Fault Diagnosis Of Diesel Engine Fuel System Based On Neural Network Research And Realization

Diesel engines are presently one of the most prevalent-applied dynamic facilities and they have been widely used in oil well, power generating electricity, railway traction, project machinery, all kinds of shipping and vehicles and so on. The diesel engine as the representative of back and forth machine, it is difficult to diagnose the faults by using traditional means be causing of its complexity and multiplicity, and it also can not satisfy our demands. According to the statistics as the correlative datum indicate, approximately 27% of the diesel engine malfunctions are caused by the faults of the oil systems. However, the oil system is most important part of the engine and its condition directly affects the normal running of the diesel engine. In other words, the malfunction of this part would apparently deteriorate the working performance of the whole engine, such as fuel-combusting, the reduction of power, economical and so on. Therefore, faults diagnosis in time is of great importance.The pressure wave in high-pressure fuel tube contains most state information and could be suitable for recognizing various malfunction information sensitively. Just by analyzing the fluctuant of the pressure waveforms in the high-pressure fuel pipe, the faults are diagnosed. With the characteristics of the pressure waves, 8 parameters are extracted, i.e. the maximum pressure, the starting injection pressure, the base pressure, the width of the aftereffect wave, the maximum aftereffect pressure, the pressure width, the area of pressure, the range of pressure, the rising edge width of the pressure. These features reflected the working condition of the fuel oil system directly.The BP neutral network is a type of large-scale parallelism nonlinear system, with the strong ability of associative learning, self-organizing, self-adaptation and nonlinear dynamic calculation, all of which make it more capable of training the malfunction parameters and diagnosing the system failures. In this thesis, BP neural network wildly applied nowadays in faults diagnosis field is studied. And the thesis introduced the model structure of the three layers BP network, choose of the parameter, the training way, and some improved approaches for BP neutral network. The features were applied as inputs of the neutral network and the diagnosis results as the outputs, and the results were pretty good.However, considering the BP network easily getting into the minimum point and its slow convergence rate, the thesis imports SOFM network into the faults diagnosis of the diesel engine, which is a method without teaching or training samples, and can do clustering reorganization in a high speed. The simulation indicates that with SOFM network the speed is high and the diagnosis error is very small.At last, the improved BP neural network is compared with the SOFM network. The result indicates that a high discrimination can be obtained by artificial neural network in diesel engine faults diagnosis. SOFM network has no use for sample training and learning, and it can do clustering reorganization directly in a high speed, and it also can hurdle the disadvantages, fuzzy output caused by lack of sample organizing with the BP network. By comparison and analysis, it is efficient and feasible to apply the neural networks into the fault diagnosis for diesel engines. The application foreground of neural networks will be very capacious.