Analysis and prediction of the mechanical properties of cold rolled Al-Li preforms using statistical and artificial neural network models

Abstract In this study, aluminum-lithium (Al-Li) alloys with <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mi>x</mml:mi> </mml:math> wt% lithium ( <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" overflow="scroll"> <mml:mi>x</mml:mi> </mml:math> = 0, 1 and 2) were synthesized using powder metallurgy and cold rolling, with response surface methodology (RSM) and artificial neural networks (ANN) employed for predictive modelling comparative analysis. Al-Li alloys are known for their high strength-to-weight ratio, making them suitable for aerospace applications. Yield strength, ultimate tensile strength (UTS), ductility, and fractional theoretical density (FTD) were analyzed based on lithium content, sintering time (60, 90, and 120 min), and rolling passes (2, 4, and 6). Microstructural analysis was performed using SEM, XRD, and optical microscopy. RSM followed the I-optimal design approach, developing empirical models through variance analysis. Findings indicated optimal sintering times of 90 min for improved ductility and FTD, and 120 min for enhanced yield strength and UTS. Cold rolling induced strain hardening, with UTS peaking at 141.170 MPa, while yield strength initially increased but declined after two roll passes due to over-straining. The ANN model, utilizing a 3-20-4 topology, exhibited strong predictive performance, achieving a determination coefficient (R 2 ) of 0.99628. Both RSM and ANN successfully predicted Al–Li alloy properties within the investigation range. However, overfitting concerns, particularly with unseen data, affected ANN’s accuracy in predicting ductility. Overall, ANN outperformed RSM in predictive accuracy, especially for nonlinear behavior, making it a more effective tool for modelling mechanical properties of Al-Li alloys.