Deep Adversarial Learning System for Fault Diagnosis in Fused Deposition Modeling with Imbalanced Data

Abstract

Fused deposition modeling (FDM) has been widely promoted as an emerging additive manufacturing technology. With the growing demand for its commercialization, the requirements for the quality of products are increasing. Quality control and fault diagnosis in the manufacturing process are becoming prominent. However, most research focuses on monitoring the process, while few studies diagnose the faults to find their causes, especially concerning the drift of process parameters. The domain-shifting problem occurs when process parameters drift in FDM process and it largely influences the diagnosis performance of a trained model. To fill this gap, this paper proposes a deep adversarial learning system for fault diagnosis in the FDM process, based on captured upper layer images during the manufacturing process. Conditional generative adversarial network is adopted to augment the original dataset and solve the between-class data imbalance problem. As for domain-shifting problems, this research utilizes a domain adversarial neural network to process features from different domains, so as to identify the process parameters with drifting values in the FDM process. A laboratory case study verifies the effectiveness and accuracy of the proposed method in diagnosing.

Keywords: Additive manufacturing, Adversarial modeling, Fault diagnosis, Fused deposition modeling, Imbalanced data