Reliable Prediction of Remaining Useful Life for Aircraft Engines: An LSTM-Based Approach with Conservative Loss Function

The remaining useful life (RUL) prediction is crucial for maintaining aircraft engines’ safety and operational efficiency, requiring precise forecasting in their complex and dynamic environments. This paper presents a novel approach for predicting RUL of aircraft engines using an advanced long short-term memory (LSTM) network. The novelty of this approach lies in the integration of a specially designed loss function and the introduction of a new evaluation metric called the Average Safe Underestimation Error (ASUE). The evaluation scheme utilizes the comprehensive C-MAPSS dataset from NASA, which includes detailed run-to-failure operational data of aircraft engines. The LSTM network, specifically adapted for time-series analysis in RUL prediction, underwent a meticulous preprocessing phase to tailor the dataset for effective LSTM analysis. The network was then trained and optimized to predict RUL with high accuracy. A key challenge in RUL prediction is balancing the risk of overestimating RUL, which can lead to unexpected failures, and under predicting it, which can result in unnecessary maintenance costs. To address this, a modified loss function is incorporated in the LSTM model that specifically penalizes overestimation, thus reducing the likelihood of unexpected engine failures. Furthermore, the ASUE metric is introduced to quantitatively evaluate the model’s performance in maintaining a delicate balance between ensuring operational safety and minimizing maintenance costs. The proposed approach, combining the modified loss function and the ASUE metric, steers the LSTM model towards more accurate, reliable, and cost-effective RUL predictions. This enhances safety and operational efficiency in aviation maintenance, contributing significantly to the field of predictive maintenance in the aerospace industry. The study demonstrates that our methodology not only improves the accuracy of RUL predictions but also plays a vital role in aligning them with economic considerations and safety requirements. The code used in this work can be found at https://github.com/AneesPeringal/rul-prediction.git.