Environment Identification in Flight using Sparse Approximation of Wing Strain

TL;DR

This study presents a bio-inspired machine learning approach that uses sparse wing strain data and POD features to accurately identify different aerodynamic environments during flight, reducing sensor requirements.

Contribution

The paper introduces a novel sparse approximation method leveraging POD modes for environment classification from wing strain data, inspired by insect sensing mechanisms.

Findings

Robust environment classification with sparse, noisy data

Effective frequency selection for discrimination

Reduced sensor count needed for accurate identification

Abstract

This paper addresses the problem of identifying different flow environments from sparse data collected by wing strain sensors. Insects regularly perform this feat using a sparse ensemble of noisy strain sensors on their wing. First, we obtain strain data from numerical simulation of a Manduca sexta hawkmoth wing undergoing different flow environments. Our data-driven method learns low-dimensional strain features originating from different aerodynamic environments using proper orthogonal decomposition (POD) modes in the frequency domain, and leverages sparse approximation to classify a set of strain frequency signatures using a dictionary of POD modes. This bio-inspired machine learning architecture for dictionary learning and sparse classification permits fewer costly physical strain sensors while being simultaneously robust to sensor noise. A measurement selection algorithm identifies frequencies that best discriminate the different aerodynamic environments in low-rank POD feature space. In this manner, sparse and noisy wing strain data can be exploited to robustly identify different aerodynamic environments encountered in flight, providing insight into the stereotyped placement of neurons that act as strain sensors on a Manduca sexta hawkmoth wing.