Condition-based maintenance (CBM) is a crucial strategy in modern industries to optimize equipment performance and prevent failures through real-time condition monitoring. Although CBM offers considerable advantages, it must tackle uncertainties that arise from noise in multiple Condition Monitoring (CM) signals and the latent degradation processes. To manage these uncertainties effectively, a Partially Observed Markov Decision Process (POMDP) offers a robust framework, though it is not scalable with high-dimensional CM data. Addressing this limitation, we propose a scalable and real-time maintenance planning framework based on the POMDP framework that efficiently processes multi-sensor CM signals and provides optimal maintenance decisions under observational uncertainty. Under certain conditions, we theoretically prove the existence of an optimal control-limit policy and introduce an efficient algorithm to overcome the computational challenges. We validate the effectiveness of the model through extensive numerical experiments and demonstrate its scalability in terms of the number of CM signals it can accommodate. Furthermore, we also present a real-time case study with CM signals from turbofan engines.