Value of Sensing: a Novel Sensor Selection Metric for Remaining Useful Life Prognostics

The rapid proliferation of sensors and the increasing complexity of industrial systems have enabled high-resolution monitoring of system states. However, not all sensors contribute meaningfully to the accurate prediction of remaining useful life (RUL), and redundant sensing can introduce uncertainty in RUL estimation while incurring unnecessary industrial expenditures. This paper presents a systematic sensor selection strategy that quantifies the Value of Sensing (VoS) based on the variance of the predicted RUL. The proposed VoS metric, analogous to causal inference, serves as a guiding index for optimizing sensor configurations in RUL prediction tasks. The approach employs an iterative sensor elimination process, sequentially removing the sensor with the lowest VoS until model performance deteriorates, thereby identifying the optimal sensor subset. Experimental validation on the FD001 sub-dataset from the C-MAPSS dataset demonstrates a significant reduction in the number of required sensors while maintaining predictive accuracy. The results highlight the effectiveness of the proposed strategy, offering a costefficient and robust framework for industrial prognostics.