Research Of Mechanical Vibration Signal Blind Separation Based On Neural Network

The vibration signal is one of the detection items must be checked as in the normal operation, run of trial run and maintenance process of the aircraft and spacecraft(especially the engine) and vehicle motion structure. In recent years, our country gained many important achievements in the research of the method in vibration signal processing and analysis.The research root in the study project of run of trial run test on aeroplane motor when I practice in the Siemens cooperated with a space research unit. I am mainly responsible for the test of multi-source excitation vibration response of aeronautical motor casing.The researches having been studied are always in the condition of linear mixed model, nonlinear mixed model is not discussed too much.This paper has made blind source separation based on the neural network be applied to engine vibration signal. For the linear mixed model, this paper add momentum term to improve the neural network, and make the blind source separation effect better, for nonlinear mixed model, this paper use RBF neural network that has good nonlinear forced nearly ability to make the mixing signals be effectively separated.This makes the engine vibration signal blind separation technology is more perfect,more comprehensive,and this is just the innovation of this paper.This paper extracts the physical characteristics of the engine vibration signal effectively directed against the complexity of the engine itself and the diversity of its work environment,to analysis the signal logically and define the cost function of blind separation process.Based on maximum likelihood estimation algorithm of single-layer feedforward neural network algorithm is applied to the vibration of the engine blind signal separation in the linear mixed model, adding momentum item after the system well limit network falling into local minimum, avoid oscillation, to further enhance the convergence rate of weights.RBF neural network correlation algorithm is applied to the vibration of the engine blind signal separation of the nonlinear mixed models, uses minimum mutual information algorithm to statistics the independence of signals, and uses fuzzy c-means clustering method to estimate the center μ of the implicit kernel function.The RBF neural network can be trained for separating the nonlinear mixed models.The two kinds of network structure of engine vibration signal blind separation algorithm is applied to an air motor casing vibration multi-source excitation response test. And through the MATLAB simulation, the result is proved the feasibility, effectiveness and accuracy of the two algorithms.This paper takes a new model and sight of the blind separation direction of research in the field of application of engine vibration is discussed.