Research On Engineering Machinery Identification Based On Statistical Analysis Of Working Sound Signals

Underground cable was widely used in our country as a kind of economic and security way.But underground cables are often damaged,and the security of the cable is seriously threatened.The statistics which from nations electric power cable fault reasons in recent 3 years,show that the external force damage is the main cause of underground cable damage,accounted for more than half the failure rate.These external forces mainly come from: hand-held electric hammer,cutting machine,hydraulic hammer,excavators and others.Therefore,we can monitor and identify all kinds of engineering machinery in real time,so as to achieve the purpose of protection of underground cable.Sound recognition system has the advantages of high efficiency,low complexity and simple collection,and it can also collaborate with video surveillance systems.Therefore,it is widely used.In this topic,the voice recognition technology applied in underground cable monitoring system against external force damage,the system can monitor and identify all kinds of engineering machinery in real-time.The reality requires the recognition system can correctly identify all kinds of engineering machinery signals under different distances,but the sound signals often contain complex noise signals,the collected signals under different distances have different SNRs,this makes the target signals recognition rates are very low.In this paper,statistical pattern recognition technique was applied to road engineering machinery sound signals recognition system by analyzing time domain and frequency spectrum characteristics for the all kinds of engineering machinery signals under different distances.In the paper,the main work are as follows:(1)This paper presents several stable statistical characteristics,such as Short-term Frames Energy Ratio(SFER),Short-term Spectrum Amplitude Ratio(SSAR),Short-term Spectrum Amplitude Ratio Rate(SSARR),Width of Pulse(WoP)and Interval of Pulse(IoP).These characteristics are less affected by the variation of distance.(2)These characteristics were separately used for corresponding classifiers,then formed a complete identification system.The recognition algorithm consisted of appropriate classifiers for the statistical characteristics of engineering machinery.This paper presented concrete steps for each classifier,and showed the effectiveness of the algorithm through simulation results.The average of recognition rates for various target signals in the recognition algorithm reaches more than 85%;(3)In the properties of the tests,the author analyzed the recognition rates under different characteristic parameters.Through the experiment,the threshold value of the characteristic parameters of each classifier was finally selected.(4)The author compared and analyzed the recognition results of Linear Prediction Cepstrum Coefficients,Mel-Frequency Cepstrum Coefficients and proposed.The results show that the algorithm proposed in this paper has a good generalization ability,the statistical characteristics are much more stable,and the recognition rates can reach the application requirements.