Vibration signal analysis for fault detection of combustion engine using neural network

A non linear relationship between an internal combustion engine and its engine parameters such as vibration signals/ exhaust gas is expected to be available. Under various fault conditions, vibration signals were collected using a test-bed to prove this. Fourier transformed vibration signals were mapped to their corresponding faults using a back propagation neural network. The network consists with about 250 input nodes and 150 hidden nodes; resilient back-propagation was used to deal with the complexity created by the high number of nodes. The collected dataset was divided and used for training and testing; and selection combination was changed to check different types of conditions. Using a neural network, creating a relationship between simulated engine faults and their corresponding vibration signals was successful. Although an engine is a complex environment with a lot of unexpected conditions, this result can be used as a start to help predicting engine faults in an efficient and accurate manner. Additional engine characteristics such as exhaust gas/ com port data can also be used to future enhance this fault predicting system.