2025 ASEE Annual Conference & Exposition

Automated Grading of Engineering Mechanics Assignments Using Large Language Models and Computer Vision: A Work in Progress

Presented at Assessments, Feedback, & Grading

This project aims to develop an automated grading framework for the Statics course at a Canadian University, which serves over 1,500 students annually. Grading these complex assignments, often involving handwritten responses and diagrams, currently requires the efforts of over 26 teaching assistants (TAs). The manual system faces challenges including grading inconsistencies, delays in providing feedback, and significant resource strain. To address these limitations, the proposed framework integrates Optical Character Recognition (OCR) and Large Language Models (LLMs) to automate the grading process. The framework begins with anonymization and preprocessing of student submissions, followed by digitization using OCR software such as Mathpix to convert handwritten content into machine-readable formats. The digitized files are evaluated using a grading platform that employs LLMs guided by predefined marking schemes and step-by-step evaluation prompts. The system generates grades, detailed feedback, and consolidated reports for each submission. Preliminary testing demonstrated significant reductions in grading variability and time, while maintaining a strong alignment with human-assigned grades. The findings show that Qwen LLM achieves higher consistency across multiple evaluation metrics, whereas GPT-4 provides superior accuracy in specific scenarios. This approach enhances the scalability and efficiency of grading practices, making it a promising solution in large-scale educational settings.

Authors

Dr. Ahmed Mowafy PEng http://orcid.org/0009-0002-4557-1176 University of Alberta [biography]

Dr. Ahmed Mowafy Saad is an Assistant Teaching Professor in Civil and Environmental Engineering at the University of Alberta. He teaches large first- and second-year courses such as Engineering Mechanics and Mechanics of Deformable Bodies, reaching over 1,800 students annually. With over 13 years of combined academic and industry experience in the Middle East and Western Canada, he integrates real-world insights into innovative teaching practices. His recent work focuses on AI-assisted grading tools and dynamic, multi-version assessments that improve academic integrity and efficiency at scale. He has also introduced self-marking strategies inspired by Indigenous assessment practices to promote reflection and deeper learning. In addition, he pioneered the Fantasy Mechanics League (FML), a gamified classroom framework that enhances student motivation and engagement in large lecture settings. His work is grounded in the belief that “teaching empowers you to create countless lives through the brilliant minds you cultivate.”
Mr. Mohammad Talebi-Kalaleh University of Alberta [biography]

Mohammad Talebi-Kalaleh is a self-motivated third-year PhD student in Structural Engineering at the University of Alberta. His research focuses on leveraging Large Language Models (LLMs) in engineering education and exploring their applications in structural engineering. Mohammad's work primarily revolves around artificial intelligence, smart infrastructure, and bridge monitoring through crowdsensing techniques. He is particularly passionate about integrating advanced AI-driven solutions into structural health monitoring to enhance infrastructure resilience and efficiency.
Mohamed Sabek http://orcid.org/0009-0005-2906-9874 University of Alberta [biography]

Mohamed Sabek is a Ph.D. student in Construction Management and Engineering at the University of Alberta. He is the lead developer of the AI-driven grading framework presented in this paper, which integrates Optical Character Recognition and Large Language Models (LLMs) to enhance assessment efficiency in large-scale engineering courses. His research lies at the intersection of artificial intelligence, computer vision, and engineering education, with a special focus on fine-tuning vision-language models (VLMs) for real-time construction site monitoring. Mohamed’s broader research goals aim to advance intelligent automation in construction environments to improve safety, productivity, and resource allocation. He holds PMP and RMP certifications and has received multiple awards recognizing his contributions to AI applications in civil engineering and education technology.
Harry Peng University of Alberta
Mohammad Aqib University of Alberta
Dr. Samer M. Adeeb P.Eng. University of Alberta
Mohamed Magdy Elgammal University of Alberta
Dr. Clayton Pettit University of Alberta

Download paper (1.59 MB)

» Download paper

« View session