Research On Pipeline Fault Feature Extraction Method Based Spectrum Fusion

Pipeline transportation has become an important transport way after highway, railway, water way, aviation. Due to aging plumbing, corrosion and man-made destructive factors, there are lots of pipeline fault, which poses a great threat to human living environment and property. So it is an urgent need for pipeline fault feature extraction and diagnosis. The existing methods based on signal feature domain can extract local variation of signal, but invalid for global feature of complex and time varying fault signal. Therefore, it has important theoretical meaning and application value to research on the pipeline fault extraction based spectrum fusion.In this paper, an improved empirical mode decomposition (EMD) method using second generation wavelets interpolation was proposed based on the analysis of the influence of sampling on the EMD. According to the characteristics of pipeline fault signal, the extraction method of statistical characteristic spectrum in time domain and EMD characteristic spectrum in time-frequency domain was given. On that basis, a pipeline fault feature extraction method based on spectrum fusion was proposed. Finally, as one of pipeline fault recognition methods, multi-classification support vector machine was developed.The experimental results show that the improved EMD method can eliminate undesirable intrinsic mode functions and extract characteristic spectrum in time-frequency domain accurately under low sampling rate. Compared with extraction methods in single domain, the method based on spectrum fusion can get pipeline fault feature more fully and accurately. The multi-SVM classification algorithm which is constructed by fusion feature spectrum can realize pipeline fault recognition, and it also provides an effective way to solve pipeline fault diagnosis problem.