Fault detection and isolation for railway vehicle suspensions

The ability to detect and isolate component faults in a railway suspension system is
important for improved train safety and maintenance. An undetected failure in the
suspension systems can cause severe wheel-rail wear, reduce ride comfort, worsen
passenger safety and increase unexpected maintenance costs. Existing fault detection
methods are limited in several respects, such as effectiveness/sensitivity for fault
detection, or robustness to external condition changes. This thesis investigates a
model-less fault detection and isolation approach using cross correlation and/or
relative variance techniques, developed to overcome these limitations.
This thesis treats a conventional bogie vehicle with a symmetrical structure. Excited
by the track irregularities, the dynamics of the vehicle are studied under the normal
conditions, with an emphasis on the vertical and related motions of the bogies and
the carbody.
Two fault detection schemes employing data processing using data directly from
measurement are discussed. One uses cross correlation evaluation of the basic bogie
motions to detect component fault; the other takes advantage of the relationship
between the relative variances of the suspension accelerations.
Finally, the fault isolation schemes are assessed based on the comparison of fault
detection performances in different conditions. The proposed approach does not
require detailed knowledge of the vehiclelbogie and external track irregularities. The
effectiveness of the approach is verified by computer simulations in
Matlab/Simulink.