A neural network architecture for vibration analysis

The method to estimate physical properties of materials by vibration analysis has been studied and has come into use. In this study, it is examined to use the neural network to analyze the vibration of the material. In general, the vibration consists of periodic fluctuation with long-term fluctuation. Therefore, we adopt the structure of the neural network with sub-networks which estimate those elements. Moreover, since the periodic fluctuation consists of the sinusoidal waveform with different frequencies, and the long-term fluctuation consists of the damping component with different time constants, the sub-networks are assumed to be the structures based on those sums of products. Since this structure is the method to obtain the solution of identification in the form of the variable weights, we call it the answer-in-weights structure. In this study, we propose to use the answer-in-weights neural network for vibration analysis. We simulated the vibration analysis by proposed neural network using actual damping waveform. As the result, we confirmed the effectiveness of the proposed neural network structure.