An Adaptive LSTM-PR Hybrid Health Status Prognostics Strategy with Balance Between Accuracy and Computational Burden

Aviation contactors are widely utilized in the aircraft electrical power distribution systems (AEPDS), which are the key equipment for the reliability and security of electrical power distribution. The accurate and rapid estimation and prognostics of the health status (HS) for aviation contactors can greatly improve the operational stability of AEPDS and ensure flight safety. Nevertheless, the limited computational resources on board restrict the application of complex methods, which have heavy computational burdens. To obtain the balance between the prognostic accuracy and computational burden of HS prognostics for aviation contactors, an adaptive LSTM-PR hybrid prognostics strategy is proposed. The proposed adaptive hybrid strategy combines the advantages of the high accuracy of the long short-term memory (LSTM) model and the light computational burden of the polynomial regression (PR) method, respectively. Moreover, a novel integrated assessment indicator (IAI) is constructed to comprehensively assess the prognostic performance. Based on the degradation dataset of aviation contactors, the proposed prognostics strategy is validated. The validation results demonstrate that the proposed strategy has a remarkable superiority in terms of integrated prognostic performance. Compared with typical baseline methods, which contain the convolutional neural networks (CNN) and back propagation neural network (BPNN) methods, the proposed strategy enhances the integrated prognostic performance by 91.06% and 74.65%, relatively.