Diesel Engine Fault Detection And Diagnosis Based On Rough Set And Kernel Principal Component Analysis

With the development of the level of contemporary industry, diesel engine isincreasingly widely used in variously important industries in the contemporary. However, dueto the complexity of the structure of diesel engine itself and the work surroundings in generalare abominable, which lead to the raising of the incidence and complexity of the fault ofdiesel engine. If these faults are not promptly recognized and handled, it will bring theirreparable economic losses, even cause secure accident. Therefore, it is meaningful todevelop the relative study about the diagnosis of diesel engine’s faults. In recent years, basedon the technology of diesel engine’s faults diagnosis of vibrated signals has rapidly developedin the field of mechanical fault diagnosis, the thesis mainly researches is the methods ofdiesel engine’s fault diagnosis which is based on the vibrated signals. To begin with, using theenergy spectrum of wavelet packet to extract the eigenvalues of vibrated signals. Thenproposed a method which combine the kernel principal component analysis (KPCA) methodand the rough set theory to further optimize the eigenvalues and finish fault monitoring. Itovercomes the shortcomings that the KPCA can’t delete redundant attributes and eigenvaluesbefore dealing with data. Besides, it also effectively decreases the calculating amount ofKPCA and enhances the rate of operation. However, this method only finishes monitoringfaults but not completes the classification of faults. Finally, combined with support vectormachine (SVM) and KPCA to finish engine failure classification, and achieved good results.Prove the validity and applicability of the proposed method, provides some theoreticalsupport for the fault diagnosis of diesel engine work. The specific research contents andconclusions of the thesis are summarized as following:(1) Familiar with the diesel engine fault diagnosis experiment platform which is mainly based on vibration signal acquisition. Features of normal condition and other three conditionsunder three rotating speed have been extracted form four mensuration points in this paper,which provides the raw data for the thesis.(2) By contrasting to analyses the eigenvalues’s sensitivity to the fault of diesel engine,the eigenvalues are produced by four layers of wavelet packet energy spectrum and timedomain analysis. Consequently, found that the former is much more suitable to diagnose thefault of diesel engine.(3) As for the disadvantage that KPCA can’t delete the redundant attribute andeigenvalues before data processing. Firstly, proposed using reduction based on Rough SetTheory to filter characteristic values. Specifically, constructed four condition informationsystem decision tables of the diesel engine’s four measurement points, it adopts the SemiNaive Scaler (SNS) discrete algorithm and equivalent frequency division to deal withcontinuous attribute values of decision tables by discreted. Thus, it effectively avoids theshortcoming that the SNS discrete algorithm could bring too many breakpoints. Then, usingthe attribute reduction algorithm based on Skowron difference matrix which summarized inthis article to reduce the decision tables.(4) It puts forward to the faults detective algorithm which using statistics of KPCA, andmonitors the four points of break-down models by using the KPCA. By simulation can provethat this algorithm could better test whether diesel engine have fault or not. Meanwhile, afterthe comparative analysis, it proves the effectiveness that the thesis used rough set to decreasethe number of eigenvalues before monitoring faults in KPCA.(5) Aiming at the weakness that KPCA can not identify the form of faults, combing theKPCA with supportive vector machine (SVM), on the basis of minimizing inputting data ofSVM, effectively identify the type of diesel engine faults and further reducing the amount ofcalculation. Besides, it realizes the complementary advantages of two algorithms.