Research On Post Nonlinear Blind Source Separation Algorithm And Its Application

Rotating device, which mainly takes on power transporting and energy converting, is one of the main vibration sources. The devices, which are complicated in structure, are always rotating at high speed when operating, so that they are easily to be ageing and it will result in serious consequences when they break down. Vibration signals, closely connected with the operating conditions of rotating device, hold an important position in equipment condition monitoring and fault diagnosis of nuclear power plants. In actual measurement, vibration signals are complex mixed signals coming from multiple unknown vibration sources, which makes vibration source identification a significant procedure before analysis to distinguish the sources of the mixed signals.Blind Source Separation, BSS for short, is an effective method to separate source signals from monitored mixed signals based on some conditions and assumptions with both source signals and the mix approach between sources unknow. In this article, BSS is studied to create a new effective method to recognize the vibratory noise sources of the multi-stage centrifugal pumps. The characteristics of vibration signals generated by complex mechanical equipment such as multi-stage centrifugal pumps are not quite clearly known, and they are mixed complicatedly into the observation signals collected by the monitoring systems. BSS model,independence criteria and learning algorithm are well studied and non-linear mixed-signal are chosen to be research foundation. Two linear BBS algorithms, JADE-SOBI joint algorithm and FastICA algorithm based on negative entropy are studied and compared. JADE-SOBI joint algorithm is chosen to apply to the linearization of post nonlinear BSS. Two post non-linear BSS algorithm are studied, including Multilayer Perceptron (MLP) based on minimum mutual information and geometric algorithm. Both algorithms are implemented through MATLAB programming and used to separate non-linear mixed signals.Sectional horizontal multi-stage centrifugal pumps are studied as research object according to pump type and layout feature of the nuclear power system. Static and dynamics analysis are done through finite element analysis to obtain the vibration characteristic and signals of multi-stage centrifugal pumps. 3D modeling is done by Pro/Engineer and the analysis is completed through ANSYS Workbench. Maximum stress and strain, first six modal natural frequencies and vibration conditions of main components during the operation of pumps are achieved. Measuring points are arranged properly to obtain vibration simulation signals. Mixed signals are achieved based on post non-linear model and separated successfully through non-linear BBS algorithm. Considering the simulation separation test process and result, the advantages and disadvantages of both algorithms are analyzed and improvement orientations are pointed out.The study of vibration source identification method based on BSS technology, to some extent, makes up for the defects of conventional spectra analysis in machinery condition monitoring and fault diagnosis, provides a reliable basis for the fault diagnosis of machinery in nuclear plants and helps to improve operational safety.