A data fusion-based dynamic prognosis model for bearing RUL prediction using vector auto-regression

Abstract Bearings experience fluctuating loads and are crucial to rotating machinery. A sudden bearing failure may result in unexpected machinery failure, which causes an unscheduled stoppage and productivity loss. Predicting remaining useful life (RUL) is essential in Predictive maintenance. In this work, a dynamic prognosis model named reference distributions-vector auto-regression (VAR) method dynamically forecasts the bearing life using the VAR method whenever the degradation trend falls outside chosen confidence intervals computed using reference distributions. Firstly, multi-domain features are computed from bearing vibration signals. The top eight features are chosen by computing the monotonicity of all features. Then, a health indicator is calculated using the locally linear embedding approach by fusing selected features. The health indicator is smoothened using the exponential weighted moving average method. The fault initiation is detected using a reference distribution. Then, the RUL of the bearing is predicted using the proposed method. The efficiency of the proposed approach is demonstrated through multiple accelerated life-test datasets from various sources.