Fault-detection on an experimental aircraft fuel rig using a Kalman filter-based FDI screen

Reliability is an important issue across industry. This is due to a number of drivers such as the requirement of high safety levels within industries such as aviation, the need for mission success with military equipment, or to avoid monetary losses (due to unplanned outage) within the process and many other industries. The application of fault detection and identification helps to identify the presence of faults to improve mission success or increase up-time of plant equipment. Implementation of such systems can take the form of pattern recognition, statistical and geometric classifiers, soft computing methods or complex model based methods. This study deals with the latter, and focuses on a specific type of model, the Kalman filter. The Kalman filter is an observer which estimates the states of a system, i.e. the physical variables, based upon its current state and knowledge of its inputs. This relies upon the creation of a mathematical model of the system in order to predict the outputs of the system at any given time. Feedback from the plant corrects minor deviation between the system and the Kalman filter model. Comparison between this prediction of outputs and the real output provides the indication of the presence of a fault. On systems with several inputs and outputs banks of these filters can used in order to detect and isolate the various faults that occur in the process and its sensors and actuators. The thesis examines the application of the diagnostic techniques to a laboratory scale aircraft fuel system test-rig. The first stage of the research project required the development of a mathematical model of the fuel rig. Test data acquired by experiment is used to validate the system model against the fuel rig. This nonlinear model is then simplified to create several linear state space models of the fuel rig. These linear models are then used to develop the Kalman filter Fault Detection and Identification (FDI) system by application of appropriate tuning of the Kalman filter gains and careful choice of residual thresholds to determine fault condition boundaries and logic to identify the location of the fault. Additional performance enhancements are also achieved by implementation of statistical evaluation of the residual signal produced and by automatic threshold calculation. The results demonstrate the positive capture of a fault condition and identification of its location in an aircraft fuel system test-rig. The types of fault captured are hard faults such sensor malfunction and actuator failure which provide great deviation of the residual signals and softer faults such as performance degradation and fluid leaks in the tanks and pipes. Faults of a smaller magnitude are captured very well albeit within a larger time range. The performance of the Fault Diagnosis and Identification was further improved by the implementation of statistically evaluating the residual signal and by the development of automatic threshold determination. Identification of the location of the fault is managed by the use of mapping the possible fault permutations and the Kalman filter behaviour, this providing full discrimination between any faults present. Overall the Kalman filter based FDI developed provided positive results in capturing and identifying a system fault on the test-rig.