A Knowledge Management System to Support Design for Additive Manufacturing Using Bayesian Networks

Abstract

Design for additive manufacturing (DfAM) is gaining increasing attention because of the unique capabilities that additive manufacturing (AM) technologies provide. While they have the ability to produce more complex shapes at no additional cost, AM technologies introduce new constraints. A detailed knowledge of the AM process plays an important role in the design of parts in order to achieve the desired print result. However, research on knowledge management in this area is still limited. The large number of different AM processes, their individual sets of critical parameters and the variation in printing all contribute to a high level of uncertainty in this knowledge domain. Applying AM at the early stages of design projects introduces another source of uncertainty, as requirements are often not well defined at that point. In this paper, a knowledge management system using Bayesian networks (BNs) is proposed to model AM knowledge in cases where there is some uncertainty and fill the knowledge gap between designers and AM technologies. The structure of the proposed model is defined here by introducing the overview layer and detailed information layer. In each layer, different types of nodes and their causal relationships are defined. The system can learn conditional probabilities in the model from different sources of information and inferences can be conducted in both forward and backward directions. To verify the accuracy of the BNs, a sample model for dimensional accuracy in the fused deposition modeling (FDM) process is presented and the results are compared with other methods. A case study is provided to illustrate how the proposed system can help designers with different design questions understand the capabilities of AM processes and find appropriate design and printing solutions.