Dissertation
Dissertation > Industrial Technology > Machinery and Instrument Industry > Mechanical operation and maintenance

Research on Weak Feature Extraction and Diagnosis Methods of ElectricMechanical Equipment

Author HeHuiLong
Tutor WangTaiYong
School Tianjin University
Course Mechanical Manufacturing and Automation
Keywords weak feature extraction cascaded bistable stochastic resonance method fusion embedded structure fault diagnosis
CLC TH17
Type PhD thesis
Year 2007
Downloads 470
Quotes 2
Download Dissertation

The complexity of vibration signals may come from the complexity of equipment structure, cooperation of multi-parts, diversity of excitation and fault sources, which brings difficulties to fault diagnosis. Aiming at the electric- mechanical equipment, this dissertation focuses on the research on weak feature extraction from complex signals and related diagnosis methods.The extraction of weak feature signal submerged in strong noise plays an important role in fault diagnosis. The non-linear filter characteristic of cascaded bistable stochastic resonance system (CBSRS) is revealed further in this dissertation. The results show that weak feature frequency components located in low frequency area can be extracted during the implementation of low-pass filter owing to the energy transfer mechanism from high frequency area to low frequency area by CBSRS. Metal cutting experiment and roller-bearings fault diagnosis examples demonstrate its practicability.The signals collected by each sensor are always the mixtures of multi-vibration sources caused by the coupling of different mechanical parts. Blind source separation (BSS) and independent component analysis (ICA) provide solutions to the separation of different sources. Due to the facts that measurement mechanical vibration signals are always corrupted with addictive background noise and the negative impact of noise is ignored in the existing blind separation arithmetic, a method of CBSRS de-noising is presented and applied to the noisy ICA problem. A simulation experiment proves its feasibility.Empirical mode decomposition (EMD) is a time-frequency analysis method and can adaptively decompose the signal into several intrinsic mode functions (IMFs) according to its characteristic time scale. Support vector machine (SVM) is a new machine learning method and can be taken as a classifier. According to the fact of different fault with different complexity, a diagnosis method, Renyi entropy regarded as complexity criterion, is put forward based-on EMD and SVM in this dissertation. The method takes the Renyi entropy of several IMFs as the characteristic vector and inputs them into SVM for training and recognition. A roller-bearings fault diagnosis example demonstrates its practicability.The terminal portable instruments play a vital role in the condition monitoring and fault diagnosis system. The dissertation presents a portable data-acquisition and analysis instrument on the basis of embedded operating system (OS) with enhanced write filter (EWF), which can effectively enhance the instrument’s OS stability. Because of the lack of effective communication mechanism between conventional condition monitoring system and plant management system, an embedded condition monitoring system for plant management is proposed under internet framework, which realizes the information fusion of condition monitoring system and plant management system.

Related Dissertations
More Dissertations