Abstract

The main objective of this work was to determine the performance of the NLMS and RLS adaptive algorithms when applied to vibration signals in machinery. The adaptive filter theory was applied to vibration signals from a bearing in a rotating machine when contaminated by external random additive noise. The signals were measured by means of two acelerometers and discretely sampled. The analysis of the performance of the NLMS and RLS algorithms was developed both in the time and frequency domains using the Fast Fourier Transform. After the convergence of the adaptive filtering it was observed that the random noise was canceled out and some frequency components could be identified, including the harmonics that are related to bearing geometry and rotation speed. From the results, it was concluded that the RLS algorithm converged in mean square faster than the NLMS, making it possible the application of this technique for machinery condition monitoring to predict damage in machinery components.