Aircraft Engine RUL Prediction
Turbofan degradation modeling with sequence learning, engineered features and comparative evaluation.
2025
Bi-LSTM features → XGBoost regression · NASA C-MAPSS FD001.
Temporal features for engine degradation
Remaining Useful Life Prediction uses NASA C-MAPSS FD001 sensor sequences to model turbofan degradation. Bi-LSTM representations feed XGBoost regression, connecting temporal context to a boosted prediction stage. Python preprocessing and exploratory correlation analysis organize multivariate inputs.
Estimated remaining cycles and true-versus-predicted views connect output to the degradation trajectory.
Hybrid modeling pipeline
- FD001 engine cycles
- Sensor preparation
- BiLSTM representation
- XGBoost regression
- RUL comparison
The models have different jobs: temporal feature extraction followed by nonlinear regression.
Engineering decisions
- Organize data by engine and cycle so temporal windows retain their meaning.
- Use exploratory correlation analysis to examine redundant sensor information before training.
- Keep representation learning and final regression as explicit modeling stages.
- Inspect predicted-versus-target behavior alongside the error calculation.
Engineering decisions
Correlation analysis removes redundant sensor information before sequence learning. Windows preserve temporal structure, while RUL target construction aligns each sample to the engine’s run-to-failure path. Bi-LSTM features encode both directions within an observed window; XGBoost models nonlinear relations over the extracted representation.
The source records RMSE 14.22 on the first 50 FD001 test samples. That result describes the scoped subset and experiment configuration. Sensor filtering, window/target alignment, temporal features and regression are separate stages, making prediction error interpretable beyond a single numerical badge. Comparative plots expose how the predicted trajectory follows the held-out target.