Health Monitoring Algorithm for Turbofan Engine Using Cascade Feedforward Neural Networks

Today, health monitoring systems for turbine engines have become a vital requirement in the aviation industry. In this paper, different fault detection methods of turbine engines are reviewed based on previous research to reveal the importance of the problem and existing challenges. The existing methods use the engine signals for diagnostics, which are heavily affected by operating conditions and disturbances. The faults effect on the performance charts of the F100-PW-220 engine is detected by neural network technique to alleviate the signal variation problem. Some common faults in this type of engine are modeled, including compressor fouling, turbine blade corrosion, and fuel injection problems. The proposed method is effective in a wide range of engine working conditions such as first moments of take-off with afterburner, take-off at 0.1 M, subsonic cruise flight at 0.8 M without afterburner in 10000, 20000, and 40000 feet altitude, supersonic cruise flight at 1.6 M with afterburner in the same altitudes. The cascade neural network with probabilistic transfer functions is used in this paper and shows satisfactory fault detection, while the required training dataset is much less than the previous works. This method facilitates the fast implementation of the system due to the small training dataset and improves the diagnostics accuracy over operational time.