Monitoring of Agglomeration in Fluidized-Bed Reactor Based on Voiceprint Features Recognition with Unbalanced Samples

Abstract

When acoustic emission detection technology is applied to detect the agglomeration in fluidized bed reactors (FBRs), the collected acoustic emission samples are usually non-stationarity and unbalanced, making it difficult to extract stable and separable classification features. In this study, the voiceprint features of collected acoustic emission signals were extracted with the Mel Frequency Cepstrum Coefficients (MFCC) and Linear Prediction Cepstrum Coefficients (LPCC). Extracted voiceprint features of LPCC and MFCC were fused with RelieF algorithm to form the stable R-LPMFCC feature, which were then compressed with principal components analysis (PCA) as input data for classification. The cost factor and GINI index-based decision-making calculation were introduced to the Adaboost algorithm to significantly improve its accuracy and F-score when classifying unbalanced samples. The comparative experimental results in a fluidized-bed pilot plant verify the effectiveness and feasibility of the proposed method.