Adaptive Remaining Useful Life Prediction Based on a System-Level Variable Structure Degradation Model

Accurate prediction of remaining useful life (RUL) for complex systems is often challenged by dynamic and non-stationary degradation behaviors, as well as model mismatches arising from fixed-structure assumptions. To address these challenges, this study proposes an adaptive RUL prediction method based on a system-level variable structure degradation model (SVSDM). In terms of model construction, the initial SVSDM is developed using historical data from multiple sensors, with calibration moments determined by the proposed triggering mechanism. When the calibration is triggered, the model structure is updated accordingly; otherwise, only the parameters of the SVSDM are updated. For RUL prediction, the probability density function (PDF) based on the SVSDM is analytically derived under the first hitting time (FHT) concept. To implement the proposed method, parameter estimation for the SVSDM is performed using maximum likelihood estimation (MLE) algorithm, with dynamic parameter updates facilitated by the Kalman filter (KF). Finally, a case study on furnace wall degradation demonstrates the effectiveness and practical applicability of the proposed method.