A time series can take the bumps out of your train ride
Thanks to sensors designed by the Monash Institute of Railway Technology, a train’s dynamic responses can be continuously measured.
Thanks to sensors designed by the Monash Institute of Railway Technology, a train’s dynamic responses can be continuously measured. (A big response generally indicates a fault in the track.) When collected at regular intervals, this data forms what’s called a time series. By analysing multiple train trips, railway companies can learn how track conditions evolve and schedule maintenance accordingly, making your commute a whole lot smoother.
That’s the thrust of the research of Chang Wei Tan, Monash Computer Science PhD student. He explores how to predict through time series the effectiveness of track maintenance. Railway companies can then use this information to decide if and when to do it.
Rather than apply traditional statistical methods, Chang Wei relies on data patterns before maintenance to determine if such work will be effective or ineffective for a given track location. In this way, rainfall, soil condition, the type of surface underlying the track (e.g. bridge, swamp or rock area) and train loading will be taken into consideration.
To address the exponential growth of big data, Chang Wei has committed himself to developing faster and scalable algorithms. And he hopes to achieve this without resorting to dataset sampling, which sacrifices accuracy. Imagine a sensor recording data every minute. That's 525,600 data points per year. And that’s only for one sensor, he says. It would take ages to process and analyse all these data with current state-of-the-art methods.
Such technology has applications well beyond railways. It could be used to predict the maintenance effectiveness of anything. It can also be used to analyse the evolution of the Earth using satellite images which can assist in developing fire spread and city pollution absorption models amongst other things. Most of the data in the world can be represented in the form of a time series, claims Chang. I think my research can be used in almost every domain that requires lots of data analysis, ranging from medicine and finance to engineering and cybersecurity.
To underscore this potential, Chang gives a perfect example: Just imagine what a time series of gene expression might teach us about cancer treatment. Well, that could take the bumps out of something even more vital than a train ride.