Experimental Study on Classification Method of Leakage Signal in Industrial Boiler Pipeline

Abstract

In this paper, a new acoustic emission detection technology for boiler pipeline leakage using random forest and KNN classifier is proposed. The signal parameter index is processed as feature vector, which overcomes the shortcoming of the traditional method which requires a large number of sample data for training and classification. First, the characteristic parameters of boiler pipeline leakage acoustic emission signal are extracted, and then the extracted characteristic parameters are input into random forest and KNN classifier as feature vectors for classification processing. Eight indexes including amplitude, ringing count, duration, energy, rise count, rise time, RMS voltage and average signal level are selected and input into the classifier as feature vectors. As a diagnostic for pipeline leak classification. The experimental results show that this method is effective and feasible in pipeline leak diagnosis, and the feasibility of applying random forest and KNN algorithm to the classification of acoustic emission signals in pipeline leak detection is verified.