Predicting and optimising aluminium alloy compositions for target mechanical properties using machine learning

This study presents a machine learning framework for aluminium alloy design, solving the long-standing challenge of predicting alloy compositions from target properties. The prediction of alloy compositions is an inverse design problem, which is addressed by the use of machine learning approaches in the present work, reducing the material development time. Eight regression algorithms are benchmarked for forward prediction, with gradient boosting regression (GBR) emerging as the most accurate model. This model was then used to find the important alloying elements for training the forward regressor. For the inverse prediction, optimisation approach was utilised by using a genetic algorithm (GA) with GBR. The proposed inverse model successfully generated alloy compositions that achieve high performance, for properties like ultimate tensile strength (UTS) and yield strength (YS), with R2 values greater than 0.84 for all models. Predicted compositions were validated using the forward model and experimentally confirmed for an Al–Si alloy, demonstrating strong agreement. The results highlight the potential of ML-assisted inverse design to accelerate alloy development, reduce experimentation and enable precise optimisation of aluminium alloys for targeted properties.