03/14/2024
By Danielle Fretwell
The Francis College of Engineering, Department of Plastics Engineering, invites you to attend a doctoral dissertation defense by Tahamina Nasrin on: Applying Machine Learning Approaches to Multilayered Polymer Composites
Candidate Name: Tahamina Nasrin
Degree: Doctoral
Defense Date: Thursday, March, 28, 2024
Time: Noon to 2 p.m.
Location: ETIC 445
Committee:
- Advisor: Amy Peterson, Associate Professor, Plastics Engineering, UMass Lowell
- Wan-Ting (Grace) Chen, Assistant Professor, Plastics Engineering, UMass Lowell
- Christopher Hansen, Professor, Mechanical Engineering, UMass Lowell
- Farhad Pourkamali Anaraki, Assistant Professor, Mathematical and Statistical Sciences, CU Denver
- Robert Jensen, Team Lead, Materials Data Science, DEVCOM Army Research Laboratory (ARL)
Brief Abstract:
Multilayered polymer composites can outperform neat polymers in terms of mechanical and chemical properties, so they have been widely used in additive manufacturing (AM) and polymer packaging. Material extrusion (MatEx) and vat photopolymerization (VP) are two AM techniques used to create complex structures and parts with a high strength-to-weight ratio. Multilayer polymer packaging materials, on the other hand, are vital for keeping foods for an extended period of time since they provide sealability as well as chemical inertness to the food items. However, because the qualities and performance of multilayered polymer composite components vary greatly depending on processing parameters and structure, it is difficult to control them. Furthermore, screening polymer materials for specific use is complicated by the multiple decision-making metrics. Physics-based models have proved effective for examining specific measures, but their computational complexity makes it impossible to include the entire process. Furthermore, screening techniques are heavily reliant on experiments, which take time and can be costly. Data-driven machine learning (ML) techniques can solve these challenges since they can understand the underlying pattern in a dataset and generate predictions for previously unseen cases. However, ML techniques fall short of being applied in material science because they require a large dataset to be trained on, which is often difficult to achieve because associated experiments to generate data are time consuming.
This dissertation explores avenues for applying ML techniques to multilayered polymer composite systems, focused on MatEx, VP, and multilayer food packaging films. The common aspect of each research project of this dissertation is that they are based on small datasets, which have been systematically acquired from physical experiments. These efforts have led to the development of end-to-end ML pipelines that encompass regression, classification, and dimensionality reduction-based algorithms. These pipelines span from data collection to the development and validation of predictive models. These ML-based pipelines were designed to predict important properties of parts in MatEx and VP, select suitable packaging materials for specific applications, and navigate the vast design space in a complex, highly filled VP system. The outcomes of this dissertation are expected to provide pathways for capturing the real-time behavior of multilayered polymer composites in their respective applications, thereby reducing both experimental and computational costs.