UNIVERSITY PARK, Pennsylvania — Manufacturing technology, the various devices and systems that enable the production of manufactured goods, is changing rapidly. Due to its digital nature, the accelerated pace at which manufacturing technology advances often makes it difficult for engineers to incorporate the latest technologies and production processes into product design.
Penn State researchers have received a $424,743 grant from the National Science Foundation to investigate how the size and quality of datasets created by numerical models relate to machine learning and how this affects the support provided to engineering designers.
The three-year project, “Investigating the Effectiveness of Machine Learning Paradigms to Support Engineering Designers in Rapidly Evolving Digital Manufacturing,” is led by Principal Investigator Christopher McComb, Assistant Professor of Engineering Design, industrial engineering and mechanical engineering. Nicholas Meisel, assistant professor of engineering design and mechanical engineering, and Timothy Simpson, Paul Morrow professor of engineering design and manufacturing, are co-principal investigators. Mechanical engineering PhD student Glen Williams also recently joined the project.
We know that today’s manufacturing machines often rely on digital models that produce large amounts of data. The researchers’ approach revolves around a two-step process that uses these existing datasets to gain design insights.
“In the first stage, we will use deep learning to extract features, or recurring patterns, from the database,” he said. “In the second stage, we’re using deep learning again to learn how to use these models to predict performance and behavior — things like strength and manufacturability.”
Datasets vary widely in quantity and quality, making the usefulness of machine learning somewhat unknown. The team will use additive manufacturing, or 3D printing, to study how the quantity and value of datasets relate to the accuracy and usefulness of machine learning capabilities and how this relates to the support provided to designers engineering.
McComb said additive manufacturing was chosen as the case study because of its status as an emerging consumer manufacturing technology.
“Using additive manufacturing, we will develop methodologies that will help us better support novice engineers and designers with the next breakthrough manufacturing technology,” he said.
Through the combination of additive manufacturing, machine learning, and explainable artificial intelligence, McComb and his team will use data collected from existing 3D printing design challenges to investigate the use of design feedback. automated. Part designs will be gathered from online open sources and engineering course design challenges.
To test the effect of dataset size on feedback accuracy and level of detail, the team will create a machine learning pipeline that extracts patterns from design datasets.
As the last stage of the project, studies with engineering students will be carried out in order to provide them with professional training and to collect data. McComb defines students as key stakeholders and said that by involving them in research, the team helps prepare them for work in the manufacturing industry.
“Ultimately, our goal is to better serve them [engineering students] helping them design new manufacturing technologies faster and more efficiently,” he said. “By participating in one of our studies, we want them to learn more about additive manufacturing and also recognize that they are helping us to better support other learners in the future.”
The search results will lead to a set of mechanical part design data stored as voxels, the 3D equivalent of pixels; a better understanding of the impact of the quantity and quality of a data set on the learning and feedback capabilities of a machine; and first-hand experience of the impact of real-time additive manufacturing feedback on solutions created by engineering designers.
“For companies, this work is going to help them understand what kinds of insights they can expect to extract from the design and engineering data they already have,” McComb said. “For students and novices, this work will provide foundational approaches to support them as they learn to use new manufacturing technologies.”