Abstract: As complex features often present in both the time and frequency domains of the vibration signals collected from engines, through which it is hard to distinguish the working condition of their valves. Particularly when multiple numbers of valve states are needed to identify, the work will become more difficult to be done. In view of above reason, a smart data mining technique based on statistic rule is developed with the aid of genetic algorithm. After analyzing a large number of vibration signals collected from a diesel engine used for experiments, a few numbers of criteria that are really effective for diagnosing the valve states are found finally. Experimental results show that the proposed technique is actually intelligent and effective for solving the kind of present problem. It paves a new way accessing to the comprehensive utilization of “data mountain” currently formed in production fields.

Key words: Data mining, Diesel engine, Fault diagnosis, Genetic algorithm, Wavelet transform