Research on the impact of activation functions on neural network models for predicting vibration responses

In the research of mechanical environments for aerospace electromechanical systems, the computational method for vibration response is crucial as it affects various aspects of product structure and systems design. In recent years, the intersection between vibration response and deep learning has been a hot research area, yet many challenges remain to be addressed. This article focuses on the research regarding the impact of different activation functions on neural network models for predicting vibration responses. Activation functions are designed for three typical models: multilayer perceptron (MLP), recurrent neural network (RNN), and convolutional neural network (CNN). The study investigates the influence of different activation functions on prediction results under the condition of noisy vibration signals. This article provides a reference for research at the intersection of deep learning and vibration response prediction.