Plagiarism detection in parametric CAD models: A supervised similarity framework for engineering education

The growing use of Computer-Aided Design (CAD) software in engineering education has reshaped assessment practices, creating new challenges for academic integrity. In parametric modelling environments, plagiarism extends beyond geometric duplication to the reuse of modelling sequences, feature-tree structures, and constraint-based design intentions, which text-oriented plagiarism-detection tools cannot capture. This article presents a similarity-screening framework for academic-integrity assessment in parametric CAD coursework. The method combines geometric comparison, feature-tree composition, the ordering of operations in the construction history, and sketch-level constraint structure into a multi-level analysis. By examining both the final artefacts and the underlying design intent, the system computes an aggregate similarity score for each model pair, ranks the pairs, and flags those exceeding an operational threshold for expert review; the score is intended for screening and does not by itself constitute evidence of plagiarism. The framework was calibrated on fourteen sessions (395 models; 7548 within-session comparisons) and evaluated retrospectively on four further sessions held out from the estimation of the weights and of the threshold (176 models; 5637 comparisons), each submission having received a manual determination independent of the score. Within the sessions examined, no confirmed case of plagiarism fell below the operational review threshold, and on average about two per cent of the within-session comparisons were flagged for expert review. Because comparisons from the same submissions are not independent, results are reported per session. Beyond screening, the approach supports training tools that promote academic integrity, and frames CAD plagiarism as a technology-mediated phenomenon requiring specialised assessment strategies.