Critical turbine maintenance: Monitoring and diagnosing planes and power plants in real time

Description

GE produces a third of the world’s power and 60% of its airplane engines—a critical portion of the world’s infrastructure that requires meticulous monitoring of the hundreds of sensors streaming data from each turbine.

The infrastructure and teams involved with processing this data have evolved over the course of decades, incorporating expert knowledge on how sensor variations act as leading indicators for maintenance issues, compliance with regulations and customer agreements, and innovative digital twin models to identify potential issues. These maintenance systems have performed spectacularly, minimizing down time and identifying critical issues early. They have also collected years of hand-labeled data connecting sensor output with downstream impact such as hazardous gas leaks and melting fan blades. With sensor technology steadily increasing the amount of data streaming from these engines, making them harder to analyze, and failure modes becoming increasingly nuanced, GE has embarked on the next chapter of innovation in maintenance, incorporating machine learning.

June Andrews and John Rutherford explain how GE’s monitoring and diagnostics teams released the first real-time ML systems used to determine turbine health into production.

Topics include:

• Understanding new and complex domains
• Navigating ML product design to maximize the probability of a successful release
• Focusing on scalable ML implementations so development efforts in one domain can benefit another
• Safely releasing and support models in critical environments distributed across continents
• How to make critical data products “walk-away safe”
• How to incorporate innovations into a platform for additional turbine applications