Architekturtechnik

Abstrakt

Classification of Industrial Processes from Engineering Drawings Using Graph Neural Networks

Zivko Nikolov

While ample scanned engineering drawings area unit received each year, the net quotation corporations for custom mechanical components have knowledgeable about a billowing got to increase their process potency by substitution the presently manual examination method with associate degree automatic system. Previous work has used ancient, and data-driven computer-vision approaches to observe symbols and text info from the drawings.However, there lacks a unified framework to work out the associated producing processes as a crucial step for realizing associate degree automatic quoting system. During this paper, we tend to propose a process framework to mechanically verify the producing methodology acceptable to provide every queried engineering drawing, like lathing, flat solid bending, and edge. We tend to gift a data-driven framework that directly processes the formation pictures with a series of pre-processing steps and accurately determines the corresponding producing strategies for the queried spare a graph neural network. We tend to propose a completely unique line tracing algorithmic rule to rework advanced geometries in engineering drawings into vectorized line segments with bottom info loss.