Feature Extraction Method of Fluidized Bed Agglomeration Based on ReliefF and PCA

Agglomeration of polymer in fluidized bed reactors (FBRs) can hinder the industrial production seriously. In order to monitor agglomerations, the acoustic method was introduced, and the ReliefF based on the principal component analysis (PCA) was proposed to extract the feature of acoustic signals. Firstly, the time-domain and frequency-domain features of acoustic signals generated by reactant particles impinging on the wall of the FBR were analyzed, and a high-dimensional feature vector was found to distinguish normal and abnormal signals. The PCA method was used for removing the correlation among the feature matrixes of training data, and the cumulative weight metrics based on ReliefF was designed for the selection of feature. Besides, a low-dimensional feature vector was selected for fault modeling. The proposed method was applied to a polyethylene pilot plant. Experimental results show that the method can effectively improve the detection accuracy of agglomeration fault, and improve the reliability of the acoustic method.