Predicting Austenite Yield Strength in Steels by using Artificial Intelligence

The accurate control of austenite hot deformation requires reliable prediction of yield strength 
in order to design thermomechanical treatments to optimize the final properties of steels. This has 
motivated extensive efforts to develop predictive models in the past. Such models have demonstrated good 
accuracy, but their applicability has been limited to narrow compositional ranges or inaccurate evolution 
with temperature, largely constrained by the scope of the databases used for calibration. Therefore, when 
tested against broader datasets encompassing wider temperature and compositional variations, these 
models exhibit significant deviations. Moreover, strain rate—an influential factor in mechanical 
behavior—has often been overlooked in earlier approaches. In this study, we present a more 
comprehensive and robust model that incorporates strain rate effects and extends the compositional and 
temperature ranges. By integrating physical based methods and artificial intelligence architectures, the 
model presented significantly enhances the accuracy of yield strength prediction.