[0127] ,

[0128] in: The topological feature vector is a k+2 dimensional vector. For the correlation dimension; Lyapunov index These are the k characteristic parameters of the recursive graph, such as the diagonal length distribution and the number of diagonals.

[0129] The normalization process uses the Min-Max method to ensure that all eigenvalues ​​are within the range of [0,1], thus eliminating the influence of dimensions.

[0130] ,

[0131] in: Let be the i-th eigenvalue after normalization, which is dimensionless; Let i be the i-th eigenvalue; and These are the minimum and maximum values ​​of the i-th feature in the sample set, respectively.

[0132] Reference Figure 3 This invention achieves accurate mapping from acoustic emission signals to crack features through multi-scale feature fusion and transfer learning techniques.

[0133] This invention employs a differentiated strategy to extract features based on the characteristics of different depth sensors:

[0134] High-frequency components (50-500kHz) were extracted from a surface sensor (0-3cm) to reflect the characteristics of surface microcracks. The high-frequency components were extracted using a bandpass filter with cutoff frequencies set at 50kHz and 500kHz. Surface microcracks primarily manifest as high-frequency acoustic emission events, with relatively low energy but a high event rate.

[0135] Mid-frequency components (10-50kHz) were extracted from the mid-layer sensor (3-6cm) to reflect the development characteristics of mid-layer cracks. These mid-frequency components were also extracted using bandpass filters with cutoff frequencies set at 10kHz and 50kHz. The acoustic emission characteristics of mid-layer cracks exhibited medium frequency and energy, with a relatively long duration.

[0136] Low-frequency components (0.1-10kHz) were extracted from deep sensors (6–10cm) to reflect changes in deep structures. These low-frequency components were extracted using bandpass filters with cutoff frequencies set at 0.1kHz and 10kHz. Changes in deep structures typically manifest as low-frequency, high-energy acoustic emission events with the longest duration.

[0137] Furthermore, the weights of each sensor feature are dynamically adjusted based on signal quality and crack depth. Signal quality assessment is based on signal-to-noise ratio (SNR) and entropy, calculated using the following formula:

[0138] ,

[0139] in: The signal quality score for the i-th sensor is dimensionless. Signal-to-noise ratio, in dB; The normalized entropy value (representing the degree of disorder in the signal) ranges from [0,1]. and Let be the weighting coefficient, satisfying In practice, it is usually taken as... The impact of signal-to-noise ratio should be given priority.

[0140] The weights of each sensor are dynamically calculated based on signal quality scores.

[0141] ,

[0142] in: Let be the weight of the i-th sensor, with a value range of [0,1], and the sum of the weights of all sensors is 1; Score the signal quality; This represents the total number of sensors; Let be the depth of the i-th sensor, in cm; The estimated crack depth is in cm. The depth influence coefficient is typically taken as 0.2–0.5 cm. -1This reflects the degree to which depth differences affect the weights.

[0143] This invention innovatively proposes a gradient domain transfer learning method to solve the adaptability problem in traditional transfer learning when the source domain and target domain are significantly different.

[0144] First, a source domain, a target domain, and a transition domain connecting the two are constructed. The source domain is the acoustic emission characteristics (laboratory data) of the reference road surface model, the target domain is the acoustic emission characteristics (field data) of the actual tested road surface, and the transition domain is a continuous transition region from the source domain to the target domain, consisting of multiple intermediate domains.

[0145] Feature space transformation maps features from the source and target domains to a shared feature space:

[0146] ,

[0147] in: This is a feature mapping function that maps the original feature space to a shared feature space; and These are the original features of the source domain and the target domain, respectively; and These are the mapped features, with the same dimension.

[0148] A multi-layered gradient structure is employed to achieve a smooth transition from the source domain to the target domain. Gradient Domain The construction formula is:

[0149] ,

[0150] in: This represents the feature representation of the i-th gradient domain; The mixing coefficients range from [0,1], gradually increasing from 0 to 1 with i, forming a smooth transition from the source domain to the target domain. Typically, 5 to 10 gradient domains are chosen. For example, when 5 gradient domains are chosen, The values ​​can be 0.0, 0.25, 0.5, 0.75, and 1.0 respectively.

[0151] The objective function for gradient domain transfer learning is designed as follows:

[0152] ,

[0153] in: This is the total loss function; For classification loss functions, cross-entropy loss is typically used. This is the maximum mean difference loss, used to reduce the difference in feature distributions between the source and target domains; This is a regularization term to prevent overfitting; , and These are weighting coefficients, which control the importance of each loss term. and These are the features and labels of the source domain, respectively; and These are the features and labels of the gradient domain, respectively. and These are the mapped features of the source and target domains, respectively. These are the model parameters.

[0154] In practical applications, Typically, the value is 0.5-1.0. Take a value of 0.1-0.5. The value is set to 0.001-0.01. The model is trained using a batch size of 32, a learning rate of 0.001, and 100-200 training epochs.

[0155] To adapt to the influence of different load conditions and environmental factors, this invention designs an environmental adaptation mechanism for load parameters.

[0156] First, vehicles are classified according to their load capacity: light load (vehicles weighing less than 5 tons), medium load (vehicles weighing 5 to 15 tons), and heavy load (vehicles weighing more than 15 tons). Different load capacity levels correspond to different stress diffusion characteristics.

[0157] Simultaneously, an environmental factor model was established, considering the influence of factors such as temperature, humidity, and traffic conditions. The effect of temperature was corrected using a temperature-speed-of-sound mapping relationship.

[0158] ,

[0159] in: The speed of sound at temperature T is expressed in m / s. Reference temperature The speed of sound at low altitudes, measured in m / s; For asphalt mixtures, the temperature coefficient is typically taken as (3-5) × 10. -4 / ℃; The current temperature is expressed in °C. For reference temperature, 20℃ is usually taken.

[0160] The effect of humidity is compensated for through a humidity-attenuation coefficient mapping:

[0161] ,

[0162] in: The attenuation coefficient at temperature H is expressed in dB / m. Reference temperature The attenuation coefficient is expressed in dB / m. The humidity influence coefficient is typically taken as 0.01-0.02%; Current humidity, in % % For reference humidity, 50% is usually taken.

[0163] Load and environmental conditions are encoded using a conditional embedding vector C:

[0164] ,

[0165] Where: C is the conditional embedding vector; Independent codes for load categories, such as [1,0,0] for light load, [0,1,0] for medium load, and [0,0,1] for heavy load; For normalized temperature, ,in and These are the lowest and highest temperatures, respectively. To normalize humidity, ,in and These represent the lowest and highest humidity levels, respectively. To normalize traffic flow, the current flow is divided by the maximum flow.

[0166] Conditional embedding vectors are fused with acoustic emission features to form conditional features:

[0167] ,

[0168] in: For conditional features, the dimension is the sum of the dimension of the topological feature vector and the dimension of the conditional embedding vector; These are topological feature vectors; For conditional embedding vectors; The feature fusion operation can be a concatenation, weighted sum, or attention mechanism. In this embodiment, a concatenation operation is used.

[0169] Reference Figure 4 This invention achieves accurate prediction of crack development trends through fractal stress field modeling and crack evolution prediction technology.

[0170] This invention models the crack propagation process as a fractal system, representing stress as an m×n matrix through equidistant division:

[0171] ,

[0172] in: The stress matrix is ​​of size m×n; The value is the stress value in the i-th row and j-th column, in MPa; m is the row number, which is related to the number of sensors and ranges from 10 to 20; n is the column number, which is related to the sampling time period and ranges from 10 to 20. Each row represents a sensor sampling point, and each column represents the stress diffusion over time.

[0173] In practical applications, m is usually related to the number of sensors, and n is related to the sampling time period. For example, for a case with 16 sensors, m can be 16; for a sampling time of 10 seconds and a sampling rate of 50 kHz, considering data downsampling, n can be 20.

[0174] Take the maximum stress value at each sampling point as the stress value to form a matrix sequence:

[0175] ,

[0176] in: is the element in the i-th row and j-th column of the stress matrix, in MPa; Let be the stress value of the i-th sensor at time t, in MPa; Let j be the set of time points in the j-th time period.

[0177] Define the depth parameter d to represent the maximum depth at which stress diffuses downwards from the surface, the diffusion coefficient p to represent the rate at which stress decays with depth, and the intensity variation y to represent the pattern of stress intensity variation with depth:

[0178] ,

[0179] in: This represents the stress value at depth z, in MPa. ρ is the surface stress value in MPa; p is the diffusion coefficient, dimensionless; d is the maximum diffusion depth in cm; z is the depth in cm.

[0180] The parameter values ​​vary depending on the load conditions: under light load, p is usually 1.2-1.5 and d is 3-5cm; under medium load, p is 0.8-1.2 and d is 5-8cm; under heavy load, p is 0.5-0.8 and d is 8-12cm.

[0181] This invention calculates the correlation coefficients between the signals of each piezoelectric ceramic sensor using the mutual information method, and constructs a correlation coefficient matrix. :

[0182] ,

[0183] in: Let be the correlation coefficient between the i-th and j-th sensors, with a value range of [0,1]. Mutual information, measured in bits; and These are the edge entropies of the i-th and j-th sensor signals, respectively, in bits.

[0184] Mutual Information The calculation formula is:

[0185] ,

[0186] in: Let be the joint probability distribution of the i-th and j-th sensor signals; and These are the edge probability distributions of the i-th and j-th sensor signals, respectively; and These are the signal sets of the i-th and j-th sensors, respectively; and These are elements in the signal set.

[0187] By tracking the changes in correlation coefficients over time, the stability of the pavement structure can be monitored. For stable pavement structures, the rate of change of the correlation coefficient matrix is ​​small; however, the correlation coefficient decreases significantly when cracks propagate.

[0188] When the correlation coefficient is below the preset threshold of 0.65, it is automatically marked as a crack propagation area. The threshold of 0.65 is selected based on a large amount of experimental data, and this value can balance detection sensitivity and false alarm rate. In practical applications, for important road sections or key monitoring points, the threshold can be appropriately increased to 0.7-0.75 to improve detection sensitivity; for non-critical areas, the threshold can be appropriately decreased to 0.55-0.6 to reduce false alarms.

[0189] The cracked area was divided into nine equidistant rectangles to accurately locate the damage. The number of divisions was chosen considering both sensor resolution and actual maintenance requirements; nine regions (3×3 grid) provided sufficient spatial resolution while keeping computational complexity within an acceptable range.

[0190] The crack depth is estimated by measuring the changes in the correlation coefficients of sensors at different depths. The formula for crack depth estimation is:

[0191] ,

[0192] in: The estimated crack depth is in cm; The burial depth of the i-th sensor is in cm; This represents the total number of sensors; The weighting coefficient is calculated using the following formula:

[0193] ,

[0194] in: Let be the correlation coefficient between the i-th sensor and the reference sensor; Let be the weight of the i-th sensor, and the sum of the weights of all sensors is 1.

[0195] Reference Figure 5 This invention sets temperature values ​​based on the severity of cracks, discretizes the damage degree into temperature values, and generates a crack temperature map.

[0196] Temperature values ​​are set using a piecewise function:

[0197] ,

[0198] in: This represents the temperature value corresponding to damage level D, in units of 30°C. ; , and These represent the minimum, intermediate, and maximum temperature values, typically taken as 30. 60 and 90 ; , and These represent the lowest, middle, and highest damage levels, typically 0.2, 0.5, and 0.8 respectively; D represents the current damage level, with a value range of [0,1].

[0199] A crack temperature map is generated based on the crack temperature value and the crack region dispersion map. The temperature map is displayed in pseudo-color, with blue indicating minor damage, yellow indicating moderate damage, and red indicating severe damage, visually reflecting the road surface damage condition.

[0200] A comprehensive damage function is constructed based on crack area, depth, and propagation rate:

[0201] ,

[0202] in: The overall damage level is determined by a value ranging from [0,1]. and These represent the normalized crack area, depth, and propagation rate, respectively, all taking values ​​in the range [0,1]. , and Let be the weighting coefficient, satisfying .

[0203] In practical applications, , and The value can be adjusted according to the importance and characteristics of the specific road section. For high-grade highways, it can be appropriately increased. The weighting should focus more on crack propagation rate; for heavily trafficked road sections, the weighting can be appropriately increased. The weighting is adjusted, with more attention paid to crack depth.

[0204] Under normal circumstances, it is advisable to The impact of various factors is considered in a balanced manner. Taking into account factors such as load level, climate conditions, and pavement structure, a mapping relationship between damage degree and remaining service life is established.

[0205] ,

[0206] in: The estimated remaining lifespan is in years; Design lifespan, in years; For overall damage level; The damage index is typically taken as 1.5-2.5; The adjustment function, which takes into account the load rating W, temperature T, and humidity H, typically ranges from 0.8 to 1.2.

[0207] When the correlation coefficient is less than the threshold of 0.65, a heatmap is output, generating road maintenance suggestions. Based on the crack development stage, maintenance decision recommendations are generated.

[0208] For early-stage microcracks (overall damage <0.3), preventative maintenance strategies are recommended, such as surface sealing or fog sealing, which are low-cost and can delay crack propagation.

[0209] For obvious cracks (overall damage grade 0.3-0.7), it is recommended to adopt restorative maintenance strategies, such as crack filling and local milling and repaving, which can effectively repair existing damage and prevent further deterioration.

[0210] For severe damage (overall damage score > 0.7), it is recommended to adopt a reconstructive maintenance strategy, such as heavy intervention measures like structural layer reconstruction, to thoroughly resolve pavement structural problems and restore pavement performance.

[0211] The acoustic emission diagnostic system for fatigue cracks in heavy-duty highways provided by this invention can be used to pre-embed sensors during the construction phase of new highways or to install during the renovation of existing highways. The system is suitable for various types of heavy-duty highways, especially high-grade highways, approach roads to important bridges, and port and wharf access roads, which are critical sections with high traffic volume and heavy loads.

[0212] In terms of hardware implementation, the sensors use durable piezoelectric ceramic materials with a service life of over 10 years; the signal transmission cables are designed to be waterproof and pressure resistant to ensure long-term reliable operation; the edge computing units are located on the roadside and are responsible for near-field data acquisition and preprocessing; the regional control center is responsible for the data aggregation and analysis of multiple edge units within the region; and the central management platform is responsible for global data storage, model training, and decision support.

[0213] In terms of software system architecture, a multi-layered design is adopted: the perception layer is responsible for data acquisition, preprocessing, and feature extraction; the analysis layer is responsible for model training, inference, and diagnosis; the decision layer is responsible for result display, early warning, and decision support; and the interaction layer provides the user interface and service interfaces. Data management adopts a hierarchical storage strategy, with hot data stored at edge nodes, warm data stored at regional centers, and cold data archived to the central platform.

[0214] This invention has been applied in numerous practical engineering projects, achieving significant economic and social benefits. For example, in its application on a heavily loaded section of a provincial highway, the system successfully detected early microcracks 35 days in advance. Through timely preventative maintenance, large-scale repairs caused by crack propagation were avoided, saving approximately 30% of maintenance costs. Furthermore, due to early intervention, the service life of this road section is expected to be extended by approximately 18%, greatly improving the utilization efficiency of road assets.

[0215] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A heavy-duty highway fatigue crack acoustic emission diagnostic method, characterized by, The method comprises the following steps: Collecting acoustic emission signals through a multi-depth piezoelectric ceramic sensor array embedded in a pavement structure, the multi-depth piezoelectric ceramic sensor array being distributed along the extension direction of the pavement and being embedded at different depths in the vertical direction; Performing time-frequency domain topological feature extraction on the acoustic emission signals, comprising: Performing wavelet packet decomposition on the acoustic emission signals and calculating the entropy value to obtain a wavelet packet entropy value mapping; Constructing a phase space of the acoustic emission signals and extracting topological invariants; Obtaining a time-frequency feature vector of the acoustic emission signals; Inputting the time-frequency feature vector into a transfer learning model to perform multi-scale feature fusion, comprising: Performing depth differentiation feature extraction on the signals collected by sensors at different depths; Fusing the features of the sensors at different depths through an attention mechanism to obtain a global feature representation; Transferring source domain knowledge to a target domain using gradual domain transfer learning; Based on the global feature representation, constructing a fractal stress field model to estimate stress diffusion depth and predict crack evolution, comprising: Expressing stress as a matrix through equidistant segmentation; Calculating the correlation coefficient between adjacent sensor signals; When the correlation coefficient is lower than a preset threshold, automatically marking the crack propagation area; Generating a crack temperature map and outputting maintenance decision recommendations; The step of constructing a phase space and extracting topological invariants specifically comprises: Determining the optimal embedding dimension by the false nearest neighbor method, with a value range of 3-8 dimensions; Determining the optimal time delay by the mutual information method, with a value range of 1-10 sampling points; Mapping one-dimensional time series to high-dimensional phase space to construct phase trajectories; Extracting correlation dimension, Lyapunov exponent and recurrence plot features as topological invariants; Combining the topological invariants into a feature vector and performing normalization processing; The step of depth differentiation feature extraction specifically comprises: Extracting high-frequency components of 50-500 kHz from surface layer sensors at 0-3 cm to reflect surface micro-crack features; Extracting medium-frequency components of 10-50 kHz from middle layer sensors at 3-6 cm to reflect middle layer crack development features; Extracting low-frequency components of 0.1-10 kHz from deep layer sensors at 6-10 cm to reflect deep layer structure change features; Dynamically adjusting the weights of the features of each sensor according to signal quality and crack depth; The step of gradual domain transfer learning specifically comprises: Constructing a gradual domain that connects the source domain and the target domain; The source domain is the acoustic emission features of a reference pavement model, the target domain is the acoustic emission features of an actual detection pavement, and the gradual domain is a continuous transition region from the source domain to the target domain; Mapping the source domain and target domain features to a shared feature space through feature space transformation; Using a multi-layer gradual structure to achieve smooth transition from the source domain to the target domain; Identifying invariant features common to the source domain and the target domain to enhance the transfer effect.

2. The heavy truck fatigue crack acoustic emission diagnostic method of claim 1, wherein, The arrangement of the multi-depth piezoelectric ceramic sensor array is as follows: Distributed along the extension direction of the pavement and embedded along the cross-section direction of the road, uniformly distributed in the top layer of the roadbed; In the vertical direction, one sensor is embedded at each adjacent depth, and the height interval of each sensor in the depth direction is 3-5 cm. The piezoelectric ceramic sensors are 10-20 in number, and are buried at a depth of 5-10 cm and at a position 3-5 cm away from the top surface of the road surface.

3. The heavy truck fatigue crack acoustic emission diagnostic method of claim 1, wherein, The wavelet packet decomposition and entropy value calculation step specifically includes: The acoustic emission signal is subjected to multi-level wavelet packet decomposition to obtain a plurality of frequency band sub-signals; The normalized entropy value Ej(k) is calculated for each frequency band sub-signal; When 0.01j(k)≤Ej(k)<0.01j(k)+ε is satisfied, it is determined that an acoustic emission event occurs, wherein ε is a tolerance coefficient adaptively adjusted according to a signal-to-noise ratio, and the value range is 0.001-0.1; An entropy value time sequence is constructed and statistical characteristics are extracted to form an entropy value change rate matrix.

4. The heavy truck fatigue crack acoustic emission diagnostic method of claim 1, wherein, The step of constructing the fractal stress field model specifically includes: The stress is expressed as an m×n matrix through equidistant segmentation, wherein m is 10-20, and n is 10-20; Each row represents a sensor sampling point, and each column represents the diffusion of stress over time; The maximum stress value of each sampling point is taken as the stress value to form a matrix sequence; A depth parameter d is defined to represent the maximum depth of the diffusion of stress from the surface downward; A diffusion coefficient p is defined to represent the rate of attenuation of stress with depth; A strength change y is defined to represent the change pattern of stress strength with depth.

5. The heavy truck fatigue crack acoustic emission diagnostic method of claim 1, wherein, The step of automatically marking the crack propagation area specifically includes: The correlation coefficient between the signals of each piezoelectric ceramic sensor is calculated by the mutual information method; A correlation coefficient matrix is constructed to reflect the topological structure of the sensor network; The change of the correlation coefficient over time is tracked to monitor the stability of the road surface structure; When the correlation coefficient is lower than a preset threshold value 0.65, it is automatically marked as a crack propagation area; The crack area is divided into 9 equidistant rectangles to accurately locate the damage position; The crack depth is estimated through the change of the correlation coefficient of sensors at different depths.

6. The heavy truck fatigue crack acoustic emission diagnostic method of claim 1, wherein, The step of generating a crack temperature map and outputting maintenance decision suggestions specifically includes: The temperature value is set according to the severity of the crack to discretize the damage degree into a temperature value; A crack temperature map is generated according to the crack temperature value and the crack area discrete map; A comprehensive damage degree function is constructed based on the crack area, depth and propagation rate; A mapping relationship between the damage degree and the remaining life is established by considering factors such as load level, climate condition and road surface structure; When the correlation coefficient is less than the threshold value 0.65, a heat map is output to generate a road surface maintenance prompt; According to the crack development stage, decision suggestions for preventive maintenance, repair maintenance or reconstruction maintenance are generated.

7. A heavy truck fatigue crack acoustic emission diagnostic system implementing the method of any one of claims 1-6, characterized in that, It includes: A hierarchical acoustic emission acquisition subsystem including a multi-depth piezoelectric ceramic sensor array and a data acquisition device for acquiring acoustic emission signals; A topological feature extraction subsystem for realizing time-frequency domain topological feature extraction of acoustic emission signals, including a wavelet packet entropy value calculation module, a phase space reconstruction module and a topological invariant extraction module; A transfer learning diagnosis subsystem for realizing multi-scale feature fusion and transfer learning, including a multi-scale feature extraction module, an attention fusion module and a gradual domain transfer module; A crack evolution prediction subsystem for realizing stress diffusion depth estimation and crack evolution prediction, including a fractal stress field modeling module, a correlation coefficient analysis module and a crack positioning module; The intelligent maintenance decision support subsystem is used for generating a crack temperature map and outputting a maintenance decision suggestion, and comprises a heat map generation module and a maintenance strategy generation module. The subsystems are connected through a data bus to form a closed-loop diagnosis system.

Images