Hybrid Deep Learning Models for Predicting Compressive Strength of FDM 3D-Printed Structures

Resumo

The correct prediction of compressive strength is vital to the successful design of fused deposition modeling (FDM) 3D-printed structures, although the traditional empirical and machine learning methods tend to fail to account for the multifaceted and nonlinear relationships among the parameters of the process and the mechanical behavior. To tackle this problem, we have generated and analyzed hybrid deep learning algorithms to predict compressive strength of FDM-fabricated specimens. Systematic variation of important printing parameters has produced experimental data which were then subjected to standard compression testing. Convolutional neural networks and recurrent neural networks have been carried out separately and then combined into hybrid to capitalize on both spatial feature extraction and sequential dependency learning. Standard measures of statistical analysis and comparative graphical analysis were used to carry out model performance. The findings indicate that hybrid deep learning model always do better than single models, recording lower prediction errors, and better generalization with unknown data. These results indicate that the use of complementary deep learning designs is indeed effective in improving the predictive power of mechanical properties of additively manufactured components. It is concluded that hybrid deep learning offers an efficient and powerful structure of prediction and optimization of compressive strength of FDM 3D-printed structures using data.