Automated Flight Maneuver Quality Assessment For Aircraft Testing

Abstract

This paper presents a dual-method approach for the automated evaluation of flight maneuver quality, addressing limitations in current flight test practices that rely heavily on subjective visual assessments, manual data quantization, or static thresholds. For steady maneuvers, a statistical formulation is introduced in which flight parameter behavior is modeled as an empirical probability distribution. A new metric, termed redundant cross-entropy, quantifies trim quality by measuring the divergence of the observed data from ideal trimmed behavior which is modeled as a Gaussian distribution. This metric enables consistent comparison across parameters with differing physical units. For transient maneuvers, a fully convolutional network augmented with Squeeze-and-Excitation blocks is proposed to classify maneuver success using real and simulator data from the T625 Gokbey multirole utility helicopter developed by Turkish Aerospace. The architecture supports variable-length time series through masked global pooling and leverages downsampling to extend the model's receptive field without increasing complexity. Evaluation results demonstrate the model's capability to identify high-quality flight segments and reliably assess maneuver outcomes based on predefined success criteria. The proposed framework offers a scalable and interpretable solution to enhance the reliability and efficiency of maneuver evaluation in aircraft flight testing.