Rapid Design Optimization of a Sheet Metal Part by Applying a Neural Network

This paper presents a neural network-based design optimization process of sheet metal parts. We used a support part with three mounting holes as an example part. There are four parameters in the design: three hole-to-hole distances and the maximum vertical force exerted on the support. The mass of the support is minimized while keeping the stress below a set limit, and the optimization process takes only 3–5 seconds. The process yields the ready design and its CNC code. The developed optimization process is integrated into a web shop of an ERP system to demonstrate its applicability in designing and producing custom-tailored parts efficiently and in real time. The neural network was validated with two validation datasets with sizes of approximately 90 and 900 datapoints. The resulting CNC code was tested by cutting 10 designed parts from an aluminum sheet with a conventional machining center. The design algorithm together with its web shop and ERP integrations forms an attracting application for especially small and medium-sized metal industry companies that manufacture large numbers of custom parts and small series and are interested in mass customization.