Research On Pipeline Leak Detection Technology Bease On GMM-SVM

Because of its character including high safety, low oil loss, continuous and stable, small occupation area and low cost, pipeline transportation has become the main way for the transportation of oil and gas. However, once the accident of pipeline transportation happens, it is easy to cause serious economic loss and heavy casualties. As a result, it seems to be important that developing a reliable pipeline leak detection technology. Nowadays, there are two types of pipeline leak detection.By consulting plenty of related literature, this paper presents a new pipeline leak detection which can judge the state of pipeline according to the sound around the pipeline. Compared with the former research which focused on the change after leakage, the object of this paper is the pipeline environment. This method can not only detect pipeline leak, but also prevent leakage. Besides, the new method also has some other advantages such as portability, economy and sensitive etc.According to the related theory in speech recognition, this paper presents a leak detection technology based on GMM-SVM model. The new technology has a strong ability of approaching and classification which can overcome the wrong classification caused by the lack of training sample. Besides, according to the comparison between regular acoustic characteristics such as LPC, LPCC, MFCC and PLP, this paper also presents a better-characterized assemblage characteristic which based on static MFCC, delta coefficients of PLP and acceleration coefficients of PLP. Instead of the characteristic samples, our method SVM-classifies the parameters which is the result of GMM-clustering the characteristic samples. And as a result, the interference from the noise is greatly reduced. At last, the experimental results show that the technique proposed in this paper is feasible in the industrial field.