A trend in up-to-date developments in service computing focuses on the theme of dynamic composition and optimization of services and its application in service-oriented networked manufacturing (SONM). The paper addresses the particularities of manufacturing resource service composition and optimization (MRSCO) in SONM and proposes a conceptual framework. In this framework, cyber-physical systems (CPS) are incorporated into the manufacturing domain, together with the sensing model and cognitive model that are proposed herein, to integrate the offline resources with online services. Then the QoS models of component manufacturing resource services (MRS), basic constructs and composite MRS are formulated, with the consideration of coexistence of online and offline service phases. Based on the theory of receding horizon control approach and all the aforementioned models, a self-adaptive mechanism is designed in response to the dynamic QoS of MRS and variation of QoS goals, ultimately to guarantee the optimality of composite manufacturing service at runtime. Finally, a prototype platform is developed. The findings suggest constructive ways to model and evaluate MRS in dynamic MRSCO and to transit from a one-off optimization to the feedback-based, closed-loop adaptive MRSCO.