Finite‐Element‐Based Deep‐Learning Model for Deformation Behavior of Digital Materials

Smart composite materials fabricated through 4D‐printing methods are attracting enormous research attention for their ability to respond (typically deform) under external stimuli. The design process for such smart materials requires iterations of finite‐element simulations that are computationally expensive. Recently, researchers have tried replacing numerical simulations with machine learning (ML) models to predict the output at a much higher speed. However, there exist very few studies that explore the model algorithm's expressive capacity and analyze the physical interpretation based on the problem. This paper focuses on using ML to predict the nonlinear deformation behavior of digital materials. Various problem construction approaches and model performance are compared and discussed. It is shown that clustering the materials helps improve the generalization of training and models that treat material features as an array of numbers still face difficulties to provide accurate predictions. Inspired by modern computer vision technologies, convolutional kernels outperform other methods by recognizing the material distribution patterns. The performance is further enhanced after reconstructing the regression problem into classification. Moreover, high‐level material design information can be extracted from the model through a sensitivity analysis. This framework may greatly improve the response prediction and design process for 4D‐printed smart materials.