Time series methods applied to failure prediction and detection

Abstract

Point mechanisms are critical track elements on railway networks. A failure in a single point mechanism causes delays, increased railway operating costs and even fatal accidents. This paper describes the development of a new robust and automatic algorithm for failure detection of point mechanisms. Failures are detected by comparing what can be considered the ‘expected’ form of signals predicted from historical records of point mechanism operation with those actually measured. The expected shape is a forecast from a combination of a VARMA (vector auto-regressive moving-average) model and a harmonic regression model. The algorithm has been tested on a large dataset taken from an in-service point mechanism at Abbotswood Junction in the UK. The results show that the faults can be predicted and detected.