Rank reduction autoencoder for the generative design of power transformer
Please login to view abstract download link
The design of power transformers is of paramount importance to ensure the distribution of energy, hence their importance in ensuring their reliability over time by estimating their remaining useful life (RUL). According to the international standard norm IEC 60076-7, the RUL of power transformer depends principally on the hot spot temperature (HST). From this, an accurate simulation of the temperature distribution is of capital importance. However, this task is challenging due to the complexity of the temperature profile for different operational parameters. To overcome this difficulty, here we propose the use of a nonlinear model reduction technique called Rank Reduction Autoencoder (RRAE) \cite{mounayer2024rank}, which is used to correctly approximate the complex temperature distribution over a parametric space. This space involves ambient temperature, power consumption and design dependencies as dimensions of the transformer. The proposed technique allows for the accurate reproduction of the temperature field dependency on design and external load conditions, which opens the door to generative design in power transformers.