Fault Detection, Isolation, and Recovery techniques for large clusters of Inertial Measurement Units

Abstract

Although Micro Electro-Mechanical Systems (MEMS) Inertial Measurement Units (IMUs) have found widespread use in a variety of navigation applications that require low-cost and/or lightweight systems, their performance is typically not suitable for precision navigation. To address this deficiency, current research is investigating large clusters (15+) of MEMS IMUs with the objective of matching the performance of a single high-quality, monolithic IMU. MEMS IMUs are small enough that a cluster of them is still smaller, less expensive, and lower power than their monolithic counterparts. With such a large cluster of sensors, there is a need for a Fault Detection, Isolation, and Recovery (FDIR) system to identify failed IMUs and prevent them from corrupting the output of the entire cluster. Therefore, the present work develops a FDIR architecture that can identify outlying or erroneous data outputs from large amounts of real-time parallel data, and then prevent erroneous outputs from being incorporated into the state estimation solution. This new work explores FDIR for large IMU clusters using a k-th nearest neighbor algorithm to identify failed IMUs. A Monte Carlo simulation is used to determine the reliability of the technique under random failures of various kinds/sizes. The result of this work is a robust FDIR architecture for use in processing large quantities of redundant IMU measurement information.