Engine Misfire Diagnosis Based On Probabilistic Neural Network

Vehicles bring convenience to peoples’ many aspects, such as life and occupation and so on, this phenomenon makes the contact of different areas more frequent and boosts integration between the urban and rural districts. The engine of vehicle is the source of vehicle’s impetus, the most critical section of vehicle. The requirement of engine is changing from the initial power accommodation to present stability, reliability, safety, economy and so on. Today, the engine is a complex system that consists of electronic control, machinery, hydraulic press and many sensors, the more and more complexity of the engine’s structure increases the probability of engine’s fault occurrence. In order to monitor the engine’s condition and the automobile exhaust timely, in the last century 80 s, western countries led by America developed on-board diagnostics system to diagnose engine’s misfire fault and so on. At the same time, with the purpose of reducing automobile exhaust of the automobile manufacturers’ vehicles, the related laws and regulations are established, our country made the third, fourth, fifth emission standard of State following the example of emission standard of European Union.Vehicle misfire fault is the phenomenon that the gas mixed of air and fuel could not burn normally in cylinders of engine and the mixed gas without sufficient combustion is discharged to the outside of cylinders. There are so many reasons result in misfire fault: such as spark plug can not ignite the mixed gas, the supply system of fuel fault, air leakage of the cylinders, and so on. Engine misfire fault not only lowers the efficiency of fuel combustion, but also pollutes the atmosphere, and does not fit the emission standard of State. When urgent avoiding cars or other conditions happening, engine misfire fault makes the power of vehicle not enough, this situation could lead to a tragedy of car crash. Take all the mentioned aspects in to consideration, it will be valuable to design an approach to diagnose engine misfire fault.At the platform of physical simulation software, AMESim, four cylinders in line engine idling model and vehicle-mounted engine model are constructed. By setting throttle control signal, communication time, misfire fault injection and other related parameters, 11 kinds conditions of engine operating are simulated. Collect the data of engine rotary velocity and crankshaft angular displacement under different operating conditions for the reason that these two classes of data can intuitively response the operation condition of engine and they are easy to measure at the same time. In the matlab circumstance, data dispose is completed: add the engine crankshaft displacement calculated by remainder arithmetic to the engine rotary velocity by somehow ratio to be the observed data, distribute data to be training samples and testing samples.Base on study of many algorithms and verification of lots of simulation experiments, a probabilistic neural network modified by Principle Component Analysis and Genetic Algorithm is devised. By means of the probabilistic neural network modified by Principle Component Analysis and Genetic Algorithm to diagnose engine misfire fault under the status of engine idling operation when throttle control signal is vibrating, the status of loading engine operation when control signal is vibrating(two engine operating statuses, each status has 11 kinds of different operation condition of engine). The function of Principle Component Analysis is to optimize the structure of the probabilistic neural network’s hidden neurons, by utilizing Principle Component Analysis, elect certain numbers of principle components to earn new training samples and testing samples. Using 11 kinds of new training samples and testing samples to train a probabilistic neural network, the optimistic spread is searched globally by employing the Genetic Algorithm of Genetic Algorithm tool box, gatool. The optimistic spread is the spread that makes the classification accuracy of the probabilistic neural network with new training samples and new testing samples to the highest level. Comparing with probabilistic neural network, general regression neural, radial basis function neural networks, these three neural network, we learn that the probabilistic neural network modified by Principle Component Analysis and Genetic Algorithm has high accuracy of classification, performance of disturbance tolerance is outstanding and some other advantages, it is suitable to diagnose engine misfire fault under the statuses of engine idling operation and loading engine operation.When contrasted with existing methods of engine misfire fault diagnosis, the approach presented in this article has the merits such as collecting the observing data easily, theory and disposal of data are simple, cost performance is high, performance of disturbance tolerance is outstanding, the accuracy of locating the misfire cylinder is high(approximate to 100 percent) and performance of disturbance tolerance is outstanding. It has abroad application prospect and value of realistic popularization.