Document Type

Article

Publication Date

2025

Abstract

Objective and scalable pilot assessment is crucial due to rising training demandsand subjective evaluation limitations. Existing data-driven methods often requireextensive data labelling and focus on maneuver classification over executionquality. Addressing the need for interpretable, label-independent proficiencymetrics, this paper introduces an unsupervised pipeline to quantify roll controlsmoothness during turns from X-Plane logs. The methodology features adaptivetrend slicing for segmentation, zero-phase Butterworth filtering (0.6-4.5 Hz) toisolate roll oscillation during turns, and extraction of time domain descriptors(RMS, Peak to Peak, and crest factor). After aggregation and outlier removal, viaisolation forest, K-means clustering identified distinct “smooth” vs “oscillatory”performance groups with high statistical validity. Principal Component Analysis(PCA) further derived a continuous “smoothness index,” capturing 79.6% of thevariance. This study confirms unsupervised oscillation metrics can differentiatecontrol finesse, offering a quantitative foundation for augmenting trainingassessment and enabling future calibration against expert evaluations.

Comments

Presented at the 23rd International Symposium on Aviation Psychology, May 27-30, 2025, Hosted by Oregon State University

Download


Share

COinS