Flight regimes recognition in actual operating conditions: A functional data analysis approach

Helicopters need adequate monitoring to prevent dynamic failures from excessively affecting components’ health status, increase the level of safety, and reduce operative costs. Health and Usage Monitoring Systems have been developed to monitor helicopters during their lifetime in the last few decades. Recent works demonstrated that despite analyzing physical components’ behavior over time, tracking the regimes performed during each flight contributes to estimating the aircraft's health and usage status, paving the way for designing accurate prognostics algorithms. However, today, most regime recognition systems rely on data recorded during certification flights. It follows that the training regimes differ from the ones proposed in the prediction phase, which are acquired during helicopter actual operating conditions. This affects these recognition system performances. Aiming at overcoming this limitation, in this work, we proposed an unsupervised regimes recognition system capable of better handling the actual helicopter usage spectrum. In detail, we proposed a system based on an unsupervised learning paradigm, which leverages a soft-membership classification technique to account even for mixed regimes and transitions. In addition, the system represents data according to functional data analysis theory, which allows for considering the temporal relationship between samples in the classification process, often neglected in state-of-the-art approaches. The proposed system was tested on experimental data, collected by Leonardo Helicopter Division, assessing outstanding capabilities in recognizing correctly standard and mixed regimes and transients. Also, the presented results demonstrate the approach capabilities in paving the way for the definition of new regimes, more consistent with the actual helicopter usage spectrum.