Multi-kernel Learning based Autonomous Fault Diagnosis for Centrifugal Pumps

Centrifugal pumps are fundamental instruments in many industry plants, whose continuously operation plays an important role in the production cycle. For improving production efficiency, autonomous pump fault diagnosis has been widely adopted by many enterprises and has also attracted great research attentions. Existing studies exploit machine learning algorithms for autonomous pump fault diagnosis, which generally needs human knowledge to select distinctive data features. To avoid the bias of human selection, this paper proposes a multi kernel learning (NIKE) based autonomous pump fault diagnosis method. It trains basic classifiers (BCs) by each feature and weightily combines the basic classifiers to form the combined classifier for fault diagnosis. An autonomous BC weighting algorithm is proposed, which trains the combination weights of the BCs autonomously. We show the NIKE based fault diagnosis method provide more accurate fault detection than the existing methods and without the need of human experts' interventions.