CN-122017136-A - Pavement disease detection method and system based on vibration signals

CN122017136ACN 122017136 ACN122017136 ACN 122017136ACN-122017136-A

Abstract

The invention discloses a pavement disease detection method and system based on vibration signals, and relates to the technical field of pavement engineering detection. The method comprises the steps of initializing a system, calibrating detection equipment, loading a preset multi-scale SVM-transducer mixed identification model, a disease evolution trend prediction model and a cross-module linkage control center, acquiring road vibration signals and corresponding acquisition position information based on a disease vibration propagation model and an evolution stage adaptive target acquisition mechanism, outputting vibration original data and position data, preprocessing the vibration original data by adopting a time domain frequency domain linkage adaptive filtering algorithm, and outputting preprocessed signals. The invention realizes comprehensive capture and accurate distinction of diseases, improves signal processing quality and recognition reliability, adapts to diseases in different evolution stages and various road surface types, provides comprehensive and practical reference information for road surface maintenance, and assists in scientific maintenance decision.

Inventors

REN TIEJUN
CHEN BING
LIU FANG
KUANG ZHIQUAN
YANG HUI
SHANG JING
WANG CHUNLEI
Qin Wenxiao
JIA JUNWEI
XU FANGHONG
LI SHANSHAN
ZHOU GUANCHEN

Assignees

万邦建工集团有限公司

Dates

Publication Date: 20260512
Application Date: 20260126

Claims (10)

1. The pavement disease detection method based on the vibration signal is characterized by comprising the following steps of: Step 100, initializing a system, calibrating detection equipment, and loading a preset multi-scale SVM-transducer mixed identification model, a disease evolution trend prediction model and a cross-module linkage control center; Step 200, collecting road surface vibration signals and corresponding collecting position information based on a target collecting mechanism adapted to a disease vibration propagation model and an evolution stage, and outputting vibration original data and position data, wherein the target collecting mechanism receives filtering feedback parameters through the cross-module linkage control center and dynamically adjusts sampling frequency, super-sampling multiple and sensor gain, and the disease vibration propagation model is that In which, in the process, For disease distance The amplitude of the vibrations at this point, As the initial amplitude value at the site of the disease, In order to be a vibration attenuation coefficient, Is a natural constant which is used for the production of the high-temperature-resistant ceramic material, The distance between the detection point and the disease; step 300, preprocessing the vibration original data by adopting a time domain frequency domain linkage adaptive filtering algorithm, and outputting a preprocessed signal; The filtering algorithm comprises improved self-adaptive Kalman filtering and disease frequency domain self-adaptive mask filtering, and the cross-module linkage control center receives characteristic distinction degree feedback and dynamically adjusts a filtering threshold and a process noise variance; The process noise variance of the improved adaptive Kalman filtering meets the following condition In which, in the process, Is that The variance of the noise during the time of day, For the initial value of the process noise variance, The characteristic degree of distinction is the mean value; Masking of the disease frequency domain adaptive masking filter When meeting the following conditions Disease frequency band and power spectral density Time of day Otherwise In which, in the process, As a function of the frequency domain mask, In order to be a frequency of the light, As a dynamic threshold value of the value, As a function of the power spectral density of the vibration signal, Maximum value of the power spectral density function; Step 400, carrying out feature extraction and dimension reduction on the preprocessed signals based on a dynamic feedback weight feature system driven by a disease mechanism, and outputting feature vectors, wherein the feature system comprises time domain features, frequency domain features, time-frequency domain features and disease evolution features, weights are dynamically updated by receiving classification confidence feedback through the cross-module linkage control center, and each feature weight is evaluated by a distinguishing degree evaluation function After calculation, distribution is carried out, and adjustment is carried out according to the classification confidence, wherein, Is the first The degree of differentiation of the individual features, Is the sample containing diseases The average value of the individual features is used, Is a disease-free sample The average value of the individual features is used, Is the sample containing diseases The standard deviation of the individual features is used, Is a disease-free sample Standard deviation of individual features; Step 500, inputting the feature vector into the multi-scale SVM-transducer mixed recognition model, and outputting a disease preliminary recognition result and a classification confidence coefficient, wherein the mixed recognition model comprises a multi-scale SVM branch and a transducer encoder branch, and a weighted attention fusion formula is adopted Realize the output fusion, wherein, , LocalScore is the local feature matching, globalScore is the global feature matching, The result is output for the recognition of the SVM branch, Output results for the identification of the transducer encoder branch, Outputting the final identification after fusion; step 600, adopting an evolution adaptive space-time consistency verification mechanism to carry out validity confirmation on the preliminary disease identification result and output a final disease identification result, wherein the verification mechanism dynamically adjusts verification parameters including time verification frame number and space effective point number according to a disease evolution stage and the classification confidence; and 700, outputting and storing a three-dimensional distribution map and an evolution trend report of the pavement disease by combining the position data, the final disease identification result and the disease evolution stage output by the disease evolution trend prediction model.
2. The method for detecting pavement damage based on vibration signals according to claim 1, wherein said step 200 comprises: step 210, dividing a sensor array into a main acquisition unit and an auxiliary acquisition unit, wherein the main acquisition unit is arranged in areas with strong disease vibration on two sides of a front axle of a detection vehicle, and the auxiliary acquisition unit is arranged on two sides of a rear axle of the detection vehicle; Step 220, the disease evolution trend prediction model is based on an LSTM algorithm, inputs historical vibration signal characteristics, disease types and environmental data, and outputs a disease evolution stage, wherein the disease evolution stage comprises a germination stage, an development stage and a maturation stage; step 230, the vibration damping coefficient Adjusting along with the disease evolution stage, and adjusting the germination stage =0.03-0.05, Development stage =0.05-0.1, Maturity stage =0.1-0.2; Step 240, the cross-module linkage control center adjusts acquisition parameters according to the disease evolution stage, wherein the acquisition parameters comprise 6kHz (kilohertz) of sampling frequency and 3 times of supersampling multiple in a germination period, 2 times of supersampling multiple in a development period, 3kHz (kilohertz) of sampling frequency and 1 time of supersampling multiple in a maturation period, and meanwhile, interference residual quantity fed back by filtering is received, and when the interference residual quantity is more than 10%, the sensor gain is improved by 20%; Step 250, synchronously acquiring acquisition position information corresponding to the vibration original data through a positioning module, wherein the positioning accuracy of the positioning module is +/-0.8 m, the vibration original data and the position data are associated according to a time stamp, and the vibration original data and the position data are output.
3. The method for detecting pavement damage based on vibration signals of claim 1, wherein the step 300 includes: Step 310, the process noise variance initial value =0.05, The state equation of the improved adaptive kalman filter is In which, in the process, Is that The time of day system state vector is used, In the form of a system state transition matrix, Is that The time of day system state vector is used, For the noise to drive the matrix, Is that Time process noise, observed noise variance ; Step 320, dynamically adjusting a target frequency interval according to the disease evolution stage, wherein the target frequency interval is 5-150Hz in germination period, 5-300Hz in development period and 5-500Hz in maturation period; Step 330, performing FFT transformation on the time domain denoised signal, and generating a power spectrum density map by using 1024 FFT points, wherein the cross-module coordinated control center receives characteristic distinguishing degree feedback to obtain characteristic distinguishing degree average value When the characteristic distinguishing degree is mean The dynamic threshold value =0.25 When the feature distinguishes the degree mean The dynamic threshold value =0.3, Shielding the corresponding narrowband interference; and step 340, fusing a time domain denoising result with the disease frequency domain self-adaptive mask filtering result, and outputting the preprocessed signal with the signal-to-noise ratio more than or equal to 55 dB.
4. The method for detecting pavement damage based on vibration signals according to claim 1, wherein said step 400 comprises: Step 410, extracting time domain features, frequency domain features, time-frequency domain features and disease evolution features of the preprocessed signals, wherein the time domain features comprise peak values, root mean square values, kurtosis, pulse factors and waveform factors, the frequency domain features comprise power spectrum density peak values, main frequency, spectrum center of gravity and spectrum bandwidth, the time-frequency domain features are energy entropy after wavelet packet decomposition, and the disease evolution features comprise signal rising edge smoothness and spectrum harmonic growth rate; Step 420, calculating the discrimination degree of each feature by the discrimination degree evaluation function Pressing down The initial weights are assigned to the weights, where, Is the first The initial weight of the individual feature(s), The total sum of all the feature distinction degrees is adjusted according to the classification confidence coefficient, wherein the feature weight of the classification confidence coefficient is less than 0.8 and is reduced by 20%, and the feature weight of the classification confidence coefficient is more than or equal to 0.95 and is improved by 30%; and 430, performing dimension reduction processing on the extracted multidimensional features by adopting a PCA algorithm, reserving 99% of feature information, and outputting the feature vectors of 12 dimensions.
5. The method for detecting pavement damage based on vibration signals according to claim 1, wherein said step 500 comprises: Step 510, the multi-scale SVM branch inputs time domain basic features and frequency domain features, captures local features, the transducer encoder branch comprises a multi-layer encoder, inputs the feature vectors, captures global features and outputs global feature vectors; Step 520, calculating LocalScore and GlobalScore, and substituting the calculated result into the weighted attention fusion formula to obtain a fusion output; and 530, training a basic model based on a core sample library, wherein the core sample library comprises labeling samples of asphalt pavement and cement pavement, adapting to the pavement of the masses through a meta-learning algorithm, realizing high recognition rate by only needing a small number of labeling samples of the pavement of the masses, and outputting a preliminary disease recognition result and classification confidence coefficient, wherein the value range of the classification confidence coefficient is 0-1.
6. The method for detecting pavement damage based on vibration signals according to claim 1, wherein said step 600 comprises: step 610, according to the disease evolution stage, adjusting the time verification frame number, namely, 4 continuous frames in the germination stage, 3 continuous frames in the development stage and 2 continuous frames in the maturation stage, judging whether the continuous frame number is the same as the preliminary identification result of the disease; Step 620, adjusting space verification parameters according to the disease evolution stage and the classification confidence coefficient, wherein the space verification parameters comprise more than or equal to 3 effective points in a germination period preset range, more than or equal to 5 effective points in a development period preset range and more than or equal to 4 effective points in a maturity period preset range, and when the classification confidence coefficient is less than 0.8, the number of space effective points is increased by 1; And 630, outputting the final disease identification result if the time verification and the space verification conditions are simultaneously met, otherwise, judging that the signal is an interference signal, and not outputting the final disease identification result.
7. The method for detecting pavement damage based on vibration signals according to claim 1, wherein said step 700 comprises: Step 710, associating the final disease identification result, the position data and the disease evolution stage, and generating the three-dimensional distribution map of the road surface disease and the evolution trend report, wherein the three-dimensional distribution map of the road surface disease comprises the disease position, the type, the severity degree, the disease evolution stage and the detection time, and the longitude and latitude precision of the disease position is +/-0.8 m; And 720, transmitting the three-dimensional distribution map of the road surface diseases and the evolution trend report to a cloud background management system through a communication module, wherein the data transmission rate of the communication module is more than or equal to 100Mbps, and the cloud background management system stores, inquires and counts the three-dimensional distribution map of the road surface diseases and the evolution trend report and automatically generates maintenance suggestions.
8. A pavement defect detection system based on vibration signals, comprising the steps of: The system initialization module is used for calibrating the detection equipment and loading a preset multi-scale SVM-transducer mixed identification model, a disease evolution trend prediction model and a cross-module linkage control center; The system comprises a target acquisition module, a sampling frequency adjustment module, a sampling time adjustment module and a sampling time adjustment module, wherein the target acquisition module is used for acquiring road surface vibration signals and corresponding acquisition position information based on a target acquisition mechanism of a disease vibration propagation model and an evolution stage adaptation and outputting vibration original data and position data; The double-domain self-adaptive preprocessing module is used for preprocessing the vibration original data by adopting a time domain frequency domain linkage self-adaptive filtering algorithm and outputting a preprocessed signal, and receives characteristic distinguishing degree feedback through the cross-module linkage control center and dynamically adjusts a filtering threshold value and a process noise variance; The dynamic feedback characteristic module is used for carrying out characteristic extraction and dimension reduction treatment on the preprocessed signals based on a dynamic feedback weight characteristic system driven by a disease mechanism and outputting characteristic vectors, wherein the characteristic system comprises time domain characteristics, frequency domain characteristics, time-frequency domain characteristics and disease evolution characteristics, and the weights are dynamically updated by receiving classification confidence feedback through the cross-module linkage control center; The mixed recognition module is used for inputting the feature vector into the multi-scale SVM-transducer mixed recognition model and outputting a disease preliminary recognition result and classification confidence, wherein the mixed recognition model comprises a multi-scale SVM branch and a transducer encoder branch, and a formula is fused through weighted attention In which, in the process, , ; The evolution adaptation verification module is used for carrying out validity confirmation on the preliminary disease identification result by adopting an evolution adaptation type space-time consistency verification mechanism and outputting a final disease identification result; And the linkage output module is used for outputting and storing a three-dimensional distribution map and an evolution trend report of the pavement diseases by combining the position data, the final disease identification result and the disease evolution stage.
9. An electronic device, the electronic device comprising: the device comprises a processor and a memory, wherein the memory is in communication connection with the processor; The memory is used for storing executable instructions executed by at least one of the processors, and the processor is used for executing the executable instructions to realize the pavement disease detection method based on the vibration signal according to any one of claims 1 to 7.
10. A computer-readable storage medium, in which a computer program is stored, which computer program, when being executed by a processor, implements a road surface disease detection method based on vibration signals as claimed in any one of claims 1 to 7.

Description

Pavement disease detection method and system based on vibration signals Technical Field The invention relates to the technical field of pavement engineering detection, in particular to a pavement disease detection method and system based on vibration signals. Background The pavement disease detection is a key link of maintenance of traffic infrastructure, accurately identifies the type, position and evolution state of the disease, and has important significance in guaranteeing driving safety, reducing maintenance cost and prolonging service life of the pavement. The detection method based on the vibration signals is one of the main technologies of road surface disease detection because of convenient operation and high detection efficiency, and the core thought is to collect the road surface vibration signals through a sensor and realize disease detection through links such as signal preprocessing, feature extraction, model identification and the like. The existing pavement disease detection method based on vibration signals mostly adopts fixed acquisition parameters, filtering strategies and characteristic weights to carry out detection work, key parameters of detection links cannot be dynamically adjusted by combining vibration propagation characteristics of disease evolution stages, and a depth linkage mechanism is lacked among functional modules. This leads to the prior art to be difficult to accurately catch the characteristic signal of different evolution stage diseases, and the disease adaptability to different development states is not enough, can't satisfy the actual demand of the accurate detection of pavement disease full life cycle. Disclosure of Invention The invention aims to provide a pavement disease detection method and system based on vibration signals, so as to solve the problems in the background technology. In order to achieve the purpose, the invention provides the following technical scheme that the pavement disease detection method based on the vibration signal comprises the following steps: Step 100, initializing a system, calibrating detection equipment, and loading a preset multi-scale SVM-transducer mixed identification model, a disease evolution trend prediction model and a cross-module linkage control center; Step 200, collecting road surface vibration signals and corresponding collecting position information based on a target collecting mechanism adapted to a disease vibration propagation model and an evolution stage, and outputting vibration original data and position data, wherein the target collecting mechanism receives filtering feedback parameters through the cross-module linkage control center and dynamically adjusts sampling frequency, super-sampling multiple and sensor gain, and the disease vibration propagation model is that In which, in the process,For disease distanceThe amplitude of the vibrations at this point,As the initial amplitude value at the site of the disease,In order to be a vibration attenuation coefficient,Is a natural constant which is used for the production of the high-temperature-resistant ceramic material,The distance between the detection point and the disease; step 300, preprocessing the vibration original data by adopting a time domain frequency domain linkage adaptive filtering algorithm, and outputting a preprocessed signal; The filtering algorithm comprises improved self-adaptive Kalman filtering and disease frequency domain self-adaptive mask filtering, and the cross-module linkage control center receives characteristic distinction degree feedback and dynamically adjusts a filtering threshold and a process noise variance; The process noise variance of the improved adaptive Kalman filtering meets the following condition In which, in the process,Is thatThe variance of the noise during the time of day,For the initial value of the process noise variance,The characteristic degree of distinction is the mean value; Masking of the disease frequency domain adaptive masking filter When meeting the following conditionsDisease frequency band and power spectral densityTime of dayOtherwiseIn which, in the process,As a function of the frequency domain mask,In order to be a frequency of the light,As a dynamic threshold value of the value,As a function of the power spectral density of the vibration signal,Maximum value of the power spectral density function; Step 400, carrying out feature extraction and dimension reduction on the preprocessed signals based on a dynamic feedback weight feature system driven by a disease mechanism, and outputting feature vectors, wherein the feature system comprises time domain features, frequency domain features, time-frequency domain features and disease evolution features, weights are dynamically updated by receiving classification confidence feedback through the cross-module linkage control center, and each feature weight is evaluated by a distinguishing degree evaluation function After calculation, distribution is carried