Online Minority Cluster-Informed Semi-Supervised Random Vector Functional Link Network for Multi-Mode Intermittent Fault Diagnosis

Abstract

Industrial intermittent fault diagnosis is crucial for maintaining efficient and safe production processes. However, existing methodologies often fail to account for practical sample imbalance constraints encountered in multi-mode scenarios. In this paper, an online minority cluster-informed semi-supervised random vector functional link network, termed OMIS-RVFL, is proposed to tackle these challenges. It incorporates a minority-cluster informed strategy, employing dimensionality reduction and minority prioritization to enhance linear separability of samples in transitional conditions and improve identification of minority instances. Multiple experiments are conducted using the multi-mode Tennessee Eastman process datasets. Experimental results verified that the effectiveness of the proposed OMIS-RVFL.