Predictive maintenance for business aircraft

📈 The Business Challenge

A global business jet manufacturer, needed to modernize how it processed and analyzed flight sensor data to support predictive maintenance, enhance customer service, and enable continuous engineering improvements.

Each new connected aircraft was equipped with over 6,000 streaming sensors and uploading after every flight.

Their data lake had grown to over 100 terabytes of flight data, but extracting value had become increasingly challenging. Key stakeholders — from flight engineers to customer support to field technicians — struggled with slow, costly queries and limited access to actionable insights. They sought a scalable architecture that could power real-time maintenance decisions and long-term ML-driven predictions.

🧩 Our Approach

As the technical lead, I worked closely with the flight analytics team and AWS’s Data Lab team and solution architects to define a cloud-native solution designed for flexibility, performance, and future growth.

Our approach focused on three core use cases:

• Powering ad-hoc engineering queries over massive datasets
• Creating pipelines to support predictive ML models for proactive maintenance
• Enabling real-time maintenance insights to aircraft owners and maintenance teams through a new customer-facing web application

Key architectural decisions included:

• Refactoring the data lake to better align with query patterns
• Partitioning datasets around serial number and flight sessions instead of raw dates
• Introducing Redshift for high-performance analytics on targeted data slices
• Designing batch inference pipelines using SageMaker for scalable ML deployment

🚀 What We Delivered

• A re-architected data pipeline separating ingestion, transformation, and consumption layers
• Session-based partitioning for ML readiness, eliminating repetitive reprocessing
• Purpose-built datasets optimized for Redshift queries and cost-efficient analytics
• A framework for batch ML inference feeding directly into operational dashboards
• Pathways for future migrations to SnowFlake if desired

🎯 The Impact

• Significantly reduced query costs by minimizing unnecessary full-data scans
• Built data marts around typical use cases and users
• Enabled engineers to run previously untenable queries across terabytes of data
• Created a scalable foundation for predictive maintenance models — supporting smarter maintenance decisions and reducing unplanned downtime
• Positioned flight data analytics as a platform for additional use cases (e.g. digital twins, physics-based models, additional maintenance coverage)

👥 Team & Collaboration

• Led a cross-functional team including cloud architects, DevOps, data engineers, data scientists, user experience, front-end developers and flight engineers
• Collaborative design sessions with AWS Data Lab, AWS Solution Architects and SnowFlake
• Ongoing coordination with flight analytics, maintenance, customer support and engineering teams

🛠 Tech & Tools

• Amazon S3 and Glacier
• AWS Simple Notification Service (SNS) and Simple Queue Service (SQS)
• AWS Lambda
• AWS Batch
• AWS EMR (Apache Spark)
• AWS Glue
• Amazon Redshift
• Amazon SageMaker

💡 Key Takeaways

This project demonstrated the importance of aligning data architecture with real-world query patterns — especially when supporting both transactional applications and exploratory analytics. By rethinking partitioning and introducing fit-for-purpose storage layers, we unlocked significant performance and cost efficiencies while laying the groundwork for future AI-driven initiatives.

• Processed more than 22,500 flights, from 150 aircraft
• 4 million lines of code
• 170 experiments
• 3 AI modules covering condition monitoring, diagnostics and prognostics
• Providing early alerting to prevent more than 32% of all aircraft grounding events