•  
  •  
 

Applied Mathematics & Information Sciences

Author Country (or Countries)

China

Abstract

Friction fault diagnosis of rotating machinery based on acoustic emission (AE) technique is a research hotspot in recent years. The rotating machinery will produce multi-source noise during the operation process, so how to correctly identify the friction acoustic emission signals has become a key factor for accurate diagnosis of the fault. In this paper, it proposes a Gaussian mixed model (GMM) based on an embedded time delay neural network (TDNN) to identify friction acoustic emission signals. It comprehensively utilizes the advantages of the learning ability of time delay neural network about data structure and data distribution presentation capability of Gaussian mixture model. Time delay neural network fully exploits the time-ordered of eigenvector set, makes the maximum likelihood probability more reasonable which needs to assume that the variables are independent of each other through the transformation of the time delay network and uses them for the training as a whole with the criteria of maximum likelihood (ML) probability. During the training process, the parameters of Gaussian mixture model and neural network update alternately. The average amplitude, maximum amplitude, amplitude dynamic range, the Hurst exponent and approximate entropy (ApEn) of friction acoustic emission signals are selected as the characteristic parameters of fault recognition and these five parameters constitutes the input parameters vector of the identification model. Through the verification the AE signals of different friction states collected on the rotor test bed, the experimental results show that the identification method of rotor friction acoustic emission signals of Gaussian mixed model based on embedded time delay neural network is an effective mean of identification with high recognition efficiency.

Suggested Reviewers

N/A

Digital Object Identifier (DOI)

http://dx.doi.org/10.12785/amis/080254

Download

Share

COinS