Date on Honors Thesis

Spring 4-29-2022


Computer Science, Mathematics

Examining Committee Member

Dr. Saptarshi Sengupta, Advisor

Examining Committee Member

Dr. Gopal Nath, Committee Member

Examining Committee Member

Dr. Christopher Mecklin, Committee Member


Failure of physical devices can cause inconvenience, loss of money, and sometimes even deaths. To improve the reliability of these devices, we need to know the remaining useful life (RUL) of a device at a given point in time. Data-driven approaches use data from a physical device to build a model that can estimate the RUL. They have shown great performance and are often simpler than traditional model-based approaches. Typical statistical and machine learning approaches are often not suited for sequential data prediction. Recurrent Neural Networks are designed to work with sequential data but suffer from the vanishing gradient problem over time. Therefore, I explore the use of Long Short-Term Memory (LSTM) networks for RUL prediction. I perform two experiments. First, I train bidirectional LSTM networks on the Backblaze hard-disk drive dataset. I achieve an accuracy of 96.4\% on a 60 day time window, state-of-the-art performance. Additionally, I use a unique standardization method that standardizes each hard drive instance independently and explore the benefits and downsides of this approach. Finally, I train LSTM models on the NASA N-CMAPSS dataset to predict aircraft engine remaining useful life. I train models on each of the eight sub-datasets, achieving a RMSE of 6.304 on one of the sub-datasets, the second-best in the current literature. I also compare an LSTM network's performance to the performance of a Random Forest and Temporal Convolutional Neural Network model, demonstrating the LSTM network's superior performance. I find that LSTM networks are capable predictors for device remaining useful life and show a thorough model development process that can be reproduced to develop LSTM models for various RUL prediction tasks. These models will be able to improve the reliability of devices such as aircraft engines and hard-disk drives.