Abstract

This study presents an AI-driven framework designed to concurrently forecast the hardness and constituent phase fractions of micro-alloyed steels, accounting for specific chemical compositions and thermomechanical processing parameters. The proposed architecture utilizes a feed-forward neural network (FNN) integrated with an ensemble learning technique to bolster predictive reliability and precision. Training was conducted using a comprehensive empirical dataset derived from the continuous cooling transformation (CCT) diagrams of 39 distinct steel grades. Model inputs incorporate the alloy’s elemental constituents and a cooling profile defined by discrete time-temperature coordinates. To evaluate the influence of alloying elements and processing conditions, a sensitivity analysis was performed on the validated model. This assessment, quantifying output variance relative to input fluctuations, demonstrated high fidelity with both experimental observations and established metallurgical literature. Consequently, the model serves as a robust tool for predicting steel properties under diverse cooling regimes, aligning consistently with fundamental theoretical principles and experimental evidence.