Anomalies recognition in a context aware architecture based on TBM approach

In the sensor-based applications context, sensor reliability is not always taken into account. Due to the uncertain nature of sensors, we must integrate to the problem of belief attached to the sensors data. This paper deals with the dysfunction detection based on a two-level approach. The first level extracts conflict information of the combination of multiple data sources. The second level is based on a prediction-observation mechanism based on symbolic data provided by the first level. With this level we analyse the conflict resulting from a fusion process. It gives information about a sensor failure and the paradigm is based on the Smetspsila transferable belief model (TBM). The second level uses a Markov chain model to describe the normal behaviour of a sensor and thus can detect the abnormal one. Furthermore, the two levels association enables characterising the cause of a failure. Then, we report a case study to show the efficiency of this method when faults appear in highly heterogeneous context.