Fault diagnosis is a crucial area of research in the industry due to diverse operating conditions that exhibit non-Gaussian, multi-mode, and center-drift characteristics. Currently, data-driven approaches are the main focus in the field, but they pose challenges for continuous fault classification and parameter updates of fault classifiers, particularly in multiple operating modes and real-time settings. Therefore, a pressing issue is to achieve real-time multi-mode fault diagnosis for industrial systems. To address this problem, this paper proposes a novel approach that utilizes an evidence reasoning (ER) algorithm to fuse information and merge outputs from different base classifiers. These base classifiers are developed using a broad learning system (BLS) to improve good fault diagnosis performance. Moreover, in this approach, the pseudo-label learning method is employed to update model parameters in real-time. To demonstrate the effectiveness of the proposed approach, we perform experiments using the multi-mode Tennessee Eastman process dataset.

翻译：故障诊断是工业领域的重要研究方向，由于工况具有非高斯、多模态和中心漂移等特性。目前，数据驱动方法虽是该领域的研究热点，但在连续故障分类和故障分类器参数更新方面仍面临挑战，特别是在多工况和实时场景下。因此，实现工业系统的实时多模态故障诊断成为亟待解决的问题。针对这一难题，本文提出一种新型方法，利用证据推理算法融合来自不同基分类器的信息与输出结果。这些基分类器基于宽度学习系统构建，可有效提升故障诊断性能。此外，该方法采用伪标签学习技术实时更新模型参数。为验证所提方法的有效性，我们采用多模态田纳西-伊斯曼过程数据集进行了实验。