An Online Fault Detection and Remaining Life Prediction Method Based on SVDD for Rolling Bearings

In industrial applications, rolling bearings operate under high-speed and high-precision conditions, and their physical and mechanical properties change with prolonged operation. Traditional diagnostic methods rely on fixed models trained offline for online diagnosis, making it difficult to adapt to the changing characteristics of bearings over time. This article proposes a dynamically updated dual-boundary support vector data description (DDSVDD) anomaly detection method. A dual-boundary support vector data description (SVDD) based on slack thresholds is introduced to select more effective support vectors, constructing a classification hypersphere and enhancing the accuracy of bearing anomaly detection. Furthermore, a fusion strategy combining spatial and temporal weights is presented to achieve real-time dynamic updates of SVDD, improving the timeliness of bearing anomaly detection. Additionally, a lifespan online prediction method based on equipment characteristic evolution and center offset is proposed to define the model's endpoint update boundary, enabling precise lifespan prediction of bearing failure models. To validate the effectiveness of the proposed method, experiments were conducted using the Case Western Reserve University (CWRU) bearing dataset and the University of Cincinnati bearing dataset. The experimental results demonstrate the effectiveness and superiority of the DDSVDD in practical applications.