03/22/2024
By Danielle Fretwell

The Francis College of Engineering, Department of Plastics Engineering, invites you to attend a Doctoral Dissertation
defense by Joshua Krantz.

Candidate Name: Joshua Krantz
Degree: Doctoral
Defense Date: Friday, April 5, 2024
Time: 2 to 4 p.m.
Location: ETIC 345

Committee:

  • Advisor Davide Masato, Assistant Professor, Department of Plastics Engineering, University of Massachusetts Lowell
  • David Kazmer, Professor, Department of Plastics Engineering, University of Massachusetts Lowell
  • Margaret Sobkowicz-Kline, Professor, Department of Plastics Engineering, University of Massachusetts Lowell
  • Peng Gao, Assistant Professor, Department of Engineering and Design, Western Washington University

Brief Abstract:

The rise of Industry 4.0 and the drive toward more sustainable manufacturing practices have prompted the plastics industry to explore sophisticated process control methodologies. Closed-loop strategies enable real-time process adjustments, thereby enhancing product quality control. This becomes particularly crucial when utilizing secondary feedstock, necessitating compensation for materials' compositional and processing variability. In essence, developing and adopting advanced process control technologies can facilitate the broader utilization of recycled plastics, elevate automation levels, and enhance global competitiveness within the injection molding industry.

This research centers on pressure-controlled injection molding as an alternative to the conventional velocity-controlled process. In the injection phase, screw movement is controlled to uphold a consistent nozzle pressure, enabling more precise filling flow and uniform cavity pressure distribution. Additionally, the integration of in-mold pressure sensors and screw movement control facilitates the implementation of adaptive process control strategies, wherein the process adapts in real time to fluctuations in material viscosity. The dissertation is divided into three main sections, which focus on 1) the benchmarking of pressure-controlled injection molding, 2) the implementation of auto-viscosity adjustments for varying grades of recycled plastics, and 3) the development of machine-learning models to predict processing behavior and product quality from sensors’ data.

The first section focuses on injection molding experiments to explore the use of pressure-controlled injection molding for recycled polyolefins alongside a direct comparison with a conventional velocity-controlled injection molding process. The findings demonstrate a reduction in energy consumption and an enhancement in mechanical performance for the pressure-controlled process. Moreover, the differences in process setup are highlighted, and the importance of defining the processing window is discussed.

Building on the knowledge of pressure-controlled molding, the second section leverages a closed-loop adaptive process control system to reduce processing variability. Using cavity pressure sensors, the polymer melt's apparent viscosity is measured, and the screw movement is dynamically adjusted to maintain it within a target value. The methodology is evaluated using five different blends of recycled polypropylene, characterized by significantly different viscosity. The parameters that control the auto-viscosity adjustment, such as trigger time and pressure setpoint are investigated. The results showed that the auto-viscosity control system successfully reduced the processing variability. The variability in the mechanical properties and weight was also analyzed, highlighting minor differences.

The third section develops and compares artificial neural networks and multivariate regression models for process control. The models are first trained and validated using experimental data. The hyperparameters are optimized to avoid overfitting, a common issue for machine learning models trained with limited datasets typical of manufacturing. Then, an optimization problem is defined to obtain the process parameters that guarantee achieving a defined product quality output. The work is carried out using the data gathered from the experiments conducted in the second section of the thesis. The models are designed for implementation in a pressure-controlled injection molding process. The results show that the artificial neural network models are the most accurate in predicting the correlation between processing conditions and quality metrics.