What question did this study set out to answer?

This research aims to enhance maintenance scheduling by accurately forecasting the Remaining Useful Life (RUL) of aircraft engines.

May 1, 2026Open Access

Proactive engine maintenance in aviation using data-driven remaining useful life estimation techniques

YBYankai BiShanghai Civil Aviation College CWChao WangShanghai University YCYuan ChiGuangzhou Chemistry (China)

Key Points

This research aims to enhance maintenance scheduling by accurately forecasting the Remaining Useful Life (RUL) of aircraft engines.
Utilized a synthetic dataset to model real engine degradation patterns.
Implemented hybrid techniques combining LGBR, DTR, and RFR with GA and PSO.
Employed Recursive Feature Elimination (RFE) for feature selection.
The LGGA model achieved an R2 of 0.983 and RMSE of 432.77.
Usage_Time_Hours and Vibration_mm_s were identified as highly predictive features for RUL.
The LGGA model demonstrated high predictive accuracy and minimal residual bias in real-time simulations.

Abstract

The precise forecasting of the Remaining Useful Life (RUL, expressed in hours) of aircraft engine-like systems using a synthetic dataset designed to mimic real engine degradation patterns is a key factor in setting up maintenance schedules that are cost-efficient and safety-enhancing. Traditional fixed-interval maintenance often results in inefficient resource allocation and increased risk of unscheduled maintenance events due to the complex, nonlinear nature of engine deterioration. Consequently, a significant portion of maintenance resources is used sub-optimally. Maintenance scheduled at fixed intervals is often inefficient in resource allocation and, at the same time, a source of unplanned failures due to the complex, nonlinear nature of the engine deterioration process. Addressing such issues requires highly powerful prediction models capable of tracking the complex patterns of engine deterioration through multiple sensor readings. This publication describes a hybrid strategy that combines Light Gradient Boosting Regression (LGBR), Decision Tree Regression (DTR), and Random Forest Regression (RFR) with Genetic Algorithm (GA) and Particle Swarm Optimization (PSO). Besides that, Recursive Feature Elimination (RFE) was used to identify features that would contribute most to the model's structure, thereby indirectly facilitating its interpretability of impact. The data used in the experiment show that UsageTimeHours (total accumulated operating hours of the engine) and Vibrationₘmₛ (measuring the vibration level within the engine system) should be the most predictive features of RUL. Among all the models, the Light Gradient Boosting Regression optimized with Genetic Algorithm (LGGA) achieved the best performance, with an R2 of 0. 983 and an RMSE of 432. 77. Furthermore, the simulation of real-time engine health monitoring and predictive maintenance has verified the LGGA model's ability to quickly reach a conclusion, exhibit minimal residual bias, and achieve high predictive accuracy.

Ask AI

Helpful

Bookmark

View Full Paper