Machine learning post processing of underwater vehicle pressure sensor array for speed measurement

Abstract

An array of pressure sensors can be used to correct the drift in inertial navigation systems for underwater vehicles (UVs) in absence of other navigation support systems such as acoustic positioning, GPS and Doppler velocity measurements. To date, multiple pressure sensor arrays have been designed, proposed, and tested to prove the concept. However, it has not been researched the inclusion of non-linearities is required in the post-processing. This paper focuses on the use of machine learning as a novel approach to improve the post-processing accuracy, including non-linearities caused by the vehicle acceleration on the estimated speed compared to the linear parametric equation methodology. A series of towing tank experiments have been conducted over an array of pressure sensors located on an UV platform. The results show that pressure measurement array requires the use of non-linear post-processing methodologies as linear methodologies are not able to accurately account for vehicle acceleration effects.