AI‐Based Segmentation for Corrosion Risk Prediction in Carline Designs

ABSTRACT With shortening development cycles, digital development rises in importance. Therefore, we present an approach to segment digital carlines based on their corrosion risk using digital data sets. A typical carline is described via STL‐Files that accumulate to approximately 25 million triangles. While physics‐based simulation models exist capable of predicting the onset of corrosion in local geometry settings, an application to a complex surface mesh would be prohibitively expensive to compute. This calls for data reduction techniques that reduce the number of triangles by identifying areas prone to corrosion, as well as those that are protected by measures such as adhesives. Therefore, the implementation of corrosion‐prone areas of a body‐in‐white part, as presented in Waibel et al., is extended by implementing a design measure segmentation for adhesive pipes. This work introduces a method to predict corrosion‐protected areas for flanges with adhesives by implementing a feature extraction and a transfer learning‐based workflow for data‐efficient training. The results of the implementation are the validated prediction on body‐in‐white parts with high precision.