Neural-fuzzy fault diagnosis in a marine propulsion shaft system

The back propagation neural network (BPNN) is applied to the fault diagnosis for the propulsive system of a ship in this paper. There exists a typical vibration signal that can be obtained as a training pattern for each type of fault. These typical vibration signals are collected as training data for the neural network to train the weights of the neural network. The input of the neural network is the characteristic parameters from the power spectrum density (PSD) of the vibration-acceleration signals measured in the shaft bearing system of a ship. By applying a combination of BPNN and fuzzy algorithm methods, an optimal propagation neural network and an optimal learning coefficient for the neural network are obtained. When faults occur, the vibration signals are collected as input of the BPNN. The BPNN will recall and judge what type of fault they belong to. Four testing types, normal condition, loose bearing, unbalanced propeller and loose base were chosen during the experiments. Forty-eight samples were taken in the range 100–500 rpm of the propeller rotational speed. The diagnosis results indicate that a correctness rate of 81.25% on average can be reached and that the best result is a correctness rate of 91.7% in the case of a loose bearing.