MFCC-DSR: A Novel Feature Extraction Approach for Small Leak Identification of Gas Pipes

Gas transportation pipes are important components in delivering gas to another stations. One drawback of these pipes is poor safety and weak maintenance, where leak identification provides a strong support of timely remedy to avoid escalated accidents. There are many techniques show their specific applications, in which acoustic emission (AE) is a particularly useful technique. However, the identification of small leak using AE becomes more difficult according to high noises and wider frequency domain. This paper proposes a MFCC-DSR approach to extract small leak features of gas pipes while the noises in circumstances are strong. In the approach, we explore a discrete stochastic resonance (DSR) model to improve signal noise ratio (SNR) levels of collected acoustic signals. Besides, we introduce a Mel frequency Cepstral coefficients (MFCC)-based scheme to extract features through calculating the gradient of power spectral density of acoustic signals. Experimental results indicate that our approach perform efficiently on small leak identification of gas pipes.