Intelligent data-driven process design for fatigue-resistant steel components using the example of bainitic microstructure

Project description

The iBain project aimed to establish artificial materials intelligence for the optimization of high-strength steels. Bainite, a specific steel microstructure, had outstanding mechanical properties due to its complex internal structure, which could be deliberately adjusted during production. This internal structure already placed the highest demands on analysis and interpretation. Therefore, automatic pattern recognition and simulations were used to supplement experimental findings. Statistical methods of experimental design (“design of experiments”) helped to plan experiments and simulations and avoid redundancies. Finally, automated control of the workflow and data flow was established. As a result, a customized production process was proposed to produce optimized products with superior properties.

A materials ontology was developed in the project, which enabled the metadata to be read and processed. In addition, data-mining methods were developed to interpret existing findings from experimental data at IEHK RWTH Aachen and Fraunhofer IWM in greater depth. Furthermore, multiscale (partly open-source) simulations were used to reconstruct information that was difficult to access experimentally. An automated, digital workflow was established that would enable industrial users without in-depth expert knowledge to analyze and evaluate their own data in the same way (data sovereignty). For this purpose, an Artificial Material Intelligence System (KMI) was developed that collected, analyzed, and structurally stored the experimental and simulated data in a database. The use of the KMI and the structured digital data made it possible to link the process parameters (temperature, degree of deformation, quenching rate) with the resulting microstructure parameters (size, shape, and composition of austenite grains, ferrite lancets/grains, and carbides). The resulting digital representation, the digital twin, allowed a predictive description of the properties of bainitic steels as a function of the microstructure and processing steps and could be used to predict optimized process parameters and thus improve material qualities.

The main areas of work  were:

• Establishment of a materials ontology for the material class (bainitic) steels, including structure-property relationships, in collaboration with the innovation platform MaterialDigital.
• Development of comprehensive semantic structures for the various data sources, simulation methods, and analysis tools with a view to embedding them in materials ontology and digital workflows.
• Creation of a database for storing and managing existing and future data obtained from experiments and simulations, with functional integration for data mining and machine learning.
• Artificial materials intelligence-supported merging of parameters of bainitic microstructures with the respective process parameters (ICAMS, integration of project partner Georgsmarienhütte).
• Verification and adaptation of a simulation environment for damage in bainitic steels in the HCF and VHCF range ((very) high cycle fatigue).
• Artificial materials intelligence-guided merging of fatigue data (Wöhler curves, service life predictions from numerical models) with the relevant microstructure parameters of bainitic steels.