Predictive Maintenance Optimization in Aviation Using Machine Learning Models for Reduced Aircraft Turnaround Time and OCC Deployment

Master Thesis

Overview: Predictive maintenance in aviation has become a critical strategy for minimizing unexpected component failures, enhancing operational reliability, and reducing costs. Traditional maintenance relies on fixed intervals, which can lead to both over-maintenance and unforeseen failures. Machine learning (ML) enables the analysis of historical and real-time data to predict failures before they occur, allowing maintenance to be scheduled dynamically and efficiently. This research focuses on developing a predictive maintenance framework that not only forecasts failures but also reduces aircraft turnaround time (TAT) by enabling the Operation Control Center (OCC) to make faster, data-driven maintenance decisions. The final system will be integrated into an OCC GUI to answer real-time operational queries and support day-of-operation decision-making.

 

Objectives:

1. Develop ML models to predict component failures with sufficient lead time to minimize TAT impact.

2. Identify critical operational, environmental, and historical factors contributing to component degradation.

3. Integrate real-time anomaly detection to capture emerging issues during turnaround.

4. Compare predictive maintenance strategies to current practices in terms of cost, TAT, and reliability.

5. Deploy the predictive maintenance model into an OCC GUI for operational use.

Tasks:

Data Collection: Gather open aviation datasets or create synthetic datasets reflecting in-cabin component failures, maintenance logs, turnaround events, and operational parameters.

Data Preprocessing: Clean data, impute missing values, normalize, and conduct exploratory analysis to identify high-impact TAT delay factors.

Feature Engineering: Time-to-failure indicators, Turnaround-phase failure likelihood features, Environmental and operational stress metrics

Model Development: Supervised learning (Random Forest, Gradient Boosting, Neural Networks) for failure prediction, Unsupervised learning (Isolation Forest, Autoencoders) for anomaly detection during turnaround

Simulation & Testing: Use historical and simulated turnaround data for training; validate with recent datasets

Evaluation Metrics: Technical: Precision, Recall, F1-score, RMSE for time-to-failure predictions, Operational: Reduction in TAT delays, % of preemptive maintenance completed within turnaround window

OCC GUI Integration: Build a dashboard with query capability (e.g., "Show high-risk aircraft in the next 24 hrs"), Include visual alerts, risk scores, and recommended actions, Ensure OCC staff can interact with the model for "what-if" scenarios

Time duration: 6 Months

Contact: 

Parth Purohit, parth-yogeshbhai.purohit@tu-braunschweig.de

Dr. Thomas Feuerle,    t.feuerle@tu-braunschweig.de