Faults Detection by Marginalized Particle Filters: Application to a Drilling Process

Real-time faults diagnosis in processes is an essential step to improve their efficiencies. In this paper, by using Marginalized Particle Filters (MPF) for parameter identification, we show the possibility of its exploitation to achieve a Faults Detection and Isolation (FDI) scheme. The idea behind the proposed methodology is motivated by the assumption based on the hypothesis that occurrence of a well-defined fault can bring a process in a well-defined domain. As a result, the whole process model is developed so as to get a part which stands for the faulty case and another that represents the safety one. Thereby, the distribution of these two process states is updated through a Kalman filter and the concerned state is involved by using a particle filter. Experiments and detailed performances analysis using both simulated and real data, obtained from a drilling process used in oilfield industry, indicate that the proposed approach is efficient in terms of tracking ability of parameters change and consequently for FDI task.