A Modified Activation Function for Deep Convolutional Neural Network and Its Application to Condition Monitoring

Convolutional Neural Network (CNN) is a deep learning model which has been an active research topic and applied extensively to vibration data for condition monitoring (CM). In CNN, hyper-parameters, such as activation function, have significant effect on the training task and, consequently, overall performance of network. The existing functions some limitations, vanishing gradient problem, dead neurons, fixed value. order address reported issues, this paper proposes improved function namely (IReLU-Tanh). It adopts advantage ReLU in covering positive region, also by taking properties negative region from Tanh function. Therefore, proposed IReLU-Tanh addresses shortcomings, both gradient, To prove its effectiveness, evaluated based simulated experimental data. Results show that enhances remarkably network two aspects; firstly, task, parameters can reach optimum values with lower errors compared other functions, so learn effectively hidden features. Secondly, it improves accuracy classification yields robust detection diagnosis when against including Tanh, ReLU, LReLU, ELU.