Diagnostic Process Rework for LEAP-1A Fuel System Alerts

In the final presentations at the end of the internship, I ranked first of all eight interns. This was my main piece of work as a diagnostics engineering intern in GE Aerospace fleet support.

The problem: fault alerts for the LEAP-1A fuel nozzle flow divider valve were firing too often without cause, producing unnecessary investigation work and eroding trust in the alert itself.

The approach: I reviewed every triggering case from the previous two years, mapped the decision boundaries between different dispositions, and rebuilt the decision flow chart. Alert thresholds and criteria were recalibrated against trends in onboard snapshot parameters such as exhaust gas temperature across operating conditions. In Dataiku I built replay simulations over historical fleet data to validate the detection-rate against false-alarm-rate trade-off on real cases. Working with product support engineers, I also established a data-driven method for grading fuel system component degradation — separating “replace now” from “keep in service, with an estimate of remaining life” — to avoid unnecessary replacement cost.

The result: a 40% reduction in false alarms, delivered as a revised disposition flow chart and alert logic, along with SOP documentation and a case review library so the same reasoning can be reproduced.

The underlying data, flow charts and deliverables are company property and cannot be shown here.

Images

At the GE Aerospace office during my internship

׋

›

← All work