Deriving Machining Processes from Technical Drawings

One of the success factors for the realization of digital platforms for the exchange of production capacities is the speed at which offers can be generated. In addition, access to and use of the platforms should be as simple as possible. On the other hand, detailed information at machining process level is required for the generation of corresponding production plans on the platform in order to enable an efficient matching of requests and offers. In the present article, we propose to analyze technical drawings of the requested part using an artificial neural network to derive the machining processes. A major challenge here is multi-label image classification, since a technical drawing usually contains several machining processes. Furthermore, the occurring problem of the unbalanced dataset and the resulting improvement of the dataset via image augmentation is discussed. Finally, the performance of the developed model on the test data is demonstrated and a recursive method for improving the system during operation is described.