An integrated approach of knowledge-driven and neural network for fatigue remaining useful life prediction within small sample conditions

Fatigue remaining useful life (RUL) prediction plays a vital role in improving the operational performance and reducing the failure risk of machinery through effective maintenance management. As a result, it has extracted increasing attention and is furtherly investigated within diverse industrial fields, wherein small sample condition poses to several challenges. Consequently, we propose an integrated fatigue RUL prediction approach based on the fusion of knowledge and neural network under small sample case. Specifically, the crack propagation mechanism is determined referring to correlated domain knowledge, and large scales of fault statistics are obtained via updated model firstly. Furthermore, a fusion approach based on multiple failure mechanisms is devised to generate pseudo-labeled fault data. Then, a three-stage pre-training model based on deep neural network oriented to RUL prediction is designed, wherein both generated and a few of experimental data are utilized fully. The proposed approach is implemented in a practical case study regarding an aircraft fuselage panel and the results demonstrate the enhancement in RUL prediction accuracy.