The Application of Machine Learning Techniques in Flight Test Applications
Correctly instrumenting test vehicles is a fundamental prerequisite for a successful test campaign. Any delays caused with sensors, wiring or data acquisition equipment on the aircraft either before or during a test flight can potentially add millions to the cost of a test campaign. However, instrumenting a test vehicle is a complicated, labor-intensive and time-consuming process. Invariably the situation is further compounded by the pressure to meet a seemingly immovable deadline. This cocktail, left unchecked, is the ideal environment in which errors and mistakes thrive.
A flight test can be broken into three phases, the pre-flight, in-flight, and post-flight phases. Strategies have been developed over the years to mitigate the various risks in each of these phases. These strategies fall into two broad categories, Processes, and Diagnostic Tools. In the pre-flight phase tools typically include Initiated Built-in Test (IBIT), Continuous Built-in Test (CBIT), Auto-shunt and the visual inspection of data on real-time displays. Processes typically include filling in checklists that incorporate things like the visual inspection of Sensors and wiring. In the in-flight phase, CBIT and monitoring data in real-time are commonly used to detect faults, while in the post-flight phase problems with wiring and sensors are generally detected during the post-processing of the data.
This paper concentrates primarily on the preflight phase. Specifically, the potential application of Machine Learning (ML) techniques is discussed with a view to automating tasks that would otherwise be performed by engineers.