Pre-trained LLM-based remaining useful life prediction of aircraft engines

Although data-driven models have achieved significant success in the domain of Prognostics and Health Management (PHM), many existing approaches remain highly task-specific, relying on specialized datasets and isolated models that lack integration and interoperability. To address these limitations, the machine learning community is increasingly emphasizing a shift towards developing foundation models. The demonstrated success of Large Language Models highlights the potential of foundation models to address a diverse array of tasks through a unified framework. This study investigates the applicability of Large Language Models to the task of Remaining Useful Life prediction. A novel methodology is proposed that adopts the “pre-train–finetune” paradigm commonly used in foundation model development. In particular, the GPT-2 model is employed as the foundational architecture and subsequently fine-tuned for the downstream task of Remaining Useful Life prediction. The efficacy of the presented method is verified by making use of the publicly accessible NASA C-MAPSS turbofan engine dataset. Comparative evaluation indicates that this approach surpasses current state-of-the-art techniques, thus showing the feasibility and potential of foundation models for applications within the PHM domain.