Detection of wheel flats using stochastic wave analysis

Abstract

In railway dynamics, irregularities and defects of wheels and track cause an increase in dynamic forces. One type of imperfection is a wheel flat --- a flat zone on the wheel tread. In this study, the aim has been to develop methods for detection of wheel flats, given a time series.

Field measurements from a train where one wagon had an artificial wheel flat have been analysed. A filtering of the process was performed by using concepts and theory from image analysis and stochastic wave analysis.

In an approach where each wheel was treated separately, the process was first convolved with a Gaussian kernel and rainflow cycles in the resulting signal were analysed. Two parameters were found necessary: one bandwidth parameter to determine the degree of smoothing and one threshold parameter. In the filtered signal, the number of local extrema will determine whether there is a wheel flat or not. An alternative way is to treat the passage of the whole train and afterwards visualize the result in a diagram.