Title: A new online hybrid learning algorithm of adaptive neural fuzzy inference system for fault prediction

Authors: Zhanlong Du; Xiaomin Li; Qiong Mao

Addresses: Department of UAV Engineering, Mechanical Engineering College, Shijiazhuang, 050003, China ' Department of UAV Engineering, Mechanical Engineering College, Shijiazhuang, 050003, China ' Department of UAV Engineering, Mechanical Engineering College, Shijiazhuang, 050003, China

Abstract: Effective prognostic tools are crucial for maintenance to predict faults before systems are completely damaged and to ensure reliability. A new online hybrid learning method for adaptive neural fuzzy inference system (ANFIS) prediction model is presented in this paper based on square-root cubature Kalman filter (SCKF) and recursive least squares (RLS). SCKF and RLS are used to recursively optimise ANFIS nonlinear and linear parameters, respectively. Fault feature sensitive to system degradation process is selected as ANFIS predicting variable for fault prediction. The simulation results indicate that the higher forecasting accuracy or the lower computational complexity is obtained compared with other three online learning methods.

Keywords: neural networks; fuzzy logic; hybrid learning; nonlinear filters; fault prediction; square-root cubature Kalman filter; SCKF; recursive least squares; RLS; maintenance; chaotic time series; continuous stirred tank reactor; CSTR; adaptive neural fuzzy inference system; ANFIS; simulation.

DOI: 10.1504/IJMIC.2015.067716

International Journal of Modelling, Identification and Control, 2015 Vol.23 No.1, pp.68 - 76

Published online: 31 Mar 2015 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article