Session: 02-01-04: 7th Annual Conference-Wide Symposium on Additive Manufacturing: Polymers II

Paper Number: 95120

95120 - Modeling the Interplay Between Process Parameters and Part Attributes in Additive Manufacturing Process With Artificial Neural Network 

In order to advance with the fourth industrial revolution as well as adapt to the expected level of modern production systems, additive manufacturing opens the door of digitalized production process by combining intelligent system to optimize the final products to the desired level. Extrusion based additive manufacturing (AM) technique is used as a practical tool for prototyping as well as making functional parts. Surface quality, dimensional accuracy, and strength are the important factors to maintain high standard in the manufactured products. These part attributes are largely dependent on the AM process parameters. The overall part build quality and attributes can be improved by varying the different AM process parameters. Since the AM processes include several process parameters that need to be controlled for building a part, it is impractical to test all possible combinations of the different levels of the process parameters to optimize the attributes of a part. In this study, we model the interplay between the process parameters and the part attributes with artificial neural networks (ANN) to predict the effect of a set of process parameters on the part attributes in extrusion-based AM process. Five process parameters including build orientation, print speed, extrusion temperature, deposition direction, and layer thickness with three levels are used in this study to fabricate parts following an orthogonal array experimental design. Three attributes including dimensional accuracy, surface roughness, and tensile strength of the fabricated parts are measured and used to train, validate, and test the proposed multilayer artificial neural network models. Based on training, validation, and testing accuracy, the appropriate number of hidden layers and the exact number of nodes in each hidden layer of the networks are determined. Four different ANN models are proposed where three of them are for the three individual part attributes and the fourth one is for the combination of all three attributes. Models are trained by using Levenberg-Marquardt, Bayesian regularization, one step secant, and gradient descent algorithms where Levenberg-Marquardt algorithm outperforms the others. The root mean square error (RMSE) and correlation coefficient analysis show that the training, validation, and testing results approximately match with the experimental data for the proposed ANN models. The ANN models for the individual part attributes outperform the model for the combination of three output part attributes in terms of the RMSE and correlation coefficient for the training data. We have also performed a comparison among the individual output (surface quality, dimensional accuracy, tensile strength) results with respect to all the process parameters and investigated which parameters have a greater effect on the individual part attributes. Therefore, the trained ANN models can be utilized to predict the resulting attributes of various parts for different unexplored process parameter levels thereby allowing us to optimize the process parameters and part attributes without costly and time-consuming fabrication experiments.

Presenting Author: Nazmul Ahsan Western Carolina University

Presenting Author Biography: Dr. Ahsan is currently an Assistant Professor of Mechanical Engineering and Engineering Technology in the School of Engineering and Technology at Western Carolina University. Dr. Ahsan achieved his Ph.D. degree in Industrial and manufacturing Engineering from North Dakota State University in 2019. Before that he completed his master’s degree in Industrial Engineering and Management from the same university. His teaching and research interest include advanced design and manufacturing, additive Manufacturing/3D printing, artificial intelligence in additive manufacturing, heterogeneous lightweight porous structure design and manufacturing, and bio-printing. Dr. Ahsan has thus far published over 30 journal papers and conference proceedings. He has received the Outstanding Early-Career Faculty Award 2020-2021 in the College of Engineering and Technology at Western Carolina University. He is a recipient of Chancellor's Gold Medal Award and Prime Minister’s Gold Medal Award for outstanding performance in his bachelor's degree.

Authors:

Jayanta Deb Western Carolina University
Nazmul Ahsan Western Carolina University
Sharmin Majumder Texas A&M University