Sensor placement optimization for multiple fault detection using bayesian approach

Abstract

Monitoring, diagnosis and prognosis in a complex system required multiple and different type of sensors to extract data form their structures. Sensors measure physical quantity of parameters of various levels of the system for preventing faults of a system. Uncertainties inherent in sensors cause uncertainty issue in data sets. Data extraction of sensors simultaneously brings with overlapping issue in the system. Whereas, current methods are considered that there are non-overlapping in the system or uncertainties of sensors are ignored. However, reducing cost or physical and technological limitations cause to constraint the number of sensors in the systems. The right placement of sensors affects on the reliability and safety of the system. This dissertation presents an application of Bayesian approach on sensor placement optimization that covers overlapping and uncertainties issues. It also recommends the best possibility placement combination of sensors in a system. The Bayesian Network methodology is introduced with likelihood function for on-demand systems. The proposed algorithm generates evidence sets on-demand for overlapping and uncertainty data. The algorithm calculate information matrix for various possible sensor placement that the most expected information gain show the best location of sensors. This approach applies on car engine that has various faults in the performance of engine with the limited number of sensors. Finally, algorithm presents the best possible placement of sensor