An advanced deep learning approach for energy absorption prediction in porous metals across diverse strain rate scenarios

Different strain rates affect yield stresses, influencing energy absorption characteristics. Thus, analyzing porous metals requires considering both quasi-static and impact loads. Current methods, like experiments and finite element method, often fail to assess all factors quickly. Hence, this study introduces a deep learning model that integrates convolutional neural network (CNN) and long short-term memory network (LSTM) to analyze the energy absorption characteristics of different porous metals under various strain rates scenarios. Firstly, ABAQUS/Explicit is developed using Python for batch computing to construct the dataset. Three types of loads, quasi-static compression and impact loads (strain rates of ${\text{92.59 s}}^{-1}$ and ${\text{231.21 s}}^{-1}$), are considered. Subsequently, the microstructure of porous metals, encoded with strain rate, is simultaneously fed into the CNN for feature extraction. Finally, the output of the CNN, in conjunction with the strain data, is utilized as input for the LSTM, which establishes the intrinsic correlation among microstructure, strain rate, and energy absorption characteristics. Results show that the CNN-LSTM model achieves an accurate and fast prediction of the energy absorption characteristics of porous metals with both random pores and circular pores in a wide range of strain rates. Additionally, the CNN-LSTM model assesses the time accumulation of energy absorption, enhancing the accuracy and comprehensiveness of the overall energy absorption evaluation. It is anticipated that a method based on deep learning can offer a fresh perspective on assessing the exceptional performance of intricate porous materials across diverse loading scenarios.