Session: 10-09-01: Multiphase Flows and Applications

Paper Number: 112373

112373 - Improving Efficiency of Automotive Coating and Curing Processes Through Deep Learning Algorithms and High-Fidelity CFD Modeling 

The automobile industry has relied on computational fluid dynamics (CFD) simulations to analyse and optimise the coating and curing processes, speed up product development, and lower the cost of product development. However, CFD simulations of these processes can be computationally expensive due to the complexity of the models and the large number of simulations needed, especially when its used complex sprays such as the electrospray. As a result, more efficient methods must be developed to reduce computing time without compromising accuracy. In this article, we analyse how deep learning techniques can be used to predict coating and curing processes using electrospray CFD simulation.

A dataset of 3D Eulerian-Lagrangian CFD simulations of coating and curing processes employing electrospray for the automotive industry has been used to train four different deep-learning models. We investigate the effectiveness of convolutional n We also looked into how hyperparameters such as batch size and layer count affected deep learning model performance compared to conventional CFD simulations, we evaluated the deep learning models' efficiency and accuracy in terms of computing time. We also looked into how hyperparameters such as batch size and layer count affected deep learning model performance. Also, we looked at the target’s final deposition and distribution was required to accurately estimate the final deposition and the Eulerian velocity field. Furthermore, we studied the percentage of snapshots of the droplet distribution electrospray necessary to predict the target’s final deposition and the Eulerian velocity field from the Lagrangian distribution.

According to our findings, deep learning models can drastically reduce the amount of time needed to run CFD simulations. Depending on the model and hyperparameters applied, we can forecast the whole CFD simulation by utilising somewhere between 10 and 15% of the initial spray development. Also, we discovered that the CNN-LSTM model outperformed the other models in terms of accuracy and computational efficiency. Where the convolutional LSTM layers can extract better the features of the input snapshots. In addition, we discovered that the performance of the deep-learning models was significantly impacted by the batch size and learning rate hyperparameters.

Overall, our research demonstrates the potential of deep learning techniques to significantly shorten the computing time of CFD simulations of coating and curing processes for the automotive sector. The results of this study have significant implications for coating and curing process design and optimization in the automobile industry as well as in other industries where CFD simulations are frequently employed.

Presenting Author: Silvio Candido University of Beira Interior

Presenting Author Biography: Sílvio Cândido is a PhD student in mechanical engineering at the University of Beira Interior (Covilhã, Portugal). His current research is on the numerical modelling of multiphase flows, actually focused on the breakup of electrohydrodynamic jets, formed by Taylor cone shapes.

Authors:

Silvio Candido University of Beira Interior
Mohammad-Reza Pendar University of Beira Interior
José Carlos Pácoa University of Beira Interior