Mechanical properties prediction of bi-metal foam sandwiches using machine learning methods and elastic deformation behaviour

Abstract

Metal foam sandwiches are a kind of ultra-lightweight material made from a porous metal core bonded to two face sheets. Friction stir welding (FSW) is utilised in welding bimetal foam sandwiches. It is worth mentioning that the exact relation between mechanical properties and process parameters is challenging to determine. The innovation lies in the non-destructive estimation of mechanical properties (Young's modulus, ultimate tensile strength and fracture strain) through elastic deformation data and the novel application of artificial intelligence techniques optimised by genetic algorithms, eliminating dependency on input process parameters. After proper network training, three methods are employed to estimate these mechanical properties: a decision tree, a feedforward neural network and long-short term memory. These are chosen to investigate the influence of both machine/deep learning methods in predicting the mechanical properties of the FSW final product. Moreover, a genetic algorithm is employed to find the optimal hyperparameters of the three investigated prediction models to reach the highest accuracy. The results prove the efficiency of the proposed feedforward neural network in the estimation of Young's modulus and ultimate tensile strength for the bi-metal foam sandwiches with lower mean absolute error (MAE) and higher correlation coefficient compared to the decision tree (63.9% lower MAE and 25.50% higher correlation coefficient) and long-short term memory (77.50% lower MAE and 25.05% higher correlation coefficient). In addition, the proposed decision tree model accurately predicts the fracture strain with R-square and root mean square error as 0.61429 and 1.3862×10-5, respectively.