Real-time monitoring and accurate predictions of machine failures are important in maintenance decision-making. Traditional policies using population-specific reliability characteristics cannot represent degradation processes of individual machines, thus result in less accurate predictions of time-to-failure (TTF). Besides, most of the existing maintenance policies focus on a manufacturing system with its fixed system structure, which means the system is designed with limited flexibility. Nowadays, the flexible structure of an adaptive manufacturing system can be adjustable to meet various product types and changeable market demands. In this paper, we try to fill these gaps and develop a prognostic and health management (PHM) framework for manufacturing systems with online sensors and flexible structures. We integrate a Bayesian updating prognostic model using sensor-based degradation information for computing each machine’s TTFs, with an opportunistic maintenance policy handling flexible system structures for optimizing the maintenance scheduling. This enables the dynamic prognosis updating, the notable cost reduction, and the rapid decision making for adaptive manufacturing systems.