Abnormal Sound Detection by Neural Network in the Diesel Engine

This paper describes the method to detect the abnormal engine condition by means of an algorithm of the neural network, which is modeled upon information processing capability of the operater.The 4-cycle diesel engine is used in this experimental study. As for kinds of engine abnormalities, the leak of the gas from the exhaust pipe and the abnormal combustion depending on F.O. cut are taken as an example.The neural network is composed of the units and the weights which link between units. When the engine is operated by the normal condition, the spectrum data of the radiation sounds from the engine are put in the input layer in the neural network. By learning those data, the weights are formed between units. We judge whether the engine is the normal condigion by using this neural network. Under the normal condition, the correct recognition rate is almost 80% or over both the exhaust leak experiment and the fuel cut experiment. On the other, in case of the abnormal condition, the correct recognition rate is 20% or less in two kinds of experiments. For the results mentioned above, this method using the neural network could satisfactory detect two kinds of abnormal condition.