Search

Resource failures frequently occur in a smart manufacturing information system (SMIS), which exerts significant impacts on the robustness of the system. From a complex network perspective, this paper develops a fresh methodology for analyzing the robustness of an SMIS suffering from resource failures. First, this methodology divides an SMIS into cyber and physical layers, dissects the resources within these layers and the relationships among these resources. Based on complex network thinking, the methodology then builds a network model incorporating different failure modes and link patterns. Finally, extensive simulations are performed using the case of an appliance manufacturer and one of its suppliers. The results show that an SMIS, along with its cyber layer, exhibits the property of being robust-yet-fragile, and that an assortative link pattern is the optimal link pattern to guarantee robustness for the SMIS under targeted failures.

Fault diagnosis is an important topic both in practice and research. There is intense pressure on industrial systems to continue reducing unscheduled downtime, performance degradation, and safety hazards, which requires detecting and recovering from potential faults as early as possible. From the historical perspective, this paper divides fault diagnosis into previous research and industrial big data era. According to primary drivers, this paper classifies fault diagnosis into knowledge-driven, data-driven, and value-driven methods. Among them, the former two approaches belong to the previous research on fault diagnosis. They mainly depend on expert experience and shallow models to detect and extract failures from relatively small size data. With the continuous exponential growth of data, it is insufficient to mine valuable fault information from massive multi-source heterogeneous data. The huge diagnostic value embodied in industrial big data has driven the emergence of the third category, which belongs to fault diagnosis based on big data. It consists of big data processing and analysis corresponding to high efficiency, cost effectiveness, and generality, which can deal well with problems that previous methods faced. We introduce the concept of a device electrocardiogram from the perspective of applicability to outline the present status of fault diagnosis for big data, and compare it with traditional diagnostic system. We also discuss issues and challenges that need to be further considered. It would be great valuable to integrate or explore more advanced diagnostic methods to handle collected industrial big data and put them into practice to mine the huge hidden diagnostic value.