The topological analysis of mechanisms is a fundamental step in mechanical design, traditionally carried out manually by engineers who must interpret functional diagrams and extract structural properties from them. This paper presents a software framework that automates this process by applying deep learning-based instance segmentation to functional diagrams of mechanisms. A YOLOv11 segmentation model was trained on a custom dataset of 200 annotated images, expanded to 501 through data augmentation, to detect and classify kinematic components such as joints and links of varying degrees. The trained model achieved an overall segmentation mAP50 of 0.951 and mAP50-95 of 0.749 on the validation set. A post-processing pipeline built upon geometric analysis using the Shapely library determines the connectivity between detected components, enabling the automatic computation of fundamental kinematic parameters including the Degree of Mobility via Gr¨ubler’s criterion and the number of independent circuits via Euler’s formula for planar graphs. The system is encapsulated in an interactive graphical interface that provides multiple visualization modes, including segmentation overlays, connectivity graphs, and topological representations generated with NetworkX. The results demonstrate the viability of using computer vision as a practical tool to assist and accelerate mechanism synthesis, serving both educational and engineering design purposes.

Mechanism synthesis instance segmentation YOLOv11 topological analysis functional diagrams kinematic chains