Deep Learning Topologies
for Error Correction in Inertial Navigation Systems
by Benjamin Walters
Heat sensitive components remain by far the largest contributing factors in error propogation in inertial navigation systems. Compounding over time, errors brought about by temperature variations in these components leads to reduced predictive capacity and confounds efforts to produce stable long-term implementations of inertial navigations systems in the field. Traditonal models have sought to correct for these errors by stress-testing entire assemblies under isothermic conditions, however, these efforts lead to dubious corrective capacity and fail to address local temperature variations within a given system. Here we examine a novel approach, using thermal studies to model numerous heat-sensitive instruments experiencing non-linear, multivariate temperature stress, and compare various deep learning topologies against traditional methods for error correction.