AI-Based Aircraft Tank Water Level Prediction for Fuel Efficiency
- Customer
- Turkish Technology
- Project manager on the customer side
-
- IT Provider
- TURKISH TECHNOLOGY
- Year of project completion
- 2024
- Project timeline
- August, 2022 - January, 2024
- Project scope
- 560 man-hours
- Goals
- The primary goal of this project was to accurately predict the amount of water to be loaded into aircraft tanks for each flight. This optimization allows for dynamic adjustments to the water load, thereby reducing excess weight and lowering fuel consumption. By ensuring high levels of prediction accuracy, we aimed to reduce waste without impacting passenger experience or in-flight operations. The target was less than 1% error in the prediction model, with a focus on using historical water usage data to optimize the process
- Project Results
- The challenge was to achieve high prediction accuracy (less than 1% error) while ensuring that operational efficiency and passenger satisfaction were maintained. The project’s success demonstrates how AI can contribute to both operational cost reduction and sustainability in aviation.
The uniqueness of the project
This project is unique in its application of AI and machine learning to a specific operational challenge—determining the optimal amount of water to load into aircraft tanks. By analyzing historical water usage data and developing a predictive model, we significantly enhanced operational efficiency. What sets this project apart is the use of AI to directly influence resource optimization in aviation, particularly in reducing fuel consumption through load management- Used software
- Machine learning algorithms for water usage prediction, with the model built as an ensemble model to enhance prediction accuracy. The model was developed, trained, and deployed on the OpenShift Data Science Platform, ensuring scalability and near real-time integration with operational systems. Additionally, a human-in-the-loop approach was implemented, allowing critical oversight and validation during the prediction process. This ensured that the automated system could benefit from human expertise, particularly in complex scenarios where manual adjustments or interventions were needed to further refine predictions and maintain operational safety and reliability.
- Difficulty of implementation
- One of the most critical aspects of this project was ensuring that the machine learning model accurately predicted the amount of water to be loaded into aircraft tanks. The accuracy of these predictions was crucial because even small errors could have a significant impact on fuel consumption and operational efficiency. As a result, the model needed to have an extremely low margin of error, operating in real-time with high precision. Another challenge was transitioning from a manual decision-making process, where human operators determined the water load, to an automated system driven by machine learning. Replacing human judgment with an algorithm introduced complexity, and managing this change required a strong change management strategy to build trust in the new system among management and operational teams. Trustable new system enhance the success of change management.
- Project Description
- This project focused on building an AI-driven machine learning model capable of predicting the exact amount of water that should be loaded into each aircraft’s tank. By analyzing patterns in historical water usage data, the model could dynamically adjust water loads to reduce excess weight and fuel consumption. This optimization led to significant monthly savings of approximately $100k. The model was integrated with the airline’s existing systems, using Kubernetes for scaling based on load demands. The challenge was to achieve high prediction accuracy (less than 1% error) while ensuring that operational efficiency and passenger satisfaction were maintained. The project’s success demonstrates how AI can contribute to both operational cost reduction and sustainability in aviation.
- Project geography
- The project was implemented across Turkish Airlines’ global flight operations, with data being used from multiple international hubs.