Variational Bayesian modified adaptive Mamdani Fuzzy Modelling for use in condition monitoring

Zhang, Yu, Chen, Jun and Bingham, Chris (2015) Variational Bayesian modified adaptive Mamdani Fuzzy Modelling for use in condition monitoring. In: IEEE international conference on Computational Intelligence and Virtual Environments for Measurement Systems and Applications (CIVEMSA), 12-14 June, 2015, Shenzhen, China.

PDF - Whole Document
Available under License Creative Commons Attribution.

Item Type:Conference or Workshop contribution (Presentation)
Item Status:Live Archive


The paper proposes a new Adaptive Mamdani Fuzzy Model (AMFM) based system modelling methodology that improves on traditional Mamdani fuzzy rule based system (FRBS) techniques through use of alternative membership functions and a defuzzification mechanism that is ‘differentiable’, allowing a back error propagation (BEP) algorithm to refine the initial fuzzy model. Moreover, a variational Bayesian (VB) method is applied to simplify the results via automatic selection of the number of input rules so that redundant rules can be removed for the initial modelling phase. The efficacy of the proposed VB modified AMFM (VB-AMFM) approach is demonstrated through experimental trials using measurements from a compressor in an industrial gas turbine (IGT).

Keywords:Adaptive Mamdani fuzzy model, back error propagation, variational Bayesian Gaussian mixture model, industrial gas turbine
Subjects:G Mathematical and Computer Sciences > G700 Artificial Intelligence
G Mathematical and Computer Sciences > G510 Information Modelling
Divisions:College of Science > School of Engineering
Related URLs:
ID Code:17828
Deposited On:09 Jul 2015 09:49

Repository Staff Only: item control page