NEURAL NETWORK BASED ON WAVELET PACKET-CHARACTERISTIC ENTROPY FOR FAULT DIAGNOSIS OF DRAFT TUBE

Fengqin Han

Proceedings of the Csee · 2005 · 20 citations · 0 references

Concepts

Abstract

To diagnose accurately vortex rope in draft tube of hydraulic turbine, this paper presents a new method of fault diagnosis based on wavelet packet and neural network. It adopts three-layer wavelet packet to decompose the monitored pressure pulsation signal of draft tube, extracts WP-CE (Wavelet Packet-Characteristic Entropy), and constructs WP-CE vectors of signals, then takes those vectors as fault samples to train three-layer BP(Back Propagation) neural network, finally realizes intelligent fault diagnosis. The practical example shows the trained BP neural network can diagnose the fault of draft tube and resort it by fault severity. This method develops a new direction for the fault diagnosis of hydraulic turbine.