Identification of NRC Bell 412 Advanced System Research Aircraft Hingeless Rotor System Response via Deep Learning Methodology

The National Research Council of Canada and International Test Pilot School collaborated in the experimentation with Supervised Deep Learning methods for prediction of high-order rotor dynamics and blade loads. Using an intentionally limited flight dataset (431 samples, 6.7 seconds) from the NRC Bell 412 Advanced Systems Research Aircraft (ASRA), an evaluation framework was developed to expose limitations of conventional validation practices. Latin Hypercube Sampling is used to improve training data coverage, restoring temporal generalization and yielding positive (R2) across all eleven rotor quantities, including hub dynamics and blade bending at two spanwise stations. Peak results include beam bending at R2 = 0.94, (CNN, (r=0.97)) and inboard chord bending at R2 = 0.81, (r = 0.90). Ensemble averaging further improves temporal-split performance from R2 = 0.33 to R2 = 0.64 at no additional cost. Results provide guidance for rotor load estimation under data-constrained conditions.