Titanium and its alloys, especially Ti6Al4V, have become essential in the production of aerospace components and biomedical applications due to their combination of low density, high strength, corrosion resistance, and biocompatibility. These properties are largely retained even at higher temperatures. For optimal production, a controlled homogeneous microstructure evolution is necessary, as there is a strong correlation between microstructure and properties in these materials. However, conventional open-die forging and close-die forging are highly manual, and the extreme process conditions, shorter forging time window, and material sensitivity make real-time microstructure control a challenge. This research introduces an integrated, data-driven approach leveraging an advanced measurement system, Finite Element Method (FEM) simulations, and Machine Learning (ML), including surrogate modeling, to enhance the efficiency and microstructure evolution of titanium forging processes.

This study addresses the key challenge of increasing microstructure homogeneity and controlling its evolution, specifically targeting a globular microstructure below β-transus forging, while simultaneously reducing processing time and resource consumption. The approach focuses on two different forging strategies:

• Automated Open-Die Forging: An open-die forging press is equipped with advanced sensors for real-time measurement of billet geometry and temperature. This data is used for FEM simulations to model deformation behaviour, which, combined with laboratory microstructure analysis, trains ML models to predict and optimize microstructure evolution. Surrogate-model-based simulations are implemented to substantially shorten process simulation time, aiming for a semi-automated, data-driven strategy.
• Scaled-Up Equal Channel Angular Pressing (ECAP): As an alternative to open-die forging, the equal channel angular pressing (ECAP) process will be considered, in which the workpiece with a constant cross-section is pressed through an angled channel. The system integrates sensors and uses FEM simulations to guide process design for billets, significantly surpassing current industry limits. AI models of the microstructure evolution will also be implemented for homogeneous microstructure production.

The methodology involves detailed analysis of existing processes, defining requirements for microstructure quality and process control, and collecting and verifying comprehensive material data. Hybrid simulations, integrating surrogate-model-based FEM and experimental results, are developed and optimized to achieve high accuracy (with 10% deviation from FEM). A unified training dataset (with a minimum 100 datapoints per process) is compiled from simulated, process, and laboratory data. Finally, ML models (SVM, Random Forest, Neural Networks) are trained to predict microstructural metrics with a target accuracy of 80%, enabling the implementation of a closed-loop process control system using real-time sensor data. This integrated approach demonstrates the potential for fully automated, efficient, and data-driven forging, representing a significant advance in materials science and industrial process control.