Acoustic field-assisted particle patterning is a promising method to control the spatial distribution of the functional particles (such as mechanical, electrical, chemical, and thermal functions) dispersed in a polymer matrix to fabricate multifunctional smart composite objects. To better understand the acoustic particle patterning process, a 3D high-fidelity multiphysics model is generally utilized. However, thousands of forward simulations are often required to determine a suitable set of input parameters for the desirable particle pattern. It is advantageous to replace the computationally expensive forward simulation model with a cheaper surrogate model for the optimization of the acoustic particle patterning process.In this study, a physics-informed deep-learning (PIDL) surrogate model is developed and applied to predict the acoustic particle patterning process’s outcome. The surrogate model is based on deep convolutional and transposed convolutional neural network architectures. The training and test samples involve 2188 input-output pairs. The inputs consisted of the acoustic actuator setups, which is also the input to a 3D physics-based model, and the acoustic pressure from a 2D physics-based model. The actuator setups of the model include the location, size, vibration frequency, and vibration amplitude of the actuator(s). The output of the surrogate model is the von Mises stress pattern of the film substrate from the 3D model. This 3D model is a simplified version of the actual experimental setup where we only simulate the von Mises stress of the film substrate. Based on previous experience, the von Mises stress pattern of the film substrate has a direct connection to the actual particle pattern. The 2D model, which is the cross-section of the 3D model, can run at many orders of magnitude faster than the 3D model. Both the 2D and the 3D models are constructed and implemented in the COMSOL Multiphysics simulation software.After training, the PIDL surrogate model with five convolutional layers and five transposed convolutional layers can accurately predict the von Mises stress pattern given a new actuator setup and acoustic pressure from a 2D model that the surrogate model has not observed. When compared to the DL model, the PIDL model is capable of accurately predicting the von Mises stress patterns at a significantly less computational cost. In particular, the DL model requires as many as nine convolutional and transposed convolutional layers to obtain a similar level of prediction accuracy as the PIDL model with five convolutional and transposed convolutional layers. The time used to train and test the PIDL model with five convolutional layers is about 5.5 and 1.9 times faster than the DL model with nine convolutional layers, respectively. These results show that the acoustic pressure from the 2D model can condition/inform the DL model to generate a von Mises stress pattern that is closer to the 3D model. The prediction time of the PIDL surrogate model is more than 16000 times faster than the 3D physics-based model, which further demonstrates the potential benefit of using the surrogate model in optimizing the particle patterning process. Parametric studies show that when the training data is reduced by 50%, the PIDL model accuracy only decreases by around 6%, which indicates the proposed model can perform well even if training data is scarce.

Surrogate Modeling of Acoustic Field-Assisted Particle Patterning With Application to Smart Polymer Composite Fabrication in Stereolithography: A Physics-Informed Deep-Learning Approach.

Paper Proceedings: https://asmedigitalcollection.asme.org/IMECE/IMECE2021/volume/85543

ASME 2021 International Mechanical Engineering Congress and Exposition (IMECE2021)

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.