With the current trend of materials science towards (semi)-autonomous material development and material databases, high-throughput testing methods are gaining importance. Therefore, fast and reproduceable experimental evaluations provided with an uncertainty estimation and reduced human interactions are needed.

In this work, an approach to automate the evaluation of phase transformation temperatures from dilatometric measurements for carbide free bainitic steels is presented.
The proposed evaluation workflow is supported by machine learning, material models and additionally, other complementary measurements like XRD or metallography can be included. The measurement are checked for artifacts. After this phases and transformation temperatures are assigned via a comparison with stored models. Phase fractions are evaluated with the lever rule and are crosschecked with available data from complementary measurements, older measurements and models. The uncertainty approximation is done via the application of Bayesian inference on the evaluation algorithms. This result is checked by an human und can be stored in an database for future use.
We expect that this approach can be expanded to multi-phase transformations of steels for a fully automatized dilatometer evaluation in the future.