Learning Bayesian Networks for Fault Detection

The prompt detection of faults in dynamic systems is essential to prevent dangerous operating conditions and physical breakdown. The early fault detection methods however, may have limitations of physical space, implantation cost, adequate modeling, operation conditions and others. In this work a new fault detection method is presented. This method explores the learning capability of Bayesian networks from measurements of the relevant signals that are present in the dynamic system by the use of a learning algorithm. The results are compared with ones obtained from a fault detection method based on a state observer