Machine Learning Investigation of the Rising Sun Magnetron Design and Operation

Many emerging applications of high-power microwave sources require operation at new and higher frequencies. Computer modeling is the main tool for the design and optimization of these sources, such as magnetrons. However, when designing a new source, the parameter space to scan can be too large to study with traditional techniques. We apply machine learning (ML) (supervised learning in particular) to search this parameter space automatically. We demonstrate this technique by optimizing the operation of a magnetron at 4 GHz, investigating selected parameters tradeoff and predicting operation space characteristics for the discovered device based on a limited amount of measurements. The combination of ML and optimization techniques that we demonstrate here has wide applicability for virtual prototyping and design of problems with sophisticated computer models and allows us to fundamentally reimagine the role of the human in the design workflow.