Battery Aging has major impact on the design of battery energy storage systems. Therefore, it is important to be able to predict the aging behavior of batteries as accurately as possible. This is achieved by using battery aging models. In order to parameterize these models, time-consuming and cost-intensive experiments are necessary. To increase the efficiency and effectiveness of these experiments, an iterative process for multi-objective optimization of model-based experimental design for battery aging studies is proposed.
The main goal of the presented process is to generate the most accurate models possible in as little time as possible by using the least number of experiments as possible. No prior knowledge is necessary, though beneficial for reducing the number of iterations.

At the beginning, a design space (DS) and target space (TS) are defined, which contain the experimental input variables, the measured quantities of interest (QoI), and their limits.
Assuming there is no knowledge of the considered system, the first iteration of experiments is based on a set of test points (TP) sampled from the DS using latin hypercube sampling (LHS). Later experimental iterations, where information of prior experiments can be used, are design with a optimized experimental design considering not only quality improvement of the model, but also the expenses connected to conducting the required experiments. The Hyper Space Exploration (HSE) provides a framework to quantify and illustrate this multi-objective tradeoff to find the paretooptimal set of experiments efficiently and effectively.
In the next step either a new empirical model is built with the gained experimental results, which is then fitted to the measurements. If a set of models already exists, these are parametrized as well. Or the experimental data is used to train surrogate models.
For model discrimination, the statistical significance is tested before comparing the models using statistical quantities like the Aikaki Information Coefficient (AIC). At the same time, all parameters of the models are tested for their statistical significance and the model is reduced to the most sensitive input variables. The best available model is then used to design the next set of experiments, as described above.
The iterative process can be repeated until a satisfactory battery aging model was generated, or the available time is over.

This workflow is applied to a calendar and cycle aging study of Samsung INR21700-50E Li-Ion cells.