Kateryna Morozovska: Applications of Physics-Informed Neural Networks for Heating and Ageing problem in High Voltage Engineering

Abstract:

Power components often face design challenges related to heat distribution and the ageing of their paper insulation. Conventional CFD methods are fairly capable of capturing heat distribution in power transformer windings in static conditions. However, with the introduction of digitalization and a higher share of intermittent electricity generators and intermittent loads, it becomes challenging to solve heat diffusion problems in power transformers and other similar components in real-time as well as achieve good predictability of their thermal behavior. Hence, Physics-Informed Neural Networks is seen as a promising method for learning the physics of the system in real-time and predicting the heat distribution of the power components for further applications in digitalization. On the other hand, the ageing of the power transformer insulation is a problem with many uncertainties, one comes from the lack of date on the insulation ageing and another one is the data quality. In the second part of the talk, we will introduce the application of PINNs to cellulose ageing problems.