Replacement of In-Orbit Spacecraft Attitude Determination Algorithms with Adaptive Neuro-Fuzzy Inference System via Subtractive Clustering

Abstract

Traditional spacecraft attitude determination algorithms suffer from accuracy versus computational load dilemma. As algorithm complexity increases, its computational load increases and also its accuracy. This limits algorithm applicability in real-time (In-Orbit) environment. In this article, a solution based on Adaptive Neuro-Fuzzy Inference System (ANFIS) has been developed. The number of inputs to ANFIS is 12, which implies an extremely large number of rules of the membership functions to be utilized. Consequently, the proposed solution blows up. Hence, a clustering technique should be utilized in order to minimize the number of rules. Design simulations of the former Egyptian spacecraft EGYPTSAT-1 have been utilized as test case. During the standby operational mode, the spacecraft is assumed to have attitude errors within ±10°. ANFIS is a multi-input single-output algorithm. Therefore, a separate ANFIS is developed for each one of the attitude angles. Each one of them is trained based on two data sets, provided by TRIAD and Q-method simulators. Thus, six ANFISs are developed. The results have shown that the developed ANFISs can work in the stand-by operational mode and provide attitude estimates of ±10°. Moreover, the average execution time for ANFIS is about 40% of the average execution time for the analogous Cascade-Forward Neural Network algorithms (CFNN) and 25% of the Extended Kalman Filter average execution time. The computational overhead of the proposed ANFIS algorithms is constant regardless of the attitude determination or estimation algorithm to be replaced by ANFIS. Consequently, very accurate and sophisticated models could be built in order to provide training data set and then replaced by ANFIS without increasing the ANFIS computational load.