Two-class model based on nonlinear manifold learning for bearing health monitoring

The health monitoring and management applied for modern industrial machinery is a trend in intelligent industrial production. In this context, a novel health monitoring method based on a two-class model and nonlinear manifold learning is proposed to assess bearing performance degradation timely and reliably. This method does not need much historical operation information and can enhance the practicability in real industrial application. The two-class model is constructed by involving healthy data and monitored data together. The manifold learning is then implemented to extract the sensitive features for good class separable measure. According to within-class and between-class scatter measure, a discriminant factor is introduced to assess the distribution of the monitoring data as compared to the healthy data, which can reflect the bearing performance degradation. In the experiments for monitoring bearings in full cycle life, the proposed health indicator has provided a much more accurate degradation path and rather monotonic profile as compared to other classical indicators. Results indicate the potential of the proposed two-class model as an effective method for bearing health monitoring.