A novelty degradation assessment method for equipment based on multi-kernel SVDD

Support vector data description (SVDD) has been applied to performance degradation assessment for years. But single kernel may not describe the varying distribution very well. Multi-kernel learning (MKL) method was developed and proved to perform better than single kernel. Previous studies have been conducted to build up a fixed model, which takes the sample distance as the assessment index. However, different condition may have the same distribution in feature space. In this paper, we proposed a new robust method for bearing performance degradation assessment based on multi-kernel SVDD, and designed a new index with hyper-sphere radius. The experiment results show that the new index can reflect the degradation´s development exactly.