Issue 45 | Uncrewed Systems Technology Aug/Sept 2022 Tidewie USV Tupan | Performance monitoring | Bayonet 350 | UAVs insight | Xponential 2022 | ULPower UL350i and UL350iHPS | Elroy Air Chaparral | Gimbals | Clogworks Dark Matter

43 Performance monitoring | Focus 300 Mbytes of data. One development team has collected 80,000 hours of data since 2004, which is several gigabytes’ worth that is not yet labelled and is not necessarily structured consistently, as the sensors and parameters have changed over time. The data then requires labelling, ideally in a way that does not need input from engineers, but this unsupervised labelling algorithm is difficult to implement accurately. A key technique is clustering, collecting groups of the same data to provide an indication of whether a data point is inside the acceptable range. The first step has been to develop a small data set that is good enough to identify alarms. An online connection to the GCS allows an operator to upload the log file after the flight to the server, and they could then use ML to produce a report and email a summary of the flight data and any recommendations, for example highlighting vibrations in the actuator on the tail that might indicate that it needs to be replaced or scheduling it for maintenance. That also helps the UAV developer to increase the data set for the ML algorithm, but that requires the removal of sensitive data such as the position of the UAV during a mission. In reality, such performance monitoring systems are a combination of supervised and unsupervised learning. The operator can add manual text, while the developer suggests key words to help the ML system identify what happened. One approach being adopted to assist the development of the ML system is to use synthetic data generated from a simulation, rather than using data from a mission. Development engineers have generated a small data set using synthetic data from a hardware-in-the- loop (HIL) simulation tool that is set up to produce situations that are unlikely to happen. The tool combines a simulation environment that provides labelled synthetic data such as position with data from the sensors in the HIL system. The ML network can then be trained to perform the mapping between the synthetic data and measured flight data, for example. That is interesting, as it could allow UAV operators to obtain parameters for their platforms that at the moment would only be possible with tests in wind tunnels or CFD. To facilitate the process of gathering data for network training, simulation techniques can be used for generating data, using automatic flight control systems coupled with mathematical models of aircraft. The procedure is performed in two phases.  During the training phase, a mathematical model (with random aerodynamic parameters) of an Uncrewed Systems Technology | August/September 2022 Connecting sensors reliably is a key part of performance monitoring. To help achieve that, UAVCAN (Uncomplicated Application-level Vehicular Computing and Networking) is a lightweight peer-to-peer comms protocol being developed for such applications, in vehicles from UAVs to driverless cars. Now called Cyphal, the protocol runs on top of a CAN FD or Ethernet network to connect sensors and processing nodes with fewer than 1000 lines of code. It has been designed for functional safety design with no master node to avoid a single point of failure, and is supported by tools to analyse the system and ensure its correct behaviour. This allows developers to build robust links between components while taking into account fault-tolerance and real-time constraints with a ‘heartbeat’ message to synchronise the data packets. Cyphal has two clearly separated major components: the transport layer that works on top of reliable vehicle networks such as Ethernet or CAN FD, and the transport- agnostic presentation (serialisation) layer based on the Data Structure Description Language. This is similar to the interface description language used in the Robot Operating System, but introduces additional constraints for real-time determinism. The protocol also supports higher-level publish/subscribe protocols (similar to the Distributed Data System specification) that allow nodes to be easily added to a network, and provides a common foundation that different vendors can use to ensure that hardware is interoperable through a generic set of standard application- agnostic communication data types. Unlike some other networking standards, Cyphal does not try to describe the application program interface. UAVCAN/Cyphal The Cyphal protocol can be used with different underlying connection technologies for sensors in vehicles such as UAVs (Courtesy of Cyphal)