Consideration of Variable Operating States in a Data-Based Prognostic Algorithm

A technical system’s availability is vital for economic efficiency. Especially, unscheduled downtime has a major impact on economic efficiency in most branches. Prognostics and health management (PHM) is an emerging discipline which enables downtime scheduling while using as most as possible of a component’s wear margin. Common approaches rely on data-based algorithms that are trained with acquired sensor data and aim to model the degradation behavior in order to predict the degradation curve as well as the remaining useful life (RUL). Most data-based approaches examine a rather static operating condition of a unit under test (UUT), i.e. the load or number of revolutions is not varied while generating run to failure data of a component like bearings. While this assumption can be held for several components, there are those that are operated in a varying state. An aircraft’s control surface actuator is an exemplary component that sees different loads through varying airspeed, weather and the flight mission. Since these variations influence the degradation behavior, they have to be considered in a prognosis algorithm. In this presentation of the underlying paper, a prognosis algorithm is proposed that incorporates a varying operating state, where an operating state is characterized by the parameters which significantly influence the degradation behavior. For a final discussion of the performance improvement, a simulated dataset is generated as input for the proposed and a reference algorithm. It is shown that the proposed algorithm achieves a better prediction performance compared to the reference algorithm that does not incorporate the varying operating state