Publication Date
Fall 2025
Degree Type
Master's Project
Degree Name
Master of Science in Computer Science (MSCS)
Department
Computer Science
First Advisor
Robert K. Chun
Second Advisor
Saptarshi Sengupta
Third Advisor
Navrati Saxena
Keywords
LLM4TS Transformer Architecture; Time-Series Foundation Model; LLM-Based Prognostics; Degradation Modeling; Sensor-Driven Health Monitoring; Remaining Useful Life (RUL)
Abstract
Modern industrial systems including Turbofan Engine(s) are susceptible to failures induced by degradation. Therefore, accurate modeling of component health & Remaining Useful Life (RUL) is vital for reducing maintenance-related downtime. In this work, predictEngineLife, a Regression Framework implemented by leveraging LLM4TS Transformer architecture is introduced, & applicability of this Framework as general-purpose Baseline Model for multivariate forecasting tasks & Remaining Useful Life (RUL) estimation is evaluated. Framework validation is performed on: Electricity Load Diagrams (ECL) 2011–2014 Dataset for large-scale forecasting of hourly electricity demand, & NASA C-MAPSS FD001 sub-dataset for sensor-driven RUL Prediction. On both the dataset(s), Uniform windowing & Z-score normalization is performed to preserve long-range temporal structure. FD001 Inference is compared against TCN, Neural/SVR Regressor(s), & LSTM baselines. Inference results demonstrate that LLM4TS provides consistent end results with recent literature & enables Low-Rank Adaption-mediated fine-tuning on Time-Series (TS) Datasets.
Recommended Citation
Chakravarty, Aniruddha Prabhash, "PREDICTENGINELIFE: RUL ESTIMATION OF C-MAPSS TURBOFAN ENGINE USING LLM4TS & LOW-RANK ADAPTATION" (2025). Master's Projects. 1611.
DOI: https://doi.org/10.31979/etd.zfsb-74yw
https://scholarworks.sjsu.edu/etd_projects/1611
