Double-machine-learning-based data-driven stochastic flow stress model for aluminium alloys at elevated temperatures

Abstract

As a widely used metal, aluminium alloys take an important role in many engineering fields. Quantifying the stochastic flow behaviour of aluminium alloys at high temperatures is inevitable for performing the reliability design in structural fire engineering. This work proposes a double-machine-learning (DML) framework for modelling the stochastic flow stress of aluminium alloys at elevated temperatures directly from the dataset. The proposed framework extracts the mean and standard deviation of the flow stress from the dataset and describes them separately by two machine learning (ML) models. The experimental data of the 6061-T651 aluminium alloys is used to validate the DML-based stochastic flow stress model. By comparison between four popular ML models, the artificial neural network (ANN) model and the Gaussian process regression (GPR) model are the most appropriate choices for predicting the mean and the standard deviation, respectively, in the DML framework. The stochastic distribution of the flow stress estimated by the DML framework, which consists of the ANN model and GPR model, is in high agreement with the experimental results. The good agreement between predictions and experimental results demonstrates that the DML framework can accurately describe the stochastic flow stress of aluminium alloys at elevated temperatures and can be used in the reliability design processes of aluminium structures. • Double-machine-learning (DML) framework is proposed for stochastic flow stress at elevated temperatures • Established stochastic flow stress model is validated by experimental data of aluminium alloys • DML framework with ANN and GPR model is the most suitable choice for aluminium alloys • Probability distribution of aluminium alloys’ flow stress is accurately reproduced by DML framework