Unmanned Systems Technology 027 l Hummingbird XRP l Gimbals l UAVs insight l AUVSI report part 2 l O’Neill Power Systems NorEaster l Kratos Defense ATMA l Performance Monitoring l Kongsberg Maritime Sounder

82 T he ability to monitor the performance of an unmanned system becomes a key design criterion when autonomy is added to it. Taking its remote operator out of the control loop means the system has to make its own decisions about the integrity and performance of its subsystems. The design and implementation of the performance monitoring in such platforms is undergoing a significant shift that is influencing changes in system architectures. Basic telemetry from a system in operation is no longer adequate for diagnosing problems or highlighting potential problems. Built-in self-test and onboard diagnostics (OBD) are used alongside fault-tolerant designs to ensure that component failures are detected early. Machine learning (ML), often using neural networks, is increasingly being used to monitor the performance of components, to monitor the signals from sensors, the vibrations in the airframe or chassis, or the current from the battery pack. While some ML techniques have included sophisticated software such as the Kalman algorithm running on microprocessors, microcontrollers are now adding neural network hardware ML accelerators to support neural networks directly. That is driving a split between centralised monitoring, for example in a domain controller that consolidates the functions of multiple electronic control units, or to have the ML monitoring taking place in smaller microcontrollers local to sensors. A third option is to include the monitoring within the fabric of a system-on-chip (SoC), tapping into the infrastructure used for debugging the performance of the chip during system development. The data can be processed locally in the fabric using a lightweight processing engine, or shipped to the cloud for analysis. This use of the cloud has become increasingly important for monitoring battery packs. There is a growing trend for data from the packs to be monitored remotely, with autonomous vehicles sending the data to the cloud for analysis by sophisticated machine network algorithms. These monitor individual packs, but collating data from all the vehicles in a fleet – or even across an entire range of vehicles – allows more subtle trends to be identified. That raises the question of reliability, however, as cloud connections are vulnerable if a comms link fails. Nick Flaherty explains the impact autonomy is having on monitoring the performance of unmanned systems How’s it going? August/September 2019 | Unmanned Systems Technology A digital twin can be used to monitor the performance of its real- world counterpart by providing more access to the digital model than is possible in the actual vehicle (Courtesy of Siemens)