Neural Network Modeling of Submarine Shell

Submarine propeller operating in unsteady wake flow often results in dynamic fluctuating force which is transmitted through rotator-raft system and leads to shell structure vibration and structure-borne noise transmission. Even though methods like Flügge’s equations of motion or FEM can be used to model cylinder shell, as for submarine equipped with devices the vibration behaviour of shell will not agree with that acquired from model developed with those methods. In this paper, Neural Network is used to analyse vibration of the submarine shell structure consisting of conical shell, hull and hemisphere shell. Firstly, a finite element model for the combined shell structure is developed for numerical simulations, whose aim is to acquire shell’s vibration information so as to represent the actual output of sensors arranged on submarine shell. Then actuating force is applied on this model based on the prior knowledge of typical blade passing frequency and fluctuating force. The vibration characteristics and responses at the interested positions on the shell are analyzed numerically. These data are used in Neural Network as training and testing data respectively to further develop a non-linear submarine shell model. Vibration responses at the same points acquired from finite element model are compared with the Neural Network model respectively. Results show that Neural Network can be used to model the complex submarine shell structure for effective vibration analysis. Future research will be concentrated on applying Neural Network with the testing data that acquired from actual submarine to develop a controlled plant.