Adaptive prognosis of bearing degradation based on wavelet decomposition assisted ARMA model

摘要

This paper proposes a prognostic model using wavelet decomposition for noise cancellation in the measured vibrational signal of the bearing. An adaptive auto-regressive and moving-average (ARMA) with Recursive Least-Square (RLS) algorithm is implemented for root mean square (RMS) value of the vibrational signal of the bearing. The RLS updates the ARMA model coefficients based on the new signal available during the online monitoring to enhance prediction accuracy. In addition, the implementation of RLS does not require any knowledge of the ARMA model coefficients. This advantage of the model eases the computation of the ARMA model coefficients in comparison with the old method, the Yule-Walker equation. To demonstrate the applicability of this adaptive model, we tested the proposed model using the experimental data of degradation trend of the bearing. The maximum absolute error (MAE) is 0.56 of the RMS value with a mean square error (MSE) of 0.0013 and a mean average percentage error (MAPE) of 1.03% The error can be further reduced by increasing the iteration of the RLS algorithm to converge to the steady state error.