Using Neural Network to Estimate the Success Percent of Spacecraft Active Fault-Tolerant Controller

Abstract

Determining the controller success percent is one of the important issues in spacecraft active fault-tolerant control. The importance of this subject is mainly related to the random and unpredictable nature of faults. On the other hand, since there exists a wide range of faults, various simulations and evaluating controller success percent will require a large amount of time. To resolve this problem, the present paper uses neural network to determine the controller success percent in various fault conditions. First, the neural network is trained and its performance in predicting controller efficiency is verified. Then, considering the high speed of the trained network, a thorough investigation is performed based on a wide range of faults. The obtained results are physically sensible and show that as the fault increases, the probability of controller success will decrease.