| In recent years,the demand for underground electric cable safety is becoming more and more urgent in our country due to the rapid development of urbanization.However,the negligence of the constructors in the road constructions and metro constructions,resulting in electric cable faults and breakdowns happened almost every day.The properties of power grid,manufacturers and people’s safety have been seriously affected.Thus,protecting the underground electric cable which is not damaged by the excavation equipments is essential and urgent for the power system department and urban construction department.This paper presents a deep analysis of the acoustic signal for four representative excavation equipments(namely,excavators,cutting machines,hydraulic hammers,and electric hammers)on the basis of the speech recognition,and then constructs an acoustic signal feature extraction approach based on spectrum dynamic characteristics and the extreme learning machine(ELM)is employed as the classifier in the proposed recognition algorithm.The algorithm can effectively detect the potential damages caused by excavation equipments in the operation to carry out the early warning and achieve the information of the accident area.The main contents are as follows:1.To derive a convincing recognition algorithm,a cross microphone sensor array with 8 channels was used to collect the acoustic samples,and they are captured at different distances between sensor to the excavation equipments in a relatively idea condition at night,which are collected as acoustic feature samples.The acoustic waves of passing automobiles are considered as the most frequently encountered background noise.In addition,to make the system more robust,the testing source data files at different distances are collected in a real construction site during the daytime.2.The feature extraction approach based on Mel Frequency Cepstrual Coefficients(MFCC),and the dynamic features between the temporal frames based on the first and second order difference features of MFCC are denoted as(35)MFCC and(35)(35)MFCC,and last the method of spectrum dynamic features between large numbers of frames is proposed.The contrast experiments are carried out by the feature extraction approach of the acoustic signal features.3.In the pattern recognition stage,the recognition rate,the training model time and the testing time are the most important evaluation standards in this paper.The classical BP feedforward neural network,the traditional KNN and the popular ELM are employed as the three pattern recognition algorithms to identify the acoustic signal types of excavation equipments.4.In the experiment,we conduct some comparison experiments for the MFCC 、(35)MFCC、(35)(35)MFCC and spectrum dynamic features acoustic feature extraction,and then the BP feedforward neural network、the KNN and the ELM are employed as the classifiers.Finally,the feature extraction method based on spectrum dynamic features and ELM as the classifier algorithm is stable by a large number of experiments verification.5.To enhance the robustness of the algorithm,collecting the acoustic data of excavation equipments to verify each kind of equipment working state in the subway site.The results show that the proposed algorithm can accurately identify the excavation equipments and achieve the pre-warning purpose.Finally,the recognition algorithm through the MATLAB software to establish a GUI interface. |
