Artificial Materials Intelligence For Microstructure-Property Relationship

Irina Roslyakova
The group works on developing an Artificial Materials Intelligence (AMI) modeling strategy and its related applications to analyze and interpret complex materials data. We define AMI as a hybrid physically-based data-driven modeling strategy that uses well-established AI-components such as mathematical algorithms, statistical analysis, machine and deep learning, natural language processing, knowledge-based systems, computer vision, optimization, robotics and others to analyze heterogeneous simulated and experimental materials data on one, several or all modeling scales.

Schematic representation of the Artificial Materials Intelligence (AMI) modeling concept.

Due to the high complexity and heterogeneity of materials data, we focus mainly on three key goals:

• Development and application of data mining and machine learning methods in multi-criteria computer-based materials design
• Material informatics incl. image analysis for data-driven prediction of materials properties of single crystal super alloys
• Management of research data incl. storage of metadata and automated processing of microstructure images using machine learning and deep learning methods

Recently, the main research activity of the group is the development and application of AMI modeling strategy with focus on two materials types:

• description and prediction of creep behavior for single crystal super alloys (SFB TR 103, T2 Transfer Project T2)
• AI-driven modeling and prediction fatigue for bainitic steels (iBain MaterialDigital Project).

To build a so-called the creep indicator model suing AMI modeling strategy, physically-based and data-driven models will be combined to identify statistically sound correlations between materials chemistry, thermodynamic, microstructure and mechanical data of single crystal Ni- and Co-based super alloys. In comparison to the existing purely phenomenological modeling methods, such hybrid modeling strategies allow to identify the influence of individual physical effects from the considered contribution on selected material properties. Moreover, they are frequently required to accelerate the computer-based design of new materials and alloys, for example, the development of rhenium-free Ni-based alloys with the desirable mechanical properties.

Flowchart of AMI modeling strategy for creep indicator model.

Another research activity includes the development of physically-based model that can distinctly account for the effect of thermal vacancies effects on materials properties, such as diffusion and precipitation, separately and be able to identify how they affects different binary and higher order systems. This would provide a consistent treatment of thermal vacancies in CALPHAD databases and their effects on the mechanics and kinetics of materials.

Proposed thermal vacancy model results for BCC-W.

Finally, the group is actively involved in developing modern machine learning based tools and software to contribute to the development of 3rd generation CALPHAD databases. Our latest research activity is the development of an open-source R-based software, named RTDB. The RTDB is an automated software for the fitting of heat capacity data uploaded by the user using either a default or a user-defined CALPHAD model. Based on the fitting results, the software will automatically derive correlated properties such as relative entalphy, entropy and the Gibbs energy function and generate a corresponding TDB file that is ready to be used in any free or commercial computational thermodynamic software available.

News

Masters Course MSS

Studiengang Materialwissenschaft