Online Abnormal Component Locating of Aircraft Fuel System Using Bayesian Method

Abstract

Aircraft systems now become more and more complex as the new technology developed. Since abnormal component (LRU) can trigger a cascade of dangerous accidents, it is very important to develop methods for locating the abnormal component online. However, current fault isolation methods for multiple component systems usually do not consider the practical constraint that only a small part of components can be monitored in real world, which is entirely possible to happen in an aircraft system. This paper develops a probabilistic framework for abnormal component locating of multiple component systems with the consideration of the practical constraint that only a small part of components can be monitored. First of all, a Gaussian Mixture Model (GMM) is introduced to describe the sensor data collected from multiple components. Secondly, based on an estimator of minimum volume set, the criteria to determine whether the monitored samples are abnormal or not is obtained. Lastly, a Bayesian method is used to calculate the posterior probabilities belonging to each possible abnormal component. The proposed method is applied in aircraft fuel system example, which illustrates the efficiency of the proposed method.