Improving RUL Estimation via Time-Series Retrieval in Multimodal LLMs

Retrieving historically similar degradation segments from training sets and converting them into visual comparison artifacts improves MLLM performance on the C-MAPSS FD001 benchmark. This retrieval-based grounding provides more accurate Remaining Useful Life (RUL) predictions compared to non-retrieval baselines.