CM: Machine-Learning Driven Decision Support in Design for Manufacturability

Iowa State University, Ames IA

Investigators

Abstract

Traditional design and manufacturing relies on the experience and training of the designer to create a component with manufacturable features. However, even after careful design, the as-manufactured part might differ from the as-designed part. In addition, the inclusion of certain features might significantly increase the manufacturing cost. For example, the inclusion of a thin feature might necessitate the use of complex jigs or fixtures to prevent the flexing of the part during machining, which increases manufacturing time and cost. This problem is also encountered in additive manufacturing, where there is no body of knowledge regarding design rules that will reduce manufacturing defects. This project aims to address this challenge by developing computer-aided design tools that can identify difficult-to-manufacture features using machine learning. The process of identification of the source of infeasibility in manufacturing in a complex part is a challenging task, even for an experienced designer. Therefore, the use of machine learning could potentially play a critical role by detecting non-intuitive patterns from examples of feasible and infeasible parts, and identifying the source of infeasibility. The results of the machine-learning framework will be used to build a decision support framework that can interactively identify manufacturability concerns during the design process and present design modifications interactively to the designer. Finally, the multidisciplinary components of the project will be integrated into a larger educational effort to offer students a solid foundation in the critical interdisciplinary area of cyber-enabled manufacturing. The objective of this project is to create a design for manufacturability tool that uses machine learning to identify difficult to machine or manufacture features in a computer-aided design model and suggest changes to the non-manufacturable features. The novelty of this research is the use of machine learning in a computer-aided design and manufacturing environment, making it accessible to designers using a familiar design interface. The research team will develop tools for loading existing models of parts and performing virtual machining simulations to create a digital voxelized representation of the as-manufactured part. The original as-designed part will also be converted to a voxelized representation that will be suitable for machine learning. The machine-learning framework will be trained using multiple machining simulations and will classify feasible and infeasible designs by learning from positive and negative examples. Furthermore, the machine-learning framework will be used to present alternative feasible designs to the designer.