The unexpected failure of modern industrial systems is often managed using data-driven predictive maintenance (PdM) tools that continuously monitor a system's health condition (HC) through raw sensor data to predict impending malfunction. However, research demonstrates that data-driven PdM tools perform poorly when applied to raw multi-sensor data, as it is not guaranteed that all sensor data describe a system's health condition. As systems become more complex and require multi-sensor measurements, to perform accurate data-driven PdM, an accurate health condition representation of a system's life cycle data is required. This work introduces a Kullback-Leibler divergence (KLD) based health indicator (HI) constructor for multi-sensor systems. This method applies information entropy to construct a HI representation that describes the occurrence of faults and their influences during a system's life cycle. Additionally, the proposed method conducts feature selection to expose and remove sensors that do not capture information related to a system's HC. The utility of the proposed method is tested on the Commercial Modular Aero-Propulsion System Simulation turbofan engine data and the OEM Group's Cintillio SAT Batch Spray semiconductor manufacturing equipment data.