Zachary Menard, Building HRC Labs | AI for heat treatment, HRC Labs

Machine learning is becoming increasingly capable of predicting heat-treatment outcomes such as hardness, case depth, and distortion, but the harder question is how to make those models usable in real heat treatment decisions. This presentation explores a practical path forward: combining data-driven methods with established metallurgical models, thereby adapting those models to the behavior of a specific furnace, process, or material. The goal is to avoid treating machine learning as a standalone black box. A key part of that framework is explicitly quantifying uncertainty. This means understanding when a model is operating within familiar territory versus when it is extrapolating, and therefore how much confidence to place in a proposed recipe change or optimization measure, such as reducing the soak time on a given recipe. The goal is not to automate metallurgy, but to make existing metallurgical knowledge easier to apply, refine, and use with confidence in day-to-day heat-treatment decisions.