Robust Real-Time Thermal Modelling of High-Speed Permanent Magnet Synchronous Machine

由于汽车工业和能源部门的电气化程度不断提高，对电机的需求从未如此之大。因此，降低这些机器的成本和尺寸，同时最大限度地提高其功率能力是至关重要的。实现这些目标的主要机会是通过对关键电机部件（如端绕组和永磁体）进行精确的实时热建模。这样的模型将能够测量电机内困难或难以接近的位置，而不需要昂贵的传感器。因此，可以增加给定机器的功率密度，因为可以减少由于机器内部温度不确定性而存在的大安全裕度。此外，这可以使电动机器小型化，同时仍然满足给定应用所需的额定功率。这个项目的动机主要是为了解决上述实时热模型可以带来的好处；然而，更确切地说，这是由于目前文献中提出的方法缺乏一致性、鲁棒性和实施。对该主题的全面回顾概述了许多不同建模技术的缺点和优点，包括机器学习、降阶热网络、状态观察者和混合方法。虽然没有单一的方法是流行的最爱，大多数方法围绕集中电容建模。此外，在文献中发现的测试是基于严格实验室条件下的数据集，通常具有简单的测试周期。如果将模型应用到现实世界的系统中，即汽车环境中，那么模型的鲁棒性和可行性就会受到质疑。由于许多出版物依赖于预先记录的实验数据集进行机器学习、参数化和测试，这一点得到了进一步加强。最后，这个实验数据集，就像文献中的许多其他数据集一样，使用了相对较低的速度（6000转/分）液冷电机，因此高速电机产生的内部现象可能没有被捕获，其他冷却方法也不清楚。该项目的科学影响将使测试和汽车应用中的电机小型化。由于电机内部温度的不确定性，这些机器目前受到较大的热安全裕度的限制。此外，通过对机器内的温度进行建模，可以避免在难以接近的位置（例如转子磁铁）安装昂贵的传感器，从而降低试验台系统的成本。最后，该项目旨在增加具有不同几何形状和冷却系统的建模机器的灵活性，减少与参数化模型相关的成本和时间，同时提供可靠和准确的结果。

As a result of increased electrification within the automotive industry and energy sector, the demands from electric machines have never been greater. Therefore, reducing the cost and size of these machines whilst maximising their power capabilities is crucial. A prime opportunity to achieve these targets is through accurate real-time thermal modelling of key motor components, such as the end windings and permanent magnets. Such a model would enable the measurement of difficult or inaccessible locations within the motor, without the need for expensive sensors. Therefore, the power density of a given machine could be increased as the large safety margin, which exists due to temperature uncertainties within the machine, could be reduced. Additionally, this could enable electric machines to be downsized, whilst still meeting the required power ratings for a given application. The motivation for this project is broadly to address the aforementioned benefits a real-time thermal model could enable; however, it is more precisely motivated by the current lack of agreement, robustness, and implementation of methods currently proposed within the literature. Comprehensive reviews of the topic outline drawbacks and benefits to many different modelling techniques, including machine learning, reduced order thermal networks, state observers, and hybrid approaches. Although no single method is yet to prevail as a favourite, most methods revolve around lumped capacitance modelling. Furthermore, testing found within the literature is based on datasets in strict laboratory conditions, often with simplistic test cycles. This brings into question the models' robustness and feasibility if implemented onto a real-world system, namely in automotive contexts. This is further reinforced as many publications rely on a pre-recorded experimental dataset for machine learning, parametrisation, and testing. Finally, this experimental dataset, much like many others in the literature, uses a relatively low speed (6000 rpm) liquid cooled motor, hence internal phenomena resulting from high-speed motors may have not been captured and other cooling methods are not understood. The scientific impact of this project will be to enable downsizing of electric machines in testing and automotive applications. These machines are currently restricted by large thermal safety margins due to temperature uncertainties within the motor. Additionally, by modelling the temperature within the machine, the cost of test bed systems can be reduced by circumventing the requirement for expensive sensors in poorly accessible locations (e.g., rotor magnets). Finally, this project seeks to increase the flexibility of modelling machines with differing geometry and cooling systems, reducing the cost and time associated with parameterising the model, whilst delivering reliable and accurate results.