This paper presents a proof of concept approach for predicting geometric deviations of CAD designs after their production. This could help designers to recognize possible quality issues or deviations as early as possible. Several machine learning models are trained on designs and CT scanned as manufactured parts produced by different AM processes. Results show that the used models and features have minor impact compared to the composition of the training data. The approach demonstrates the feasibility of ML based deviation prediction to support earlier and more informed design decisions.

This paper presents exploration of early-stage gripper designs for robotic grasping of fish and evaluate prototype performance through load capacity benchmark tests. The research is motivated by the need for automation within the Norwegian fishing industry, and designing for Physical-AI based control methods. We explore several gripping technologies, including compliant fingers, pneumatic soft robotic fingers, standard suction, Bernoulli suction and freezing, and present four hybrid-design prototypes combining suction with compliant and pneumatic soft robotic fingers. We find that a hybrid design with suction and compliant fingers meet industrial design requirements by grasping a polyurethane fish model with a normal vertical load capacity of 215 N, and a lateral horizontal load capacity of 65 N.

This study explores how patent drawings, combined with image processing and Vision-Language Models (VLMs), can be used to analyse dominant designs and product variants. Using eyeglasses as a case study, the workflow collects, classifies, and crops drawings, colours functional components with SORA, and estimates areas from pixel counts. Validated by a mechanical engineer, the results show an average relative error below 10%, highlighting both the potential and limitations of using patent drawings and VLMs for modularity analysis, dominant design studies, and parametric CAD generation.