Remaining Useful Life Prediction for Aircraft Engines using LSTM

TL;DR

This paper demonstrates that LSTM networks significantly improve the accuracy of predicting aircraft engine remaining useful life from sensor time-series data compared to traditional MLP models.

Contribution

It introduces an LSTM-based approach for RUL prediction and compares its performance with MLP, showing superior results on the NASA C-MAPSS dataset.

Findings

LSTM outperforms MLP in RUL prediction accuracy

LSTM effectively captures temporal dependencies in sensor data

The approach enhances aircraft maintenance planning

Abstract

This study uses a Long Short-Term Memory (LSTM) network to predict the remaining useful life (RUL) of jet engines from time-series data, crucial for aircraft maintenance and safety. The LSTM model's performance is compared with a Multilayer Perceptron (MLP) on the C-MAPSS dataset from NASA, which contains jet engine run-to-failure events. The LSTM learns from temporal sequences of sensor data, while the MLP learns from static data snapshots. The LSTM model consistently outperforms the MLP in prediction accuracy, demonstrating its superior ability to capture temporal dependencies in jet engine degradation patterns. The software for this project is in https://github.com/AneesPeringal/rul-prediction.git.