Diagnosis On Misfire Fault Of Diesel Engine Based On Segment Angular Acceleration And Neural Networks

Diesel engines are often used in large engineering machinery,and it is prone to failure because of its bad working environment.At the same time,diesel engines fault often cause huge economic losses and casualties.So it is very important for diesel engine to have a good security assurance.What's more,recently,people pay more attention to environmental conservation,which gives new goals to engines emission performance.State? emission standard stipulates that all cars should be equipped OBD(On-Board Diagnostics)to monitor the engine operation,and reduce emissions.Misfire is commonly referred to the phenomenon that the combustion cannot be ignited in the cylinder.It is one of the frequent faults of diesel engines and caused by the failure of the fuel system and insufficient of air intake.Diesel engine misfire can influent the performance of diesel engines,and deteriorate emission,even damage the tail gas treatment device.However,sometimes misfire does not affect the operation of the engine,and it can not be distinguished easily from appearance.Therefore,misfire fault is often ignored by people,which is also an important factor of that misfire is a great danger to engines.Based on a careful analysis of mechanism of misfire fault and the defects of OBD on misfire diagnosis,the threshold method of OBD is changed into the neural network method,and a new misfire diagnosis method based on segment angular acceleration and BP neural network is proposed.In order to verify the effectiveness of the method,an experiment was carried out on a 6102 diesel engine.A neural network fault diagnosis model was set up,which realizes the diagnostic test.And different degrees misfire,a complete misfire of a single cylinder and a two cylinder misfire at the same time are considered.The result shows that the method proposed can realize the diagnosis of misfire,and the accuracy is high.Even in the case of high speed and light load conditions,the method also has a good performance.It is worth mentioning that,in the learning phase,the method only needs to select a few representative points to study because of the strong generalization ability of neural network,and less workload compared with OBD.The latter requires a lot of database to accomplish the learning phase.What's more,the phenomenon of local optimization in network training is also explored and genetic algorithm which has the capability of global search is applied to optimize the initial weights of neural network and good result is achieved.