MACHINE LEARNING TECHNIQUES

APPLIED TO FLIGHT TEST DATA EVALUATION

This paper proposes an application of machine-learning methods to the analysis of flight test data. A set of training data is used to develop relationships between measurands and generate predicted behavior. These relationships are forecast onto data from the same aircraft model to identify unpredicted measurand behavior. The application of this method may significantly reduce post-test identification time of problematic measurands. Statistical analysis methods are used to determine quality of identified relationships. The importance of a carefully selected training data set and development of robust relationships with low collinearity is emphasized. The developed application demonstrates faster instrumentation failures diagnosis than traditional methods. Areas of continued research include real-time flight safety forecasting, and reduction of required instrumentation. READ NOW