Material combination optimization for brazed ceramic-metal composites using Artificial Intelligence

Abstract

This study proposes an Artificial Intelligence (AI) driven methodology for predicting a combination of brazed ceramic-metal composite materials. Multiple machine learning (ML) algorithms are compared with the deep learning (DL) model. The developed models are tested using k-fold validation. Nine different input-output feature configurations are evaluated to assess the model performance. The input-output feature comprises material properties, namely, the coefficient of thermal expansion (CTE) and molecular mass of brazed ceramic-metal composite materials obtained from literature and the strength parameter (average Von Mises Stress (VMS)) estimated from Finite Element Method (FEM) simulation for joint assembly structure. A multi-output model, Autoencoder (AE), has also been developed and tested to predict various features. The ML model, namely the polynomial regression (PR), outperforms the other ML/DL models with a Mean square Error (MSE) of 0.01 for the test data. The autoencoder model with a 32-16-32 structure outperforms LR, PR, RF, and ANN with an MSE of 0.04% for the prediction of unseen data. The developed multi-output model accurately predicts all the features (single and multiple), while PR fails to accurately predict multi-output features of low importance. The developed AE model predicts the different material properties with an average error of ~0.16-3.78% with literature-reported values.