Fault Diagnosis Method Of Diesel Engine Based On Improved KPCA And KNN Algorithm

Diesel engine is a power device that relies on fuel combustion to release kinetic energy.It is widely used in many fields,such as ships,mines,aviation and so on,which plays an important role in the development of national economy.The diesel engine used by ship is one of diesel engine applications,which is the propulsion power device of the ship.It has the advantages of high thermal efficiency,high economy,easy starting and strong adaptability.However,the operating conditions and internal structure of the marine diesel engine are complex and it has an influence on many other equipments on the ship.Therefore the fault detection of the marine diesel engine is of great significance.The fault detection of the marine diesel engine is to analyze the data collected by the sensor equipment,combining the historical data to identify the operation state.The key problems include the effective information extraction and compression of the historical data signal.The data of marine diesel engine has nonlinear multidimensional characteristics typically and the traditional linear methods have limitation for the nonlinear data.Kernel principal component analysis(KPCA)and kernel local preserving projection(KLPP)methods have good effect on the dimensionality reduction of nonlinear data.The K nearest neighbor(KNN)algorithm can distinguish data classes effectively.In addition,it is combined with least squares support vector machine(LSSVM)for fault detection effectively.In this paper,the AVL BOOST simulation software is used to establish the low speed two stroke marine diesel engine model.Three kinds of fault types and one king of normal type are built to provide the original data for the research work.The diesel engine fault data is nonlinear and it's difficult to extract the characteristic information.So an improved algorithm for global and local structure preserving is proposed to extract the feature of data.KPCA is used to extract features of nonlinear data,only considering global structure.The improved feature extraction algorithm combining KPCA and KLPP considers the global and local feature structure.But it can only complete the effective extraction and dimensionality reduction of the fault data,but cannot complete the classification of the fault.On the basis of above method,an improved K nearest neighbor algorithm is proposed to realize the classification of the fault,and the data aliasing occurs in the projection space,and the boundary data fuzzy classification problem is solved effectively.Besides,a fault detection algorithm based on block KPCA and LSSVM is proposed.By dividing the large sample data into block processing,the detection time is reduced and the fault type is identified by the least squares support vector machine.Finally,the simulation experiments based on AVL BOOST data show that the proposed method can effectively extract the structural information of the data,and improve the accuracy of fault detection.