PSI - Issue 4

Uwe Oßberger et al. / Procedia Structural Integrity 4 (2017) 106–114

109

Author name / Structural Integrity Procedia 00 (2017) 000 – 000

4

Fig. 3a is a picture of a ÖBB 4024 train advancing the instrumented turnout. An arrow indicates the position of the check rail (see also Fig. 1 and Fig. 2b). Such passenger train type, run by the Austrian Railways ÖBB, passes the instrumented crossing several times per day. For this passenger train the arrangement of driven/ undriven axles remains always the same and the actual axle load can change based on the passenger loading of the train. The forward simulations by Pletz (2012, 2013) and their validation by Ossberger (2013) clearly showed that for a fixed axle load, fixed unsprung / sprung mass ratio of the vehicle and fixed wheel geometry the remaining influencing parameters on the strain signals generated at the crossing nose by the passing axles are the train speed, the crossing geometry and the bedding. As the train speed in this section showed only little variations (upcoming railway station next to test site), all strain signals produced at the crossing nose by the transition of the first two axles of the ÖBB 4024 passenger trains showed similar strain results.

Fig. 3a. (a) Picture of a ÖBB 4024 train advancing the instrumented turnout (b) Strain gauge signals measured on a strain gauge at the bottom of the crossing nose while a ÖBB 4024 was passing, each axle passing the crossing nose generated a clearly distinguishable strain maximum; the same graph has already been published by Kollment et al. (2016). Fig. 3b shows the strain signal recorded with one of the strain gauges placed on the bottom of the crossing nose near the transition region. The background colors indicate that the strain signal for each passing train can be segmented into a signal before, during and after the passing, after this segmentation, signal processing applied to the segments allows to extract various bits of information from the strain data. Based on the strain database from a single (!) strain gauge placed at the bottom of the crossing nose, Kollment et al. (2016) developed a signal processing chain that enables to:

 identify the direction and speed of the passing train  identify different types of trains  compare data from trains of the same type but at slightly different speeds  identify and exclude trains with irregular wheels  compare the signal generated by a train with a preceding passing of the same train type

Fig. 4 sketches a signal processing chain to compare the strain signals at the crossing nose for recurring transits of the same train type, i.e. an ÖBB 4024 train. Fig. 4a shows the original signals recorded from 6 different passings of the ÖBB 4024 train (all of them in forward direction on the main route) within a short period (< 1 month). As the trains pass the crossing nose at different speeds and/or acceleration the signals had to be transformed before they could be directly compared. Kollment et al. (2016) used a dynamic time warping algorithm combined with local polynomial resampling to derive the characteristic load pattern for the 6 ÖBB 4024 trains. Fig. 4b shows a superposition of the corresponding transformed signals. Fig. 4c gives the consensus curve of the 6 signals on the top and shows the 50% confidence interval below.