Bottom-up capacities inference for health indicator fusion within multi-level industrial systems

In Prognostics and Health Management (PHM) considerations, data fusion technique is a corner step as it enables gathering non homogeneous condition information to provide a synthesizing view on component situations. Nevertheless with regards to complex industrial systems, efficient health monitoring should be addressed not only at component level but also at higher system abstraction levels since the decision making procedure relies on the global industrial system state. To face this issu e, one way consists first in gathering, in the form of vector, the different indicators carrying most important health information related to each system function. Then, a health state has to be computed by fusing each indicator vector taking into account their relevance and their relationships in accordance with the considered system level. It is proposed in this paper to use Choquet integral as an aggregation operator to support this health state development for each function within the system breaking up. It implies to formalize the capacities (Choquet integral parameters) from a bottom-up approach allowing to infer capacities of each system function from those computed at component level. It leads to trace the relevance of health information across the system levels. The feasibility and interest of this fusion approach are shown on an application structured with two system levels.