Feature Extraction and Recognition of Ventilator Vibration Signal Based on ICA/SVM

Abstract

Ventilator vibration signal is usually mixed with some signals and shows strong nonlinearity, nonstationarity and nonGaussian. It presents a great challenge to feature extraction and recognition. We applied the independent component analysis (ICA) to ventilator vibration signal analysis, used FastICA algorithm to get a group of independent variables with the useful feature information, adopted residual self-information (RSI) to compress further for the group of independent variables, and chose the larger RSI to form the new estimating component. And then we used support vector machine (SVM) to find the ventilator healthy pattern and/or the ventilator fault pattern. The experiment result shows that by using the methods above the correct identification rate of ventilator healthy and fault state reaches 100%. Keywords-independent component analysis(ICA); residual self information(RSI); support vector machine (SVM); fault diagnosis

Fault Feature Extraction Method of Large Rotating Machinery

“…So we can translate the problem of maximizingJ(w) into the problem of maximizing;[<(= > ?)]. That is to say, we should find a separation matrix w so each independent component has maximum non-gaussianity and attains the best result of separation [5].Hyvarinen calculate w in iteration Eq.8.…”