Application of Distributed Computing for High Fidelity AC Machine Analysis

The research investigates the opportunities presented by high performance computing (HPC), both on-premises and cloud based, when applied to electrical machine design and computational electromagnetics tools. Optimisation pipelines involving electromagnetics, thermal and structural analyses require traversal of very large search spaces, which can lead to intractable solution times on typical personal computers. By moving the solution calculations to large distributed systems solutions times fall precipitously. This results in more solutions in the same time; increasing design iteration rate, or it can result in higher fidelity solutions in the same; leading to more optimal machine designs. This concept has been widely adopted by the likes of Ansys, Autodesk, etc.

This research is of particular interest to Motor Design Ltd, the proprietors of Ansys MotorCAD. MotorCAD is a parametric electrical machine design suite for use in end-to-end electrical machine design. At present, the MotorCAD meshing pipeline and FEA solver are tightly coupled to its graphical user interface (GUI). This makes abstracting MotorCAD's FEA solution logic to distributed systems very difficult and carries with it several other issues including licensing and operating system lock-in as MotorCAD's GUI runs only within Windows based operating systems. Part of this research aims to work alongside Motor Design Ltd. to separate the solution logic from the GUI and build a robust cloud-based infrastructure to greatly boost the productivity of MotorCAD users.

The novelty of the research methodology stems from the investigation into the potential benefits of building computational electromagnetics tools designed specifically to be run on large distributed systems, utilising not only the large CPU counts and memory allocation of these systems, but also the caching, monitoring, data storage, and a plethora of other tools made available by web service providers such as AWS, Google Cloud Platform, Azure, etc. A good tool of this kind must optimise for either cost or time at the request of the user. Pricing for cloud compute resources can be used with detailed information about specific compute instances (such as virtual CPU count, memory allocation, geographical location, etc) to optimise for cost with a given time constraint, or conversely optimise for fast time to solution within a specified budget. Cloud systems require a network to communicate, therefore network times and instance instantiation times must also be considered.

Furthermore, this research will investigate the application of distributed computing to additively manufactured (AM) electrical machine design tools. This presents further opportunities within the context of machine design as highly complex stator and winding geometries can be analysed within tractable timeframes compared with the solution times when run on local hardware.

This project falls within the EPSRC "Electrical motors and drives / electromagnetics" research area.