A Hybrid Transformer-GRU Network for Remain Useful Life Prediction of Aero-Engine Systems

Condition-based maintenance (CBM) is a maintenance method that has promise within the prognostics and health management (PHM) discipline. The determination of the Remaining Useful Life (RUL) of equipment is widely recognized as a complex prognostic task, generating significant concern in both corporate and academic contexts. The precision of basic machine learning and statistical approaches is insufficient for practical applications. Given the limitations of the CNN in effectively capturing implicit temporal information for precise prediction, we suggest the use of a Transformer-GRU Network including an encoder and decoder component. The Transformer encoder use the Multi-head Self-Attention mechanism to effectively capture the important temporal information. Following this, the GRU decoder is used to rebuild the merged sequence order that was first encoded by the Transformer encoder. The experimental results indicate that the proposed technique exhibits higher performance in comparison to the existing state-of-the-art algorithms.