Satellites are essential in the modern humans’ everyday life. Television, telecommunication, navigation, and more are all made possible by satellites. We all know that these complex structures float around in space, but rarely do we consider, that they can be utilized solely because we are able to send and receive information through them. If the antenna system on a satellite breaks, we have millions of dollars floating around in space and no way of communicating with it – in other words, the satellite becomes worthless. This is why satellite antenna systems are essential.
The last decade has seen an increased interest in utilising space for search and rescue, environmental science missions, internet, telecommunication, and more. This has pushed the requirements for space-born vehicles including their antenna systems – the systems not only have extremely high-performance requirements, but also have to be compact, low-weight, and low-power. The increased requirements lead to increased complexity which in turn leads to an increase in simulation time. This has made it very time-consuming, even impossible in some cases, to optimize the antenna performance or quantify uncertainties in the system due to e.g. manufacturing or assembly errors. Without the possibility to thoroughly optimize or quantify uncertainties, we may end up with sub-optimal antennas which can negatively impact a mission. This industrial-PhD project will try to diminish the time-consumption in the analysis phase giving way to more accurate and robust antenna systems.
The essential question is how the time-consumption can be decreased? Well, the basic idea is simple to understand, but have a wide range of challenges that make it difficult to apply. The idea is to replace the true complex and time-consuming antenna model with a simpler approximate model, which in turn is much faster to evaluate. If this could be done with a high degree of accuracy, then optimisation and uncertainty quantification could be carried out using the surrogate instead. We expect that such a surrogate would facilitate a speed-up of at least 10.
The real challenge lies in the construction of both accurate and robust surrogate models. To learn the true model behavior, it is necessary to evaluate it at a collection of points. Using adaptive sampling, we actively choose the points, which maximize the information retrieved from the true model. This allows us to build an accurate surrogate with as few true model evaluations as possible. Other approaches such as sparse representations will also be examined. The goal is to be able to construct robust and accurate surrogates for a wide range of real-world antenna designs.
Ticra is a Danish company that have worked with antenna design software for decades. This industrial-PhD is carried out as a collaboration between Ticra and DTU.