A Loosely-coupled Neural-symbolic approach to Compliance of Electric Panels

This paper presents an ongoing work on project MAP4ID ``Multipurpose Analytics Platform 4 Industrial Data'', where one of the objectives is to propose suitable combinations of machine learning and Answer Set Programming (ASP) to cope with industrial problems. In particular, we focus on a specific use case of the project, where we combine deep learning and ASP to solve a problem of compliance to blueprints of electric panels. The use case data was provided by Elettrocablaggi srl, a SME leader in the market. Our proposed solution couples an object-recognition layer, implemented resorting to deep neural networks, that identifies components in an image of an electric panel, and sends this information to a a logic program, that checks the compliance of the panel in the picture with the blueprint of the circuit.