Research On Pipeline Fault Detection Method Based On Multi Source Heterogeneous Data

As one of most widely used energy sources in the world,oil has played a play a very important role in the development of our national economy.However,with the increase of pipeline operation time,pipeline corrosion and other reasons will cause pipeline leakage,which will cause serious pollution to the environment and great loss to the national economy,the detection accuracy of pipeline fault detection method based on single sensor can not meet the requirements of precision and accuracy of fault detection in industrial field.The main reason is that the anti-interference ability of the single sensor is weak.In some extreme cases,the fault detection of the pipe can not be realized completely.Based on the above problems,this paper uses the relevant theory of information fusion,and proposes a pipeline fault detection method based on multi source heterogeneous signals based on sound wave and pressure.The main work of this article is as follows:The first,the wave and the propagation mechanism of the sound wave and pressure signal in the pipeline are analyzed.The phenomenon of wave aliasing in sound and pressure signals is found.In order to fully and effectively show the information contained in sound and pressure signals,A signal decoupling method(ICEEMD-ApEn)based on improved complete ensemble empirical mode decomposition with adaptive noise(ICEEMD)and approximate entropy(ApEn)is proposed.the ICEEMD of signal decoupling,the intrinsic mode function(IMF)of different frequencies is obtained by decoupling the signals by ICEEMD.By calculating the approximate entropy of each IMF component,the physical meaning of each IMF component is judged according to the approximate entropy value.Second,according to the large amount of data in the normal working condition of the pipeline,the leakage of data is less,the shortcomings of the unsupervised fault diagnosis method,the principal component analysis(PCA)and the supervised fault diagnosis method,Fisher discriminant analysis(FDA),are analyzed in pipeline fault diagnosis.Meanwhile,in order to adapt to the characteristics of nonlinear and non-stationary signals of acoustic and pressure signals,the concept of kernel function is introduced.A semi supervised kernel Fisher discriminant analysis(KPFDA)method combining kernel PCA and kernel FDA is proposed.The accuracy of fault detection in non complex conditions is improved.Third,to solve the problem of low accuracy of KPFDA detection in complex conditions.The multi-scale and semi supervised kernel Fisher discriminant analysis(MKPFDA)method is proposed by combining the multi scale decomposition concept of the signal with the KPFDA.The accuracy of pipeline fault detection under complex conditions is improved.Fourth,analysis is the requirement of experiment.A pipeline simulation experimental platform is built,and the actual effect of ICEEMD-Apen decoupling in acoustic and pressure signals of pipeline is verified.The effectiveness of KPFDA and SKPFDA in pipeline fault detection is tested.