Identification of abnormal events by data monitoring: Application to complex systems

Many maintenance actions, such as mechanical, electrical, and hydraulic skills, are mandatory to maintain complex systems in operational conditions. Considerable research has been conducted in these fields to optimize maintenance actions. Most research proposes approaches based on physics: physical model of a specific failure, law of aging, etc. In spite of their performance, these approaches are quite difficult to implement on a complex integrated system. Each field of expertise assesses the good health of a system part using its own experts, its own methods, and, in some cases, its own data. Nevertheless, these fields all make up the same machine, and no interaction between systems is considered. Our study is not based on physical approaches but uses operational data and mathematical tools to diagnose, off-line, the current state of the system. The proposed paper concerns a new concept consisting in characterizing normal system functioning by using data recorded during monitoring. The life profile of this complex system is described by employing all the available data to determine, on the one hand, all normal events and, on the other, to identify abnormal events according to their position compared to the normal envelope defined. The recorded data are then specifically analyzed to characterize the level of criticism of an event considered to be abnormal. This abnormal event could then be assimilated to a global behavioral drift of the studied behavior, which is different to usual behavior. This approach is applied to helicopters by use of all flight recorded data.