Health index-based remaining useful life prediction using functional principal component analysis with multivariate sensor data

Accurate prediction of remaining useful life (RUL) is crucial for effective prognostics of industrial devices and health management. With the development of sensor technology, multiple sensor data points can be collected during the device monitoring process. To fully exploit the information from different sensors, various methods have been proposed in the literature to construct a health index (HI) by combining multivariate sensor data to predict RUL. However, most of these methods assume that the sensors are independent. In reality, there are correlations between sensor signals due to shared degradation processes, common environmental factors, or sensor interactions. Ignoring these correlations when constructing the HI can lead to inaccurate RUL predictions. To address this issue, this paper proposes a novel approach that utilizes functional principal component analysis (FPCA) to integrate multivariate sensor data while considering sensor correlations. Compared with other data fusion methods, FPCA can extract orthogonal principal component functions from multiple sensors, effectively avoiding multiple correlations between sensor signals. Furthermore, to correct the systematic bias caused by traditional models that primarily capture global trends, we introduce a residual correction module based on random forest (RF) to refine the RUL prediction. Simulation studies and two real-world case studies demonstrate the superiority of the proposed approach in terms of both accuracy and robustness.