Fluidized Bed Agglomeration Diagnose Based on Wavelet Packet Entropy and Gaussian Test

Sound waves in audible range are used for agglomeration monitoring and early warning of fluidized bed for the ethylene polymerization. According to the energy distribution in the frequency domain and the shift characteristics of the power spectral centroid of the normal and agglomeration signals, the sampling rate and signal pre-processing approaches are determined. Energy ratios of each wavelet packet frequency sub-band is calculated with wavelet packet decomposition and then the wavelet packet entropy is calculated based on the energy ratios and is taken as voiceprint of a frame signal. Diagnosis of fluidized bed agglomeration is realized by testing the Gaussian of the wavelet packet entropy within a determined diagnosis period. Compared with the agglomeration diagnosis approaches based on pattern recognition, the proposed approach overcomes the defections of susceptible to fail alarm and missing alarm caused by insufficient fault samples and weak generalization capability of diagnosis model. The approach is simple and easy to be implemented. The effectiveness and reliability of the proposed method is verified with experiments performed on a pilot plant. Agglomeration early warning can be implemented 18-45 minutes ahead compared with the existing approaches based on pressure and temperature monitoring.