Compound rub malfunctions feature extraction based on full-spectrum cascade analysis and SVM

Abstract

The acceleration signals of operational rotor vibration provide a lot of information about its running behaviour. The acceleration signal features of identification and extraction in the process of speed change are important for the fault diagnosis of rotating machinery. The full-spectrum cascade analysis of rotating machinery vibrations is an efficient method that enables the symptoms of some special types of fault (especially for rub) to be clearly detected. Some typical compound rub malfunctions have been researched by experiments in this study. Acceleration signals have been received by the experimental apparatus and analysed by full spectrum. The abrupt changes in surging acceleration signals of rotor malfunctions can be detected and their fault feature spectra shown in full-spectrum cascade plots. The full-spectrum experimental data are applied to the support vector machine (SVM) training to be classified. The results indicate the potential and feasibility of this approach for the diagnosis of rotor malfunctions. The full-spectrum cascade plot can enhance the feature information for the knowledge base of the rotating machinery rub fault diagnosis system and is of great significance to diagnose compound rub faults in a rotor more accurately.