RUL Estimation (Turbofan Engine)

Author: Augustine Osaigbevo

---

Table of Contents

• Project Description
• Repository Structure
• Instructions (How to Run)
• Technical Approach
• Results & Performance
• Challenges & Mitigations
• Next Steps
• Credits

---

Project Description

This project implements a similarity-based Remaining Useful Life (RUL) estimation workflow for turbofan jet engines.

By analysing run-to-failure trajectories from historical engines, the method estimates how much useful life a current engine has left, along with confidence bounds.

Why it matters

• Safety: Early and reliable failure prediction prevents in-flight shutdowns and unscheduled removals, ensuring higher operational safety.
• Cost savings: Predictive maintenance avoids unnecessary preventive part replacements, while also minimising costly unplanned downtime (aircraft-on-ground events).
• Operational efficiency: Condition-based maintenance allows airlines and MROs to schedule interventions just-in-time, improving fleet utilisation.

In short: predictive maintenance = safer flights, lower costs, smarter scheduling.

---

Repository Structure

Similarity_RUL.mlx               # Main MATLAB Live Script: end-to-end workflow
helperLoadData.m                 # Load/convert NASA turbofan data into timetables
helperPlotEnsemble.m             # Visualize run-to-failure ensemble
helperPlotClusters.m             # Visualize operating-regime clusters
helperPlotRULDistribution.m      # Plot RUL PDF, estimate, and confidence bands
README.md

The dataset (NASA C-MAPSS / PHM08) is automatically downloaded and preprocessed by the live script.

---

Instructions (How to Run)

1) Requirements

• MATLAB (R2022b or later recommended)
• Predictive Maintenance Toolbox

2) Clone the repo

git clone https://github.com/austineaero/Remaining-Useful-Life-RUL-Estimation.git
cd Remaining-Useful-Life-RUL-Estimation

3) Open and run

• Open Similarity_RUL.mlx in MATLAB.
• Run all sections sequentially:
  1. Load and preprocess data
  2. Normalise across operating regimes
  3. Construct health indicators
  4. Train residual-similarity RUL model
  5. Evaluate predictions at different observation horizons

---

Technical Approach

1. Data Preparation

• Use NASA turbofan run-to-failure dataset.
• Split into training and validation engine sets.

2. Handling Operating Regimes

• Cluster trajectories by operating condition (e.g., throttle/altitude).
• Apply regime-specific normalisation to remove biases.

3. Feature Screening

• Compute trendability, monotonicity, and prognosability scores.
• Retain only features that evolve consistently with degradation.

4. Health Indicator (HI) Fusion

• Fuse selected sensor features into a single HI.
• Apply offsetting and smoothing so engines align at time zero.

5. Residual-Similarity RUL Estimation

• Fit a polynomial model to each training HI.
• For a test engine, compute residuals against each trained model.
• Use L1 distance to find K most similar histories.
• Estimate RUL distribution from the selected neighbors → provides both point estimate and confidence intervals.

6. Performance Evaluation

• Compare RUL estimates when 50%, 70%, 90% of trajectory observed.
• Assess accuracy, bias, and spread of prediction intervals.

---

Results & Performance

Key Observations

• Early in life: RUL distributions are wide (greater uncertainty).
• Closer to failure: estimates tighten significantly, giving high confidence.
• Method balances accuracy with uncertainty quantification, providing safer decision-making for maintenance.

1. Ensemble of run-to-failure trajectories (health indicator over time)

2. Clustered operating regimes visualisation

3. Trendability/monotonicity/prognosability scores for feature selection

4. KNN vs. RUL (by % life observed)

% Observed	KNN (nearest neighbours in HI space)	RUL PDF (mean + 90% CI)
50% life
70% life
90% life

Observation	Mean Absolute Error (MAE)	90% CI Coverage
50% life	Higher error, wide CI	~95%
70% life	Improved accuracy	~92%
90% life	High accuracy, narrow CI	~90%

--

Challenges & Mitigations

Challenge	Mitigation
Engines operate under multiple regimes → sensor drift	Cluster regimes and normalise within each cluster
Noisy or inconsistent sensor channels	Feature scoring → select only robust indicators
Misaligned degradation progression	Offset and smooth health indicators
Mid-life predictions have wide uncertainty	Provide RUL PDFs with confidence bands, update as new data arrives

---

Next Steps

• Experiment with alternative distance metrics (Dynamic Time Warping, Mahalanobis).
• Explore deep learning-based HI extraction (autoencoders, LSTMs).
• Calibrate confidence intervals via conformal prediction.
• Extend to Python implementation (PyTorch/tslearn) for open-source portability.
• Package workflow as a deployable web service for maintenance engineers.

---

Credits

• MATLAB Predictive Maintenance Toolbox

• NASA Ames PCoE Datasets (C-MAPSS / PHM08 turbofan engine run-to-failure data):
  https://www.nasa.gov/intelligent-systems-division/discovery-and-systems-health/pcoe/pcoe-data-set-repository/