Aircraft sensor estimation for fault tolerant flight control system using fully connected cascade neural network

Flight control systems that are tolerant to failures can increase the endurance of an aircraft in case of a failure. The two major types of failure are sensor and actuator failures. This paper focuses on the failure of the gyro sensors in an aircraft. The neuron by neuron (NBN) learning algorithm, which is an improved version of the Levenberg-Marquardt (LM) algorithm, is combined with the fully connected cascade (FCC) neural network architecture to estimate an aircraft's sensor measurements. Compared to other neural networks and learning algorithms, this combination can produce good sensor estimates with relatively few neurons. The estimators are developed and evaluated using flight data collected from the X-Plane flight simulator. The developed sensor estimators can replicate a sensor's measurements with as little as 2 neurons. The results reflect the combined power of the NBN algorithm and the FCC neural network architecture.