Methodical Approach to Reducing Design Time by using Neural Networks in Early Stages of Concept Development

Modern companies often face various challenges in concept development of products or systems. Design engineers prepare initial concepts as 3D models. These are then simulated by computational engineers. If requirements are not met, this necessitates an iterative process that runs between the design and computation departments until a valid concept is created. Design methods such as topology optimization are often used here. The upcoming result is then attempted to be adapted to certain manufacturing processes. These iteration loops can sometimes take a very long time, since the model construction and structural optimization generate large computational efforts. The present work shows on an example a methodical approach, which represents a first proof of concept, to solving this problem, including a description of methods and techniques, as well as possible problems in a detailed analysis concerning training data for neural networks and their abstraction capabilities. It is evident that additional research work needs to be conducted for further utilization in order to address all arising questions.