Our client is a leading provider of railcars to the US/Canada rail industry, owning over 200,000 railcars and 130 wayside sensors to detect wheel failures.
The sensors tend to exhibit aberrant behavior during events such as flooding, extreme weather, and mechanical failure. A single faulty sensor could lead to the removal of 200-500 healthy wheels, resulting in more than USD 1 million in avoidable costs. They needed an early warning system which could predict sensor failure and also an advanced reporting system to supplement it.
- Large data volumes – 300,000 rows every day; 100 GB data over three years
- Alerts from the sensors varied over time and across detectors due to seasonal variations in traffic, weather patterns, and car usage
- Integration of the analytics solution with existing infrastructure, so that the results are actionable without delays
- Organized and processed site level data – the number of wheels which passed through the site, the number of alerts that were generated by the sensors, the speed at which the wheels passed through, the type of railcar that passed through, etc.
- Created a normalized metric called Alerts per Wheel (APW) – since the number of wheel-sand alerts at a particular site depends on the traffic
- Accounted for seasonality and temperature related effects because the number of alerts also depended on seasonality, with more alerts expected during winter than summer and during colder winters than during normal winters
- Developed benchmarks for each site and scored each sensor against the benchmark to flag potentially faulty ones
- Developed an interactive UI with drill-down capabilities. Also, built a weekly email notification system to alert relevant personnel
- The deployed centralized solution saves the client USD 6 million in costs every year
- The success of this project led to the American Association of Railroads to adopt the same solution and expand it to an industry-wide implementation.