Intelligent Diagnosis on Flexible Material Windings in a Double-Suction Centrifugal Pump With Boosting Capsule Network (BCN)

The double-suction centrifugal pumps (DCPs) are important hydraulic machinery for agricultural, industrial, and urban life. However, they are suffering a high risk of flexible winding material due to the increasing pollution of rivers. Unfortunately, this serious problem has not yet received the attention of researchers. To solve this foreseen issue, a novel deep-learning (DL) model, namely the boosting capsule network (BCN), is proposed to solve the weaknesses of the convolutional neural network (CNN) and capsule network (CN). The boosting part fully refines the kernel thought of CNNs to achieve an optimal tradeoff between time-consuming and accuracy of diagnostic tasks. The computational burden and generality of BCN are dramatically reduced and improved from the traditional CN. Experimental studies, including different flow rates under different rotational speeds, are conducted to analyze the diagnostic accuracy, time consumption, and layer visualization. The results show that BCN achieves the best generalized diagnostic results for different operational conditions and data types than the compared models.