Mechanical drawings are crucial in industry for manufacturing and information retrieval. Companies face a recurring issue with managing their hand-drawn drawings produced in the mid-20th century. To address this, the engineers convert hand-drawn drawings into CAD drawings. This transformation process must be error-free to avoid material waste and significant costs. To ensure this, quality assurance engineers carefully verify the final CAD drawings to ensure all information is preserved. This verification process is time-consuming and laborious, often requiring the dedicated effort of two engineers for an entire week per drawing.

We propose to automate this verification step to identify dissimilarities, defined as divergences causing inconsistencies between CAD and original hand-drawn versions. Identifying these dissimilarities will reduce the effort of the engineers to perform the necessary corrections.

We outline our planned approach to identify dissimilarities between mechanical drawings. Our approach is divided into two aspects: interpreting the drawings to extract information and comparing the drawings based on this information. We also present an initial contribution by proposing a formalism to describe the information contained in mechanical drawings, distinguishing between pure geometric shape and annotations. This formalism is instantiated through a graph-based structure that represents this information, with the future goal of allowing effective segmentation and comparison of the drawings.