Machine Learning Approaches for Failures Detection in Inertial Navigation Systems

Inertial Navigation Systems (INS) are essential for the navigation of both surface and underwater crafts, playing a critical role particularly in autonomous and automated control systems. The reliability of these systems is paramount for ensuring the safe maneuvering of such vessels. Monitoring the technical condition of INS is crucial for operational safety. However, traditional diagnostic approaches often fall short in effectively tackling INS challenges due to dynamic uncertainties, external interferences, and susceptibility to informational disruptions. This paper presents a comparative study on the application of various machine learning strategies for diagnosing failures in INS. It specifically examines gyroscope and accelerometer data from surface vessels, employing the Nomoto model to describe their dynamics. The paper suggests methods for feature engineering and evaluation of their significance. Multiple machine learning algorithms were developed, refined, and evaluated to address the diagnostic challenge. An analysis comparing the effectiveness of these models is also included.