Symbol Detection in Mechanical Engineering Sketches
Digital transformation is pervasive in our daily lives, evident in new technologies like smart devices or AI chatbots. This digitalization extends to product development to enhance product quality and reduce development time and costs. However, design data is diverse and phase-dependent. This paper presents an approach for detecting principle sketches in early development phases, aiming to automatically recognize symbols using object detection models. Existing approaches were analyzed, and a new procedure with synthetic training data generation was developed. Six different generation types were analyzed and tested with different one- and two-stage detection models. The procedure was evaluated on two unknown test datasets: one focusing on gearbox variants and another derived from CAD assemblies.