Unmanned Systems Technology 009 | Ocean Aero Submaran S10 | Simulation and testing | Farnborough report | 3W-110xi b2 TS HFE FI | USVs | Data storage | Eurosatory/UGS 2016 report

39 where the performance of the control system can be shown to operate in the way it is intended but at the same time testing for faults or problems that can occur. That means simulating the electronics or hardware faults, or causing faults to occur by changing the model while it is running. This is done by changing parameters while the system is running instead of doing a true hardware failure. It can be done as a static event by injecting the fault, or by causing faults to occur while going through a specific performance loop. That is critical for flight testing or unmanned vehicles, to show how the system handles failures while it is operating. Here it is more than just the model; there’s the stimulus that is provided to the system to emulate the activity of the vehicle. In this case the HIL system has to simulate not just a pseudo-manoeuvre but a stimulus of the environment, the surface it is driving on and what it is encountering. One HIL system supplier has developed an animation tool to show what a driver or camera would see out of the front of the vehicle. Putting a camera in front of this screen provides the data for testing adaptive cruise control and lane departure warning systems. Now the challenge is to find ways to inject faults into the radar and Lidar systems within the control loop. FPGAs running high-level models are also used to do that failure insertion and emulate factors such as dirt on a lens, fog or changing lighting conditions. The hardest things are manipulating radar and Lidar because of the high bandwidth of the data stream. Another challenge is providing the comms interfaces at full speed so that the hardware in the loop can be fully exercised. Complete HIL systems are developed by working with a customer directly, depending on the system specification – how to run tests, how to stream data, how to manage the tests and so on. For example, virtual vehicles are built for global car makers, connecting up to 70 ECUs simultaneously in real time to test out braking with engine control and steering. These functions are applied across three ECUs simultaneously, and there is no way to test them without connecting them together with the data traffic from the CAN automotive bus. Instead, a gateway module is used to intercept the comms between the ECUs ‘on the fly’ and modify it to test for different types of faults. It can be carried out in a single cycle (from 1 ms to 250 µs), which is more than sufficient for controlling the mechanical elements in the control loop but not for a video interface or an electric motor drive simulation, so FPGAs are used to obtain the faster loop times and cycle times required. Unmanned Systems Technology | August/September 2016 Simulation and testing | Focus Computational fluid dynamics (CFD) software now allows engineers to model the behaviour of air over the wings of a UAV or water across an AUV without having to build a wind tunnel or large water tank. Wind tunnel technology for one is becoming more expensive compared to simulation, as computer hardware falls in price and the software improves. CFD works well for simple cases, as in a UAV flying at a cruising speed where the airflow is in the linear range, which is around ±10 º for the angle of attack for the wing. This calculation uses the Reynolds average case and gives good correlation with the results from a wind tunnel. The challenge comes in more difficult situations in the other 340 º of operation that can lead to stalling, or in hybrid operation where there is vertical movement converting to horizontal movement. This increases the simulation time by a factor of ten, and the results don’t map as well as the industry would like, as the software tools are not proving to be as accurate, although that is improving. Outside the steady-state analysis, the most popular algorithms are detached eddy simulation, and that is the way the industry is heading right now. The most accurate approach is direct numerical simulation, but that is projected to be 60 or 70 years away before computing power is up to the task. Using computing clusters for CFD simulation is not simple. Some developers have written web applications to upload the data to the cluster, starting with clusters of 48 cores as a minimum. The apps set the accuracy of the result, sort out the grids used for the simulation and split the calculations across the different cores. Have wind tunnels had their day? CFD can help with the design and development of airframes and underwater craft (Courtesy of TotalSim)