The rolling stock company was unable to predict engine failures in winter months.

Predicting engine failures early would allow the rolling stock owner to carry out the necessary maintenance when the train was scheduled to be in the depot and avoid the penalty payments for delay or disruption.

Through our empirical approach, we set-up a good understanding of the train data and maintenance information. Our platform is able to cope with over 600 sensor parameters acquired every few seconds. We analysed 100 billion data points.

We were able to categorise historical engine shutdown into different root causes. This enabled us to apply focused methods for different categories of engine shutdown.

We added terrain slope analysis by using GPS co-ordinates available in the orbita data and slope information via open public data. This increased the effectiveness of our methods in predicting engine failures.

Amygda was able to predict engine failures between 2 and 11 days before failure. This enabled the rolling stock owner to optimise train operations and reduce maintenance costs.

We were even able to categorise to a high level of accuracy the sub-system within the engine that was at risk of failure.