This paper focus on the diagnosis problem of the misfire and the misfire degree caused at high speed and low loads

compare and analyse the transient speed characteristics under normal and misfire conditions

and find it can effectively improve characteristics sensitivity of engine misfire by shortening the segment length of angular acceleration speed. According to the analysis on the misfire fault diagnosis

it is found that the threshold value method is replaced with the neural network method can effectively diagnose the misfire by employing the relationship between different cylinder characteristic value. Based on the above

a improved misfire diagnosis method of combination of improved segment angular acceleration with the neural network is proposed finally. This method can carry out accurate single cylinder misfire diagnosis over the full speed range

and can determine the misfire degree effectively with the second level diagnosis. And the new approach has a good accuracy rate of misfire fault diagnosis. Meanwhile

it requires less data in the network learning phase and is easy to grasp

which is suitable for on-line diagnosis of engine misfire faults.