Self-supervised representation learning anomaly detection methodology based on boosting algorithms enhanced by data augmentation using StyleGAN for manufacturing imbalanced data

• DOI: 10.1016/j.compind.2023.104024
• Publication Date: 2023-10-09T09:03:55Z
• Authors: Yoonseok Kim; Taeheon Lee; Youngjoo Hyun; Eric Coatanea; Siren Mika; Jeonghoon Mo; YoungJun Yoo
• Keywords: Boosting; External Data Representation; Hyperparameter

This study proposes a methodology for detecting anomalies in the manufacturing industry using self-supervised representation learning approach based on deep generative models. The challenge arises from limited availability of data defective products compared with normal data, leading to degradation performance models owing imbalances. To address this limitation, we propose process that leverages Gramian angular field transform time-series into images, applies StyleGAN image augmentation anomalous and utilizes boosting algorithm classifier selection supervised learning. Additionally, accuracy before after augmentation. In experimental cases involving CNC milling machine wire arc additive proposed outperformed augmentation, resulting improved precision, recall, F1-score anomaly detection. Furthermore, Bayesian optimization hyperparameters further enhanced metrics. effectively addresses imbalance problem, demonstrates its applicability various industries.