A method and system for intelligent diagnosis and life prediction of hidden damage of heavy-load highway based on dynamic response perception

Abstract

The application discloses a kind of based on dynamic response perception's heavy load highway implicit damage intelligent diagnosis and life prediction method and system, method includes: collection real-time vibration signal and image data and carry out denoising processing, obtain denoised vibration waveform and image sequence;Obtain the axle load data of heavy load highway heavy load vehicle, based on vibration waveform and image sequence, extract the vibration peak feature and crack texture feature corresponding to axle load of heavy load vehicle and fusion, obtain multi-source data matrix;Multi-source data matrix is handled using convolutional neural network and the implicit damage of road surface is identified, and the damage degree is graded based on road surface implicit damage identification result, obtain damage grading result;According to road surface implicit damage identification result, based on the temperature and humidity change data obtained by sensor array Evolution speed judgment, predict damage evolution speed;Based on damage grading result and damage evolution speed, life prediction is carried out in combination with multi-source data matrix, and road surface service life prediction result is obtained.

Description

Technical Field

[0001] This invention relates to the field of highway maintenance technology, specifically to a method and system for intelligent diagnosis and life prediction of latent damage in heavy-load highways based on dynamic response perception. Background Technology

[0002] Highways, as the lifeline of national infrastructure, bear significant responsibility for economic development and social operation. Their service performance and lifespan directly impact traffic safety and transportation efficiency. The prevalence of heavy-load traffic has placed unprecedented pressure on highway pavements, leading to increasingly prominent hidden damage such as cracks and delamination, seriously threatening the long-term stability of roads. Current detection and maintenance methods largely rely on manual inspections or single-device scanning, making it difficult to comprehensively capture the dynamic response characteristics of the pavement under heavy vehicle loads, and they lack adaptability to complex environmental factors. This results in difficulties in the early identification of hidden damage, often leading to delayed maintenance decisions, wasted resources, and shortened road lifespan.

[0003] The core challenge lies in accurately capturing the dynamic correlation between heavy traffic and pavement response, and in achieving precise damage grading and life prediction in complex environments. There is a complex interaction between the axle load characteristics of heavy-duty vehicles and the pavement material response. Existing technologies struggle to integrate multi-source data in real time and to reconstruct the evolution of damage in dynamic traffic scenarios. For example, repeated compaction by overloaded trucks can loosen the base layer, but existing detection methods struggle to distinguish the degree and location of this latent damage, thus failing to provide accurate data for maintenance. Furthermore, environmental factors such as temperature and humidity changes accelerate material fatigue; existing prediction models often ignore the dynamic impact of these variables, leading to biased life assessments and hindering scientific decision-making for preventative maintenance.

[0004] Therefore, how to integrate multi-source dynamic data to achieve accurate identification and classification of hidden damage, and at the same time combine environmental factors to build an accurate life prediction system, has become a key issue in intelligent highway maintenance. Summary of the Invention

[0005] To address the above technical problems, this invention provides a method for intelligent diagnosis and life prediction of latent damage in heavy-load highways based on dynamic response sensing, comprising the following steps: Real-time vibration signals and image data are collected from heavy-duty highways using a sensor array, and the vibration signals and image data are denoised to obtain denoised vibration waveforms and image sequences. Acquire axle load data of heavy-duty vehicles on heavy-duty highways, extract vibration peak features and crack texture features corresponding to the axle load of heavy-duty vehicles based on the vibration waveform and the image sequence, and fuse the vibration peak features and crack texture features to obtain a multi-source data matrix; The multi-source data matrix is ​​processed using a convolutional neural network to identify hidden road surface damage, resulting in a hidden road surface damage identification result. Based on the hidden road surface damage identification result, the degree of damage is classified to obtain a damage classification result. Based on the identification results of the hidden road surface damage, the evolution rate is determined based on the temperature and humidity change data obtained by the sensor array, and the damage evolution rate is predicted. Based on the damage grading results and the damage evolution rate, and combined with the multi-source data matrix, life prediction is performed to obtain the pavement service life prediction result.

[0006] Preferably, the method for obtaining the multi-source data matrix includes: Based on the vibration waveform, the vibration peak feature corresponding to the axle load data is extracted using Fourier transform; Based on the image sequence, a convolutional neural network is used to extract the crack texture features corresponding to the axle load data; The multi-source data matrix is ​​obtained by fusing the vibration peak features and the crack texture features using principal component analysis.

[0007] Preferably, the method for obtaining the damage grading result includes: The multi-source data matrix is ​​standardized and denoised to obtain a preprocessed data matrix; The convolutional neural network extracts spatial features from the preprocessed data matrix to obtain a set of feature vectors, and performs damage pattern analysis on the set of feature vectors to obtain the damage pattern distribution. If the damage pattern distribution matches the preset damage type template, the hidden damage type of the road surface is determined, and a classification label is output to obtain the hidden damage identification result of the road surface. The hidden damage types of the road surface include: interlayer peeling and base layer loosening. The results of the hidden road surface damage identification are input into a pre-trained grading model to classify the degree of damage, thereby obtaining the damage grading results.

[0008] Preferably, the method for predicting the rate of damage evolution includes: The sensor array collects environmental temperature and humidity change data, and the temperature and humidity change data is cleaned to obtain cleaned temperature and humidity data. The temperature and humidity data after cleaning are fused with the road surface hidden damage identification results using a data fusion algorithm to obtain fused label data; If the fused label data meets the preset damage evolution conditions, the random forest algorithm is used to determine the damage evolution speed and obtain the damage evolution speed.

[0009] Preferably, the method for obtaining the pavement service life prediction result includes: The damage grading results and the damage evolution rate are reduced in dimensionality using feature extraction to obtain a simplified feature set; By integrating the multi-source data matrix and the simplified feature set, a structured time series dataset is generated using time series analysis methods. Using a pre-trained damage evolution model, the damage evolution trend is simulated based on the time series dataset to generate an evolution trend sequence; Based on the evolution trend sequence, a multinomial regression method is used to fit the prediction curve to obtain the pavement service life prediction result.

[0010] The present invention also provides an intelligent diagnosis and life prediction system for hidden damage of heavy-load highways based on dynamic response perception. The prediction system applies the above-mentioned prediction method and includes: a data acquisition module, a data fusion module, a damage grading module, a damage evolution module and a life prediction module. The data acquisition module collects real-time vibration signals and image data from the heavy-load highway through a sensor array, and performs noise reduction processing on the vibration signals and image data to obtain the noise-reduced vibration waveform and image sequence. The data fusion module is used to acquire heavy-duty vehicle axle load data on heavy-duty highways. Based on the vibration waveform and the image sequence, it extracts the vibration peak features and crack texture features corresponding to the heavy-duty vehicle axle load, and fuses the vibration peak features and crack texture features to obtain a multi-source data matrix. The damage grading module uses a convolutional neural network to process the multi-source data matrix and identify hidden road surface damage, obtains the hidden road surface damage identification result, and grades the degree of damage based on the hidden road surface damage identification result to obtain the damage grading result. The damage evolution module determines the evolution rate based on the hidden road surface damage identification results and the temperature and humidity change data acquired by the sensor array, and predicts the damage evolution rate. The life prediction module performs life prediction based on the damage classification results and the damage evolution rate, combined with the multi-source data matrix, to obtain the pavement life prediction result.

[0011] Preferably, the workflow of the data fusion module includes: Based on the vibration waveform, the vibration peak feature corresponding to the axle load data is extracted using Fourier transform; Based on the image sequence, a convolutional neural network is used to extract the crack texture features corresponding to the axle load data; The multi-source data matrix is ​​obtained by fusing the vibration peak features and the crack texture features using principal component analysis.

[0012] Preferably, the workflow of the damage grading module includes: The multi-source data matrix is ​​standardized and denoised to obtain a preprocessed data matrix; The convolutional neural network extracts spatial features from the preprocessed data matrix to obtain a set of feature vectors, and performs damage pattern analysis on the set of feature vectors to obtain the damage pattern distribution. If the damage pattern distribution matches the preset damage type template, the hidden damage type of the road surface is determined, and a classification label is output to obtain the hidden damage identification result of the road surface. The hidden damage types of the road surface include: interlayer peeling and base layer loosening. The results of the hidden road surface damage identification are input into a pre-trained grading model to classify the degree of damage, thereby obtaining the damage grading results.

[0013] Preferably, the workflow of the damage evolution module includes: The sensor array collects environmental temperature and humidity change data, and the temperature and humidity change data is cleaned to obtain cleaned temperature and humidity data. The temperature and humidity data after cleaning are fused with the road surface hidden damage identification results using a data fusion algorithm to obtain fused label data; If the fused label data meets the preset damage evolution conditions, the random forest algorithm is used to determine the damage evolution speed and obtain the damage evolution speed.

[0014] Preferably, the workflow of the lifetime prediction module includes: The damage grading results and the damage evolution rate are reduced in dimensionality using feature extraction to obtain a simplified feature set; By integrating the multi-source data matrix and the simplified feature set, a structured time series dataset is generated using time series analysis methods. Using a pre-trained damage evolution model, the damage evolution trend is simulated based on the time series dataset to generate an evolution trend sequence; Based on the evolution trend sequence, a multinomial regression method is used to fit the prediction curve to obtain the pavement service life prediction result.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention addresses the challenge of accurately identifying and predicting the evolution of latent road damage under heavy vehicle loads. It collects road vibration signals and image data, employs filtering algorithms to purify the data, extracts peak axle load and crack texture features, and constructs a multi-source data matrix. A convolutional neural network is then used for deep learning of the matrix to identify damage types such as interlayer delamination and base layer loosening. Temperature and humidity data are integrated to determine the rate of damage evolution, and a support vector machine is used to classify the degree of damage. Combined with historical data, the evolution process is simulated to generate a remaining life prediction curve. Through multi-source data fusion and intelligent analysis, this invention achieves accurate location, classification, and evolution prediction of latent damage, significantly improving the scientific nature and efficiency of road maintenance and providing an intelligent solution for infrastructure management. Attached Figure Description

[0016] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the method flow according to an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this disclosure should have the ordinary meaning understood by one of ordinary skill in the art to which this disclosure pertains. The terms "first," "second," and similar terms used in the embodiments of this disclosure do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0020] Example 1 In this embodiment, as Figure 1 As shown, a method for intelligent diagnosis and life prediction of latent damage in heavy-load highways based on dynamic response perception includes the following steps: S1. Real-time vibration signals and image data are collected from heavy-duty highways through a sensor array, and the vibration signals and image data are denoised to obtain denoised vibration waveforms and image sequences.

[0021] In this embodiment, the sensor array includes a vibration sensor, an image sensor, and a temperature and humidity sensor. When monitoring heavy-duty highways, the vibration signal frequency range acquired by the vibration sensor is 0-1000Hz, with low-frequency noise mainly concentrated in the 0-50Hz range. Using a high-pass filter with a cutoff frequency of 50Hz effectively removes low-frequency interference while retaining the 200-800Hz high-frequency vibrations caused by road surface cracks or potholes, resulting in a purified vibration waveform. The road surface image sequence captured by the image sensor shows the expansion of cracks over time. Wavelet transform can decompose the image into different frequency sub-bands, extracting the texture change features corresponding to the 300Hz vibration to obtain the image sequence.

[0022] S2. Obtain axle load data of heavy-duty vehicles on heavy-duty highways. Based on vibration waveforms and image sequences, extract vibration peak features and crack texture features corresponding to the axle load of heavy-duty vehicles. Then, fuse the vibration peak features and crack texture features to obtain a multi-source data matrix.

[0023] The methods for obtaining the multi-source data matrix include: extracting vibration peak features corresponding to axle load data using Fourier transform based on vibration waveforms; extracting crack texture features corresponding to axle load data using convolutional neural networks based on image sequences; and fusing vibration peak features and crack texture features using principal component analysis to obtain the multi-source data matrix.

[0024] In this embodiment, firstly, the vibration waveform data is transformed to the frequency domain using Fast Fourier Transform (FFT) to extract vibration peak features corresponding to the vehicle axle load data. Vibration signal segments 0.5 seconds before and after the time point corresponding to the identified vehicle axle load data are selected. After windowing and spectral analysis, vibration peaks significantly higher than the background noise in the 50–1000 Hz frequency band and their corresponding frequencies are extracted to form vibration peak features, including indicators such as dominant frequency amplitude, energy integral, and frequency band power. Secondly, road surface images of 5 frames before and after the time when the vehicle axle load data was identified are located in the image sequence. Crack texture features are extracted using a pre-trained convolutional neural network (such as ResNet-50). Based on the ImageNet pre-trained weights, the network is fine-tuned using a crack image dataset. Finally, 1024-dimensional crack texture features are extracted from before the fully connected layer to characterize the crack morphology, direction, width variation, and texture complexity. Finally, principal component analysis (PCA) was used to fuse and reduce the dimensionality of vibration peak features and image texture features: first, the two types of feature vectors were standardized and spliced ​​into high-dimensional mixed features, and then the main components were extracted by PCA to eliminate redundant information and form a low-dimensional multi-source data matrix with good separability.

[0025] S3. Use a convolutional neural network to process the multi-source data matrix and identify hidden road surface damage, obtain the hidden road surface damage identification results, and classify the degree of damage based on the hidden road surface damage identification results to obtain the damage classification results.

[0026] The method for obtaining damage grading results includes: standardizing and denoising the multi-source data matrix to obtain a preprocessed data matrix; extracting spatial features from the preprocessed data matrix using a convolutional neural network to obtain a set of feature vectors, and performing damage pattern analysis on the feature vector set to obtain a damage pattern distribution; if the damage pattern distribution matches a preset damage type template, the type of hidden pavement damage is determined, and a classification label is output to obtain the pavement hidden damage identification result, which includes: interlayer peeling and base layer loosening; inputting the pavement hidden damage identification result into a pre-trained grading model to classify the degree of damage to obtain the damage grading result.

[0027] In this embodiment, firstly, the Z-score normalization method is used to adjust the feature dimensions of the multi-source data matrix to zero mean and unit variance to eliminate dimensional differences; then, the wavelet threshold denoising method is used to filter out residual noise in the matrix, enhance the data signal-to-noise ratio, and finally obtain the preprocessed data matrix. Next, a convolutional neural network (CNN) structure with multi-scale receptive fields is constructed. This network consists of three convolutional layers and two max-pooling layers arranged alternately, as well as a fully connected layer. The convolutional kernel sizes are set to 3×3 and 5×5 to extract spatial features at different scales. Batch normalization and ReLU activation functions are applied after each convolutional layer to accelerate model convergence and enhance nonlinear expressive power. This network extracts high-order features from the preprocessed data matrix, which are compressed by a global average pooling layer to output a 256-dimensional feature vector set. Then, the extracted feature vectors are input into the fully connected layer for damage pattern analysis. This layer uses a Softmax classifier to calculate the matching probability between the feature vector and a preset damage type template. The template includes two typical latent damage types: "interlayer peeling" and "loose basal layer". If the cosine similarity between the feature distribution of a sample and a template exceeds a threshold of 0.85, it is determined to be a damage of that type, and the corresponding classification label is output as the latent damage identification result. Finally, the identification result is input into a pre-trained damage grading model. This model employs a Support Vector Machine (SVM) structure, trained based on vibration peak features, texture features, and temperature and humidity data of different damage levels from historical data. It categorizes damage levels into three grades: "mild," "moderate," and "severe." By partitioning the feature space using a hyperplane, it outputs the damage grading results and their probability confidence scores.

[0028] S4. Based on the results of identifying hidden road surface damage, the evolution rate is determined by the temperature and humidity change data obtained from the sensor array, and the damage evolution rate is predicted.

[0029] The method for predicting the rate of damage evolution includes: collecting environmental temperature and humidity change data through a sensor array, cleaning the temperature and humidity change data to obtain cleaned temperature and humidity data; using a data fusion algorithm to fuse the cleaned temperature and humidity data with the road surface hidden damage identification results to obtain fused label data; if the fused label data meets the preset damage evolution conditions, then the random forest algorithm is used to determine the damage evolution rate to obtain the damage evolution rate.

[0030] In this embodiment, firstly, ambient temperature and humidity change data are continuously collected using a temperature and humidity sensor at a sampling frequency of 1Hz. Due to environmental interference, the original temperature and humidity data contains outliers and transient impulse noise. A sliding window median filter is used for data cleaning: the window length is set to 60 seconds, and the median of the temperature and humidity data within the window is calculated. Outliers deviating from the median by more than a threshold (temperature: ±2℃, humidity: ±5%) are replaced, resulting in a smooth and consistent cleaned temperature and humidity data sequence. Subsequently, a Kalman filter-based data fusion algorithm is used to fuse the cleaned temperature and humidity data with the pavement hidden damage identification results. Specifically, the damage type label (e.g., "interlayer peeling" or "loose base layer") corresponding to each time point is combined with the current temperature and humidity data to form a multi-dimensional observation vector. The damage-environment joint state is estimated using a state-space model, generating spatiotemporally consistent fused label data, where each sample point includes the damage type, temperature, humidity, and corresponding timestamp. Next, the damage evolution judgment conditions are set: if the same damage type appears continuously in the fused label data, and the temperature and humidity data meet any of the following conditions, then the damage evolution conditions are met: (1) the temperature is higher than 30℃ and the humidity is greater than 70%; (2) the temperature change rate exceeds 0.5℃ / min; (3) the humidity remains above 80% for more than 10 minutes. For the fused label data segment that meets the above conditions, the random forest algorithm is used to judge the damage evolution speed. The random forest model has been trained on the historical dataset. The input is the features of the current damage type, the mean temperature and humidity and the change rate, and the output is the evolution speed level ("slow", "medium", "fast") or the expansion rate in the form of continuous numerical values ​​(such as mm / day). The model votes or regresses the features through multiple decision trees, and the final aggregation result is the damage evolution speed.

[0031] S5. Based on the damage grading results and damage evolution rate, and combined with the multi-source data matrix, life prediction is performed to obtain the pavement service life prediction results.

[0032] The methods for obtaining pavement service life prediction results include: using feature extraction to reduce the dimensionality of damage grading results and damage evolution rate to obtain a simplified feature set; fusing multi-source data matrices and the simplified feature set, and using time series analysis methods to generate a structured time series dataset; using a pre-trained damage evolution model to simulate the damage evolution trend based on the time series dataset to generate an evolution trend sequence; and using a multinomial regression method to fit the prediction curve based on the evolution trend sequence to obtain the pavement service life prediction results.

[0033] In this embodiment, firstly, principal component analysis is used to perform feature dimensionality reduction on the damage grading results (including three categories of labels: "mild," "moderate," and "severe" and their probability confidence levels) and damage evolution speed (including levels such as "slow," "medium," and "fast," or continuous values ​​such as millimeters per day). Categorical and numerical variables are uniformly encoded into low-dimensional feature vectors to eliminate redundant information and obtain a simplified feature set. Then, the multi-source data matrix and the simplified feature set are fused: the feature dimensions of the two are aligned on the time axis using tensor splicing technology, and a sliding time window (window length of 10 minutes, step size of 1 minute) is used to segment and integrate the fused high-dimensional data to generate a time-ordered multidimensional structured dataset. Based on this, the autoregressive integral moving average (ARIMA) model is used to perform time series analysis on the structured dataset to extract the trend components, periodic components, and residual components of damage indicators over time, constructing an enhanced time series dataset that can be used for life prediction. Subsequently, the enhanced time-series dataset is input into a pre-trained damage evolution model for trend simulation. In this embodiment, the damage evolution model employs a recurrent neural network structure based on Long Short-Term Memory (LSTM), which has been trained on historical damage evolution data. It can simulate the expansion process of damage such as interlayer delamination and base layer loosening under different temperature, humidity, and load conditions. The model output is the evolution sequence of damage indicators (such as crack width and vibration energy) over a future period, i.e., the evolution trend sequence. Finally, based on the evolution trend sequence, a multinomial regression method is used to fit the relationship curve between damage indicators and time. The optimal polynomial order is determined using the least squares method, and the time point when the damage indicators exceed a preset threshold (such as crack width greater than 25 mm or vibration energy increasing by 50%) is used as the life termination critical point. The remaining service life under the current state is then calculated in reverse, and the final pavement service life prediction result is output.

[0034] It should be noted that the method of this disclosure embodiment can be executed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method of this disclosure embodiment, and the multiple devices will interact with each other to complete the method described.

[0035] It should be noted that the above description describes some embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, it should be understood that the sequence number of each step in the above embodiments does not imply the order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention. The actions or steps recorded in the claims can be performed in a different order than that in the above embodiments and can still achieve the desired result. In addition, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0036] Example 2 In this embodiment, a system for intelligent diagnosis and life prediction of latent damage in heavy-load highways based on dynamic response perception includes: a data acquisition module, a data fusion module, a damage grading module, a damage evolution module, and a life prediction module.

[0037] The data acquisition module collects real-time vibration signals and image data from the heavy-duty highway through a sensor array, and performs noise reduction processing on the vibration signals and image data to obtain the noise-reduced vibration waveforms and image sequences.

[0038] The data fusion module is used to acquire axle load data of heavy-duty vehicles on heavy-duty highways. Based on vibration waveforms and image sequences, it extracts vibration peak features and crack texture features corresponding to the axle load of heavy-duty vehicles, and fuses the vibration peak features and crack texture features to obtain a multi-source data matrix.

[0039] The workflow of the data fusion module includes: extracting vibration peak features corresponding to axle load data using Fourier transform based on vibration waveforms; extracting crack texture features corresponding to axle load data using convolutional neural networks based on image sequences; and fusing vibration peak features and crack texture features using principal component analysis to obtain a multi-source data matrix.

[0040] The damage grading module uses a convolutional neural network to process the multi-source data matrix and identify hidden road surface damage, obtaining the hidden road surface damage identification results. Based on the hidden road surface damage identification results, the degree of damage is graded to obtain the damage grading results.

[0041] The workflow of the damage grading module includes: standardizing and denoising the multi-source data matrix to obtain a preprocessed data matrix; extracting spatial features from the preprocessed data matrix using a convolutional neural network to obtain a set of feature vectors, and performing damage pattern analysis on the feature vector set to obtain a damage pattern distribution; if the damage pattern distribution matches a preset damage type template, the type of hidden pavement damage is determined, and a classification label is output to obtain the pavement hidden damage identification result, which includes: interlayer peeling and base layer loosening; inputting the pavement hidden damage identification result into a pre-trained grading model to classify the degree of damage to obtain the damage grading result.

[0042] The damage evolution module determines the rate of damage evolution based on the results of identifying hidden road surface damage and the temperature and humidity change data obtained by the sensor array.

[0043] The workflow of the damage evolution module includes: collecting environmental temperature and humidity change data through a sensor array, cleaning the temperature and humidity change data to obtain cleaned temperature and humidity data; using a data fusion algorithm to fuse the cleaned temperature and humidity data with the road surface hidden damage identification results to obtain fused label data; if the fused label data meets the preset damage evolution conditions, the random forest algorithm is used to determine the damage evolution speed to obtain the damage evolution speed.

[0044] The life prediction module performs life prediction based on damage grading results and damage evolution rate, combined with a multi-source data matrix, to obtain the pavement life prediction result.

[0045] The workflow of the life prediction module includes: using feature extraction to reduce the dimensionality of damage grading results and damage evolution rate to obtain a simplified feature set; fusing multi-source data matrices and the simplified feature set, and using time series analysis methods to generate a structured time series dataset; using a pre-trained damage evolution model to simulate the damage evolution trend based on the time series dataset to generate an evolution trend sequence; and using a multinomial regression method to fit the prediction curve based on the evolution trend sequence to obtain the pavement life prediction result.

[0046] The system described in the above embodiments is used to implement the corresponding method for intelligent diagnosis and life prediction of hidden damage of heavy-load highways based on dynamic response perception in any of the foregoing embodiments, and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0047] It should be noted that the aforementioned intelligent diagnosis and life prediction system for latent damage on heavy-duty highways based on dynamic response perception is embodied in the form of functional units. The term "module" here can be implemented in software and / or hardware, without specific limitations.

[0048] For example, a "module" can be a software program, hardware circuit, or a combination of both that implements the above functions. Hardware circuits may include application-specific integrated circuits (ASICs), electronic circuits, processors (e.g., shared processors, proprietary processors, or group processors) and memory for executing one or more software or firmware programs, combined logic circuits, and / or other suitable components that support the described functions.

Claims

1. A method for intelligent diagnosis and life prediction of latent damage in heavy-load highways based on dynamic response sensing, characterized in that, Includes the following steps: Real-time vibration signals and image data are collected from heavy-duty highways using a sensor array, and the vibration signals and image data are denoised to obtain denoised vibration waveforms and image sequences. Acquire axle load data of heavy-duty vehicles on heavy-duty highways, extract vibration peak features and crack texture features corresponding to the axle load of heavy-duty vehicles based on the vibration waveform and the image sequence, and fuse the vibration peak features and crack texture features to obtain a multi-source data matrix; The multi-source data matrix is ​​processed using a convolutional neural network to identify hidden road surface damage, resulting in a hidden road surface damage identification result. Based on the hidden road surface damage identification result, the degree of damage is classified to obtain a damage classification result. Based on the identification results of the hidden road surface damage, the evolution rate is determined based on the temperature and humidity change data obtained by the sensor array, and the damage evolution rate is predicted. Based on the damage classification results and the damage evolution rate, and combined with the multi-source data matrix, life prediction is performed to obtain the pavement life prediction results. The methods for obtaining the multi-source data matrix include: Based on the vibration waveform, the vibration peak feature corresponding to the axle load data is extracted using Fourier transform; Based on the image sequence, a convolutional neural network is used to extract the crack texture features corresponding to the axle load data; The multi-source data matrix is ​​obtained by fusing the vibration peak features and the crack texture features using principal component analysis; The methods for obtaining the damage grading results include: The multi-source data matrix is ​​standardized and denoised to obtain a preprocessed data matrix; The convolutional neural network extracts spatial features from the preprocessed data matrix to obtain a set of feature vectors, and performs damage pattern analysis on the set of feature vectors to obtain the damage pattern distribution. If the damage pattern distribution matches the preset damage type template, the hidden damage type of the road surface is determined, and a classification label is output to obtain the hidden damage identification result of the road surface. The hidden damage types of the road surface include: interlayer peeling and base layer loosening. The results of the hidden road surface damage identification are input into a pre-trained grading model to classify the degree of damage, thereby obtaining the damage grading results. Methods for predicting the rate of damage evolution include: The sensor array collects environmental temperature and humidity change data, and the temperature and humidity change data is cleaned to obtain cleaned temperature and humidity data. The temperature and humidity data after cleaning are fused with the road surface hidden damage identification results using a data fusion algorithm to obtain fused label data; If the fused label data meets the preset damage evolution conditions, the random forest algorithm is used to determine the damage evolution speed and obtain the damage evolution speed. The methods for obtaining the predicted service life of the road surface include: The damage grading results and the damage evolution rate are reduced in dimensionality using feature extraction to obtain a simplified feature set; By integrating the multi-source data matrix and the simplified feature set, a structured time series dataset is generated using time series analysis methods. Using a pre-trained damage evolution model, the damage evolution trend is simulated based on the time series dataset to generate an evolution trend sequence; Based on the evolution trend sequence, a multinomial regression method is used to fit the prediction curve to obtain the pavement service life prediction result.

2. A smart diagnostic and life prediction system for latent damage on heavy-load highways based on dynamic response sensing, wherein the prediction system applies the prediction method described in claim 1, characterized in that, include: The system includes a data acquisition module, a data fusion module, a damage grading module, a damage evolution module, and a lifespan prediction module. The data acquisition module collects real-time vibration signals and image data from the heavy-load highway through a sensor array, and performs noise reduction processing on the vibration signals and image data to obtain the noise-reduced vibration waveform and image sequence. The data fusion module is used to acquire axle load data of heavy-duty vehicles on heavy-duty highways. Based on the vibration waveform and the image sequence, it extracts the vibration peak features and crack texture features corresponding to the axle load of heavy-duty vehicles, and fuses the vibration peak features and crack texture features to obtain a multi-source data matrix. The damage grading module uses a convolutional neural network to process the multi-source data matrix and identify hidden road surface damage, obtains the hidden road surface damage identification result, and grades the degree of damage based on the hidden road surface damage identification result to obtain the damage grading result. The damage evolution module determines the evolution rate based on the hidden road surface damage identification results and the temperature and humidity change data acquired by the sensor array, and predicts the damage evolution rate. The life prediction module performs life prediction based on the damage classification results and the damage evolution rate, combined with the multi-source data matrix, to obtain the pavement life prediction result.

3. The intelligent diagnosis and life prediction system for latent damage of heavy-load highways based on dynamic response perception according to claim 2, characterized in that, The workflow of the data fusion module includes: Based on the vibration waveform, the vibration peak feature corresponding to the axle load data is extracted using Fourier transform; Based on the image sequence, a convolutional neural network is used to extract the crack texture features corresponding to the axle load data; The multi-source data matrix is ​​obtained by fusing the vibration peak features and the crack texture features using principal component analysis.

4. The intelligent diagnosis and life prediction system for latent damage of heavy-load highways based on dynamic response perception according to claim 2, characterized in that, The workflow of the damage grading module includes: The multi-source data matrix is ​​standardized and denoised to obtain a preprocessed data matrix; The convolutional neural network extracts spatial features from the preprocessed data matrix to obtain a set of feature vectors, and performs damage pattern analysis on the set of feature vectors to obtain the damage pattern distribution. If the damage pattern distribution matches the preset damage type template, the hidden damage type of the road surface is determined, and a classification label is output to obtain the hidden damage identification result of the road surface. The hidden damage types of the road surface include: interlayer peeling and base layer loosening. The results of the hidden road surface damage identification are input into a pre-trained grading model to classify the degree of damage, thereby obtaining the damage grading results.

5. The intelligent diagnosis and life prediction system for latent damage of heavy-load highways based on dynamic response perception according to claim 2, characterized in that, The workflow of the damage evolution module includes: The sensor array collects environmental temperature and humidity change data, and the temperature and humidity change data is cleaned to obtain cleaned temperature and humidity data. The temperature and humidity data after cleaning are fused with the road surface hidden damage identification results using a data fusion algorithm to obtain fused label data; If the fused label data meets the preset damage evolution conditions, the random forest algorithm is used to determine the damage evolution speed and obtain the damage evolution speed.

6. The intelligent diagnosis and life prediction system for latent damage of heavy-load highways based on dynamic response perception according to claim 2, characterized in that, The workflow of the lifetime prediction module includes: The damage grading results and the damage evolution rate are reduced in dimensionality using feature extraction to obtain a simplified feature set; By integrating the multi-source data matrix and the simplified feature set, a structured time series dataset is generated using time series analysis methods. Using a pre-trained damage evolution model, the damage evolution trend is simulated based on the time series dataset to generate an evolution trend sequence; Based on the evolution trend sequence, a multinomial regression method is used to fit the prediction curve to obtain the pavement service life prediction result.

Images