IMU Data for Pitch and Roll Compensation and Vertical Acceleration Estimation: A Machine Learning Approach for Strapdown Gravimeters

Strapdown gravimeters are multi-sensor systems capable of providing all necessary data to reconstruct gravity measurements collected onboard a moving platform (e.g., an airplane, a boat, or a submarine). The primary challenge lies in removing from the gravimeter readings the effects of disturbances, such as unaccounted rotations of the moving platform. This dataset was generated using an experimental setup comprising a three-axial gyroscope and a three-axial accelerometer (i.e., an Inertial Measurement Unit, IMU) installed on a “training platform”. The training platform, equipped with three linear actuators acting as feet, aimed to replicate the operational environment of a moving platform, such as a boat. In this version of the training platform, the IMU was subjected to tilt oscillations (pitch and roll) as well as vertical oscillations, with the latter being monitored by three linear encoders attached to the linear actuators. By analyzing the linear encoders data, one can infer the vertical acceleration and, according to the Equivalence Principle of Einstein, utilize it as a proxy for gravity variations to be measured. The key challenge is to reconstruct from the IMU data the vertical acceleration derived from the linear encoders data. The methodologies developed to address this challenge will play a pivotal role in the advancement of a novel type of strapdown gravimeter known as “Gravimetro Aereo INtelligente” (GAIN), which relies on Machine Learning techniques for data processing.