Phd Proposal 2016 Title: Data-driven Approaches for Prognosis and Health Monitoring (phm) – from Theory to Application

Background, Context: Optimizing the process availability cannot be reached by reducing the global cost of the process life cycle when maintenance is not predictive. On the one hand, corrective maintenance is commonly used even if it has many limitations due to production stops and downtime, when a failure occurs; on the other hand, preventive maintenance is used to avoid any production stops and to prevent the safety. Neither corrective nor preventive maintenances give good results in industrial processes. Indeed, the costs of the downtime and the equipment replacement can be substantial as well as unnecessary planned maintenances. In order to increase to maintenance performances, the operating conditions have to be taken into account. In last decade, a new field in automatic control, called prognostics, has been developed. It is not just an extension of fault detection and isolation (diagnosis), it is a complete new field. In the diagnosis point of view, the goal is to detect, to isolate and to identify faults when they occur on a process. In the prognostics point of view, the goal is different and consists in the prediction of the process breakdown by taking into account the operating conditions. A first step consists in the health assessment, which leads to a second step, the health monitoring, and to a third step, the prediction of the Remaining Useful Life (RUL). This new tool in automatic control has a large impact on industrial plants; it prevents to large damages and enables to adapt the process maintenance.