About this Abstract |
| Meeting |
MS&T25: Materials Science & Technology
|
| Symposium
|
Integrated Computational Materials Engineering for Physics-Based Machine Learning Models
|
| Presentation Title |
Bayesian Optimization of KWN Precipitation Model Parameters for Improved Predictive Performance |
| Author(s) |
Emre Cinkilic, Batuhan Dogdu, Tomas Manik, Bjorn Holmedal |
| On-Site Speaker (Planned) |
Emre Cinkilic |
| Abstract Scope |
The Kampmann–Wagner Numerical (KWN) model is a widely used physics-based framework for simulating precipitation kinetics and mechanical properties in age-hardening alloys. However, its predictive accuracy is limited by difficult-to-measure inputs such as interfacial energy and nucleation site density. These limitations are further amplified under complex thermal histories, common in additive manufacturing, welding, and quenching, where rapid thermal fluctuations challenge model assumptions. This work presents a data-driven framework that combines Bayesian optimization and active learning to calibrate KWN model parameters using limited experimental and literature data. The approach is demonstrated on the A356 (Al7SiMg) alloy system for both isothermal (aging, paint baking) and non-isothermal (quenching) heat treatments. The optimized parameters are used to predict precipitation behavior and resulting properties such as yield strength, ultimate tensile strength, and hardness. This framework enhances the reliability and applicability of precipitation modeling in industrial alloys exposed to complex thermal cycles. |