Graph neural network for shape recognition in the aluminium extrusion sector

The company operates in the extrusion industry and produces 300 dies per day. We are approaching the production of a die for our customer; therefore we access the database where around about 100.000+die profiles used in previous designs are manually stored. We search by keywords in the library and the outcome is that previous profiles do not fit. As a result, we review the delivery schedule for the day. At the end, we realize that the library was not properly organized, and we lost 2hours of work (25% of daily worktime) because the same shape already existed but could not be found since it had not been classified under the proper keywords.

The previous scenario represents a recurrent routine for several manufacturing companies whose production process is based on a MTO/ETO approach and seeks for efficiency gains in design-by-reuse (as literature on knowledge sharing describes). Companies potentially rely on thousands of models/data that they generated in the past but still fail in incorporating them optimally in their daily processes. According to its observation in the aluminium extrusion field, Intellico developed a use-case that current project “GRAPHO” underpins, in order to improve the search algorithm in the aluminium extrusion field by analysing images of past projects and to facilitate the images incorporation into the production process. The experiment bases its algorithm on automatic Graph-Structures generation to enable 2 sub-use-cases: (1) Image Similarity - the experiment, when given a profile as an input, returns profiles that are similar to it from the database; (2) Smart Database – the experiment re- organizes the profiles according to agreed-upon families of profiles, and uploads/stores new objects under these families. Therefore, while it scans for a similarity it also keeps the library organized.

Given the aforementioned use-cases, GRAPHO pursues the following objectives: (Str.) build a tool which can automatize the building of graphs in a dynamic way in combination with a Graph Neural Network (GNN) with the final result of (Spec. 1) increasing images reuse, (Spec. 2) running the code in a limited amount of time compared to as-is and (Spec. 3) automizing profile storage.

Overall GRAPHO objectives are strongly aligned with both Vanguard Region initiatives and AI Regio objectives. Intellico is intertwined into the economical ecosystem of the Lombardy Region, with 10+ collaborations with manufacturing companies. Moreover, GRAPHO designs a novel way to organize Data Spaces for Manufacturing (Call2, Topic1) as it will deliver (1) a novel tool based on Graph Neural Networks and Graph generation for a Data Space preparation in the aluminium industry. The tool will comprise a data model and a library; (2) a Data Space and a validated use-case in the aluminium extrusion industry.