Although Fault Detection and Isolation systems have been widely studied in recent years, it is still a very active research field due to its relevance in industrial production systems. In this paper, a new approach for multiple fault detection by using residual evaluation is proposed. First, an analytical redundancy scheme for residual generation is applied using nonlinear autoregressive networks with exogenousinputs for normal and faulty conditions. Simultaneous fault data is included in the training set in order to ensure multiple fault detection.Then, an adaptive filter considering statistic measures from input is used to increase sensibility and robustness. Filter coefficients are obtained off-line through genetic algorithm optimization. Finally, a neural network classifier is used for fault isolation. The proposed algorithm is tested on a rotary mechatronic test bench for backlash, bearing static friction and transmission faults to show the effectiveness of the proposed detection.
Read full abstract