Remaining Useful Life Prediction Using Multiobjective Evolutionary Deep Networks for Electromechanical Actuator

Electromechanical actuators (EMAs), vital in contemporary aircraft for landing gear and servo systems, are pivotal for flight safety. Ensuring their safety involves exploring remaining useful life (RUL) prediction methods. However, current research on EMA RUL prediction methods faces with challenges in balancing accuracy and computational complexity, leading to issues such as low accuracy or high computational complexity. To deal with the above challenges, this article presents a novel method for RUL prediction using multiobjective evolutionary deep networks (MEDNs). Initially, a health indicator (HI) and its corresponding RUL label are selected and constructed. Subsequently, an RUL prediction framework based on deep networks is established. Optimization is implemented using nondominated sorting genetic algorithm-II (NSGA-II), facilitating accurate and efficient RUL prediction for EMA. Experimental results, using the dataset from NASA flyable EMA test stand, demonstrate the superiority of the proposed method over alternatives, with reduced computational complexity, rendering it suitable for predicting the RUL of EMA.