Master Thesis Fitting Physical Models to the test measurements of the Batteries or Fuel Cells are a powerful tool in capturing their inner characteristics. However, the fidelity of the physical model is highly dependent on the set of physical phenomena coved by mathematical formalism. Differently, testing based on data-driven models, like artificial neural networks (ANN) does not require the use of prior electrochemical knowledge and their inference relies entirely on the data collected during the testing. Although such data-driven models can be very accurate, they also require a large training dataset and do not generalize well outside the training data domain. Development of machine learning methods Implementation of the developed methods into AVL testing pipeline of Batteries and Fuel Cells Overcoming limitations of sparse and out-of-distribution training datasets