RAIL SWITCH PREDICTIVE MAINTENANCE (SPM)
The goal of this project was to convert sensor readings into actionable insights, allowing rail infrastructure managers to reduce their maintenance costs and improve asset availability, thus helping operators to increase their punctuality and avoid unpleasant train delays.
We consider rail switches being one of the most relevant applications of our predictive maintenance know-how in a rail infrastructure domain due to the following two facts:
- By being itself an advanced data processing layer, Logicx SPM reuses already existing sensor data collection & storage systems, which makes Logicx SPM introduction a lite and cost efficient project
- According to our research, maintenance of switches, on average, takes up to one third of maintenance budget of rail infrastructure companies
The Logicx SPM system has been tested for 9 months on 9 frequently used switches and demonstrated prediction precision and reliability both in absolute and relative terms. SPM produced 15 times less false negative predictions and 3 times less false positive predictions in comparison to the native analytic system delivered by the switch manufacturer. The pilot version of SPM has predicted 95% of failures and produced 18% false positive alarms.
Here are a few research challenges we have faced during the system design face:
- Giving an operator an actionable insight, saying in how many days a switch is likely to fail (alternative systems did not provide any safety window information).
- Showing the switching phase in which the failure is likely to occur.
- Developing a generic prediction method that would work both for electric and hydraulic switches.
- Providing technicians with visual information that would allow them trace back and understand issues.
- Using data from pre-installed power consumption sensors, and avoiding installation of proprietary sensors.
- Development of a deep-learning system for automatic model parameter tuning for new switch types and instances.