Abstract

Usually, more than one sensor is placed to collect vibration signals for reactor coolant pump condition monitoring. The traditional method of reactor coolant pump fault diagnosis does not make full use of the relativity of all vibration signals. In order to make full use of all vibration signals, multi-sensor data fusion is introduced to reactor coolant pump fault diagnosis and a universal reactor coolant pump fault diagnosis model is built up. The reactor coolant pump vibration data fusion diagnosis model is divided into three modules. The three modules are the data level fusion module, the BP (back-propagation) neural networks feature level fusion diagnosis module, the D-S (dempster-shafer) evidence theory decision level fusion module. The data level fusion module is to eliminate the disturbance and extract the feature information about reactor coolant pump faults. The feature information handled by the data level fusion module is used as the inputs of BP neural networks. The neural networks feature level fusion diagnosis module is composed by more than one BP neural networks in condition that the number of input nodes is too large. The feature information is divided into several troops and input into BP neural networks respectively. The outputs of neural networks serve as the basic probability assignment of D-S evidence theory. The D-S evidence theory decision level fusion module fuses the outputs of neural networks and gives the final fusion diagnosis result. The experiment results show that multi-sensor data fusion is successful and promising in reactor coolant pump fault diagnosis.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call