A Multi-Label Method of State Partition and Fault Diagnosis Based on Binary Relevance Algorithm

Prognostics and health management (PHM) is an important topic of rolling bearing, so partition of healthy stage and different degree of fault states are equally important as fault types. This study uses a multi-label learning method, Binary Relevance (BR) algorithm, for fault diagnosis of bearing data based on both of states partition and fault types judgment. The Binary Relevance algorithm simplifies the classification process by transforming the multi-classification problem into multiple binary classification problems. Using extreme learning machine ensures the classification speed and effect. By comparative experiments on XJTU-SY bearing datasets, the effectiveness of our method is proved to be superior.