Abstract
Predictive maintenance is rapidly becoming a familiar concept in industrial fault detection. The ability to detect early warning signals in systems in the form of small changes in dynamic behavior is essential to anticipate failures. In general, accurate system models are an essential part of residual based fault detection. However, in complex nonlinear systems, the development of accurate models can be very difficult, thus usually other approaches are often selected. As an alternative to the nonlinear analytical models, neural networks have shown significant potential in accurately representing nonlinear systems. In this paper we show how a system identified by a neural network, and a nonlinear observer can be used to detect changes in system dynamics. The neural network structure and identification have a significant impact on the observer performance. Different methods for observer design, and appropriate neural network structures for fault detection are discussed. The experimental section shows the observer implemented on a thermo fluid system. Several faults are introduced, and the observer prediction is compared to actual data.
