A bearing signal adaptive denoising technique based on manifold learning and genetic algorithm

Signal denoising can be effectively achieved by manifold learning which is a nonlinear technique for reducing dimensionality. However, denoising results based on manifold learning are not only sensitive to relevant parameters, but also there is a strong coupling relationship between relevant parameters. Manifold learning cannot effectively achieve signal denoising based on independent and fixed parameters. To address this problem, this study introduces a denoising technique based on parameter adaptive manifold learning. First initialize parameters embedding dimension, time delay, number of nearest neighbors, and intrinsic dimension. Next, manifold learning is used for noise reduction according to the parameter. Finally, the objective function for parameter updates in the genetic algorithm is the estimated signal-to-noise ratio (SNR) derived from the denoised signal. The effectiveness of the proposed method is confirmed by the examination of the Lorenz signals, the simulated bearing signals, and the real bearing signals. The findings demonstrate that, despite requiring a significant amount of computing time, the proposed method is capable of effectively obtaining the ideal parameters and reducing bearing signal noise.