Fault Detection, Isolation and Prognosis of Aerospace Systems Using Adaptive Growing Recurrent Neural Networks

Abstract

Due to the increase in complexity in aerospace systems, developing a diagnosis, prognosis and health monitoring (DPHM) framework is a challenge that must be considered to assure the safety of such systems. This thesis discusses this problem by proposing a novel growing neural network model to automate the process of DPHM for aerospace systems. The model optimizes the architecture of a recurrent neural network and was used to make Remaining Useful Lifetime (RUL) predictions for aircraft engines and detect failure for satellite attitude actuators (Reaction Wheels). It was tested on the CMAPSS and PHM08 aircraft engine datasets simulated by NASA, and it was able to make RUL predictions with root mean square errors as low as 14.31 engine cycles. Another application to test the proposed model was on the Kepler Spacecraft’s reaction wheels from which two have failed. The model detected the failure of the two failed reaction wheels by estimating a Health Index value which indicates the probability of failure of the reaction wheels using the residuals between the speed predictions made by the model and measured speed values. Failure was predicted using the model 105 days and 54 days before it occurred for reaction wheels two and four respectively. Prognostics were also applied on the Kepler Mission reaction wheels and RUL predictions were made with mean absolute errors ranging between 2-13 days depending on how close the reaction wheel is to failure. The proposed algorithm showed results in both applications that could regard it as a promising approach for DPHM models.