Digitalization and optimization of train infrastructure

Important results from the project

In the project, the possibility of implementing, for metrology, new analysis methods of time series has been investigated. The wavelet transform is used for noise reduction and is adapted for non-stationary time series. With continuous collection of measurement data, a detailed digital track profile can be built up later thanks to analysis of the consistency between time series through dynamic time warping (DTW).

Expected long term effects

The project has collected and analyzed accelerometer data. Förutom RISE-intern knowledge development around time series analysis and accelerometer data has also built an algorithm for noise reduction of non-stationary time series through wavelet transform and a way to study the consistency between time series and their aggregation through dynamic time warping. The algorithm should be well suited for the analysis of data collected in regular train traffic and can be seen as the main result of the project.

Approach and implementation

Data collection was done by moving the sled on the crash course back and forth over two sets of cracks, about three cracks each, and simultaneously logging accelerometer data. The sled was moved along the rail at the desired speed. The time series analysis algorithm performs noise reduction with the wavelet transform and time series matching and aggregation with dynamic time warping. The algorithm is written in Python 3 and, in addition to the standard libraries, Pandas, NumPy, Matplotlib, SciPy and PyWavelets have been used.