Vehicle fault detection and diagnosis combining AANN and ANFIS uri icon

Abstract

  • A new fault detection and diagnosis approach to deal with noisy measurements and correlated variables is presented. Using Auto-Associative Neural Networks (AANN) the correlation of the variables is learned from normal operating conditions data. Then, residuals can be computed with this AANN model. By comparing residuals against normal operating conditions thresholds, faults can be detected (and sometimes diagnosed) using a Non-Linear Principal Component Analysis. For noisy variables were previous procedure fails, a second sequential step based on several Adaptive Neuro-Fuzzy Inference Systems (ANFIS) is needed. The performance of the approach was validated with a vehicle model. Early results are promising. © 2009 IFAC.

Publication date

  • December 1, 2009