Multistate Reliability Analysis of Solid Oxide Fuel Cells Using Automatic Spectral Clustering and Neighborhood Rough Sets

Abstract

To optimize the reliability of solid oxide fuel cells (SOFCs), an analysis method based on automatic spectral clustering and neighborhood rough sets is proposed in this paper. This is the first application of multi-state reliability theory to SOFC systems. Firstly, a state partitioning method based on automatic spectral clustering is proposed to partition the operational data into different classes. Then, the feature extraction method based on the neighborhood rough set is used to find the most sensitive variables to system state changes. Finally, the accuracy of state segmentation and feature extraction is verified by training state classifiers. In addition, the generality of the method is verified by migrating it to another pure hydrogen-fueled SOFC system. The results show that the state segmentation method successfully partitions the electrical characteristics into 10 states. Even the data in the early stage of failure can be well segmented. Moreover, the feature extraction method extracts 6 variables that are most sensitive to state changes, which reduces the number of the variables by 85.7%. The state classifier can achieve over 94% correct state recognition rate within the 30s. Meanwhile, the method has good generality and transferability.