The manufacturing industry is facing multi-dimensional, ever-growing challenges ranging from the lack of real-time manufacturing resource data, the inability of catching production exceptions, to the occurrence of cascading failures. This paper proposes a network-based cyber-physical production system to model, diagnose and control complex production systems subject to cascading failures. The goal is to study and characterise the evolution of cascading failure mechanisms and further mitigate the vulnerability of the manufacturing system. This is achieved through the deployment and synergistic integration of the Internet of technology with the reliability importance theory. The paper contributes to the network reliability theory and applications by proposing new importance measures and strategies to support the operation of cyber-physical production systems.