Fault Diagnosis And Health Management Of Diesel Fuel System

The diesel engine is one of the most commonly used power machinery nowadays,and it is widely used in railway traction,oil well excavation,various ships,automobiles and other mechanical fields.Due to the complex structure of the diesel engine,the fault of diesel engine always exhibits the characteristics of complexity and diversity.The fuel system is of paramount importance to the operation of the diesel engine,which actually determines the economy and reliability of the diesel engine to a great extent.Nonetheless,the fault rate of the fuel system is relatively high,and the fuel system faults occupies approximately 27% in the overall faults of the diesel engine system.Therefore,performing a fault diagnosis quickly and effectively plays a pivotal role in the fuel system.In this thesis,a fault diagnosis system for diesel engine fuel system is proposed and developed on the basis of wavelet transform and neural network technology.The specific work is listed as follows:Through literature reviewing and data sorting on fuel system fault,the causes of common fuel system faults and the methods of troubleshooting were summarized and used as the technical database of the health management module.The oil pressure waveform of the diesel high-pressure oil pipe was indirectly obtained by using an external clamp pressure-sensing device to provide data support for the subsequent diagnosis system.The author researched the basic theories and methods of wavelet transform and removed the mechanical noise of the oil pressure waveform with the help of wavelet threshold denoising technology.Meanwhile,through combining the characteristics of the oil pressure waveform and taking full advantage of the multiple experiments,the author also compared the extraction schemes of the two signal features:(1)The oil pressure signal was decomposed into different frequency bands through the technology of wavelet packet frequency band analysis,and the signal energy of each frequency band of the oil pressure signal was statistically analyzed to distinguish the different faults of the fuel system.(2)As the oil pressure waveform contains abundant status information,the waveform width,waveform amplitude,maximum pressure,initial injection pressure and other data in the fuel pressure waveform were further extracted as characteristic parameter.The experiment testified that this method is simple to operate and the extracted fault feature information is abundant,which is much more suitable to be treated as the input vector of the neural network.The author researched and developed fault diagnosis methods for fuel system based on the neural network and utilized the experiments to make comparative studies on the advantages,disadvantages as well as the diagnostic precision of different neural network models.The unsupervised learning of fault data through the SOM neural network can effectively conduct pattern recognition on different faults and perform quick diagnoses.The experimental comparison of SOM and BP neural network models was carried out in this thesis,and the results proclaimed that the BP neural network can meet the precision requirement,whereas the BP neural network is easy to fall into the local optimum.Therefore,the author developed the SOMBP series neural network model for the purpose of fuel system fault diagnosis,at the same period,the relevant experiments also indicated that the SOM-BP series neural network model is able to make up for the shortcomings of the single neural network,and remarkable improvement can also be seen in the diagnosis precision.A fuel system fault diagnosis system was further developed in the research.The common fault causes and solutions of the fuel system,wavelet denoising technology,and the SOM-BP series neural network model were applied to the fault diagnosis system,and the GUI module in MATLAB software was also taken to deal with the completion of the system development.This diesel engine fault diagnosis system can realize the functions of data import,wavelet threshold denoising and fault diagnosis,and additionally establish a health management module to see about the fault causes and the related solutions at any moment.