Explaining how a multi-layer perceptron predicts helicopter airframe load spectra from continuously valued flight parameter data

摘要

This study explains how a multi-layer perceptron predicts helicopter airframe load spectra from continuously valued flight parameter data for 2 example regression outputs - a high strain case and a low strain case. In each case, the input channels are discovered that determine the output activation values. The underlying mechanism that drives the MLP regression output for the two cases is determined.