Fault diagnosis based on relevance vector machine for fuel regulator of aircraft engine

Fuel regulator is one of the most important components in aircraft engine fuel system and its reliability has great impact on flight safety. Fault diagnosis techniques have been proven as a good solution to improving and guaranteeing the reliability of critical unit. In this paper a fault diagnosis approach for fuel regulator of aircraft engine is developed. In the fault diagnosis strategy, a fuel regulator model is used to predict fuel flow according to engine controller command, an engine inverse model is built to precisely estimate the fuel flowing through the regulator, and the deviations among the regulator model output, engine inverse model output and feedback sensor measurement are used to detect and isolate regulator faults. Due to engine structure complexities and strong nonlinearity, it is very difficult to build exact mathematical engine inverse model. To tackle this problem, an emerging machine leaning technique, relevance vector machine, is adopted to construct the relationship between sensor readings and fuel consumption. Moreover, the proposed method is assessed by hardware-in-the-loop simulation test on an experiment platform. The experimental results verify the satisfying estimation accuracy of RVM-based engine inverse model and show that the developed fault diagnosis method for fuel regulator is effective and feasible, which is promising for fuel system reliability enhancement and aero-engine condition based maintenance.