A Remaining Useful Life Prediction Model Based on Hybrid Long-Short Sequences for Engines

In transportation fields, safety, efficiency, and reliability of engines are primary concerns. Remaining useful life (RUL) estimation technology is used to assess the current health status and make effective maintenance plans for engines. How to estimate RUL accurately to increasing the reliability and safety of systems is a challenge issue. However, the existing estimation methods focus on single size sequence. To address it, this paper proposes a remaining useful life prediction model based on hybrid long-short sequences for engines. For long sequence, long short-term memory neural network is used to predict RUL by mining smoothed data potential association rules. For short sequence, time window is used to determine the size of short-term sequence, and a direct relation between engine state and RUL is modeled by gradient boosting regression. Finally, back propagation neural network is used to analyze the RUL results of long sequence and short sequence to obtain hybrid sequence prediction result. The proposed model is implemented in a turbofan engine degradation dataset, and the experimental result demonstrate the proposed model is better than other in terms of accuracy.