The Development of Hierarchical Surrogate Models of Low Frequency Electromagnetic Devices for Robust Design Systems

点击翻译按钮获取中文摘要

The increasing use of energy from carbon based fuels by society is having a detrimental effect on the environment in terms of pollution, global warming, etc. Reducing these effects requires a reduction in energy needs even as the demand for devices consuming energy increases. A movement towards electrical machines is seen as one method of reducing some of the effects, resulting in a growing demand for such devices. In fact, the majority of the world's electrical energy is both produced by electromechanical energy conversion (generators producing 98% of the supply) and used by electric motors both industrially and domestically. Electric motors are estimated to consume about 70% of the world's electrical energy (this is 80% in Canada). Reducing energy demands, requires that the efficiencies of these devices are as high as possible and regulations exist, in Europe and North America, on minimum efficiency levels. With the introduction of low cost power electronics controls and new materials, it is possible to design machines to meet the requirements. However, many of the traditional design tools are no longer applicable. Thus there is a need to develop new tools and processes capable of predicting the performance of a device accurately before it is actually constructed. One approach to doing this is to solve for the electromagnetic field in the device and use this to predict the performance. This is computationally expensive and not viable for embedding in an iterative design process. Another approach is to develop a high level model (a surrogate) representing the behavior of the device. While moving from a field solution to a surrogate is relatively easy, moving in the reverse direction is generally difficult due to information loss in surrogate creation. For a design process which is capable of improving quantities such as efficiency, both mappings must exist. The innovation in this proposal is to generate these mappings automatically and then implement the simplest model directly on a core of a Graphics Processing Unit (GPU), thus allowing a low cost massively parallel system to be built. A series of increasingly detailed surrogates will be constructed to allow the design process to operate at several levels and a design system to be created which can generate an optimized device meeting the requirements. This work will be implemented by a team of 5 Master's and 2 Ph.D. students and a Postdoctoral researcher and will build on existing research at McGill through a sequence of short term objectives. The HQP thus trained will be of value both to the research community and industry. The work will benefit Canadian companies, e.g. Magna, TM4, GE, etc., by creating a new range of powerful design tools capable of creating electrical machines which are easier and cheaper to manufacture. It will advance the state-of-the-art in electrical machine design and move Canada into a leading position in this technology in the world.