Fault Diagnosis Using Channelized Encoder-Decoder Frame for Distributed Redundant Inertial Navigation System

Device redundancy measurement systems (DRMS) are gaining popularity in most industrial scenarios. In the area of unmanned aerial vehicles (UAVs), DRMS can provide measurement data from various maneuvering channels and increase the reliability of upper level systems notably. Most fault diagnosis studies for DRMS treat a single device as the basic fault diagnosis unit and assume that the sensitivity centers of multiple devices are the same. However, the distributed redundant inertial navigation systems (DRINS) on UAVs are dominated by three-channel state characteristics, which amplifies the complexity of obtained data and needs more detailed fault diagnosis approaches. What's more, devices in the DRINS have different sensitivity centers, which means much more spatial information needs to be considered for fault diagnosis. To cope with the aforementioned problems, we present a Channelized Encoder-Decoder (CED) framework in this paper to achieve the fault diagnosis for DRINS on UAVs. In CED framework, the basic fault diagnosis unit is refined into each channel, and the data captured by DRINS is cross correlated according to different channels and different measurement devices. The data-driven strategy using deep learning methods is integrated to realize the feature self-extraction of channelized data and the self-determination of fault location. The feasibility of CED is verified through extensive experiments including comparison with other methods. It is validated that, after parameter fine-tuning, CED frame can achieve fairly accurate fault location results when faced with multi-channel and multi-device failures, providing an inspiring way for UAV DRINS fault diagnosis.