A Vibration Signal Denoising Method Based on Whale Optimization Algorithm and Batch Normalized Convolutional Neural Network

The random selection of the initial weights of convolutional neural network is prone to cause the network to fall into local optimum, which leads to the difficulty in the extraction of clean signals from noisy vibration signals. To address this problem, a vibration signal denoising method based on whale optimization algorithm and batch normalized convolutional neural network (WOA-BNCNN) is proposed. A batch normalization layer is added after the convolutional neural network to normalize the parameter distribution in the hidden layer. The whale optimization algorithm is applied to optimize the network weight parameters. The amplitude spectrum of the noisy vibration signal is utilized as the training feature and the time-domain waveform of the noisy signal is utilized as the training target to learn the noise characteristics of the noisy vibration signal. The residual learning is adopted as the difference to obtain a clean vibration signal. The superior features of the method lie in that WOA optimizes the network parameters of BNCNN to learn the noise features in the noisy signal, which enhances the denoising ability of the network. The experimental results show that the proposed method improves the signal noise ratio by 2.42dB on average and reduces the mean square error and mean absolute error compared with convolutional neural network, which enhances the denoising ability.