Diesel Fault Diagnosis Based On Particle Filters And Support Vector Machine

Diesel engine is an important part of the mechanical equipment. And the diesel engine’snormal operating status is essential to the smooth performance of the whole mechanical system. The maximizing of the machinery properties, efficiency, and reliability is the general trendof mechanical equipment. However, diesel engine is exposed to high vulnerability to mechanical problems. Therefore, the condition monitoring and fault diagnosing of diesel engine is crucial. But there are some challenges in diagnosing mechanical problems. For example, the existence of the excessive noise interferes the detection for useful signal, making it difficult to accurately identify machinery fault based on the vibration signal collected. Thus, noise reduction is needed.This thesis provides a solution for the fault diagnosis of diesel engine. First of all, thisthesis investigates the structure of diesel engine, possible causes for the fault, and themechanism of faulting. Secondly, particle filter algorithm is introduced and used in thede-noising process. This paper also proposes the use of local mean decomposition (LMD) todecompose the original signal, and then correlation analysis is conducted between each PFcomponent and the original signal. Through the calculation of correlation coefficient, thosethat show less correlation with the original signal, i.e. illusive component are eliminated. Andthe original signal is reconstructed. Then the autoregressive model, i.e. AR model, isestablished based on the reconstructed signal. The coefficient remains the same in both theAR model and the state equation of particle filter. The noise signal extracted bywavelet-based de-nosing is used in the observation equation. Based on the observationequation and the equation of state, the de-noised signal is obtained after the estimation of theoriginal signal. And the relative energy extracted by wavelet packet energy spectrum is thecharacteristic for fault diagnosis.This study uses support vector machine, i.e. SVM, to conduct fault diagnosis because of its advantage in detecting the small-sampled, non-linear, and high-dimensional data. Theresult shows the solution that this thesis provided can effectively extract the faultcharacteristic and detect machinery fault, thus tackles the challenge in fault diagnosing ofdiesel engine.