Gas turbine fault and wear prediction system
Predicting emergency stops and identifying their causes
- Oil and gas
- Chemical industry
- Artificial intelligence
- IBM SPSS
An industrial-grade methanol plant produces up to 3,000 metric tons of output daily and almost 1,000,000 metric tons annually. Every day of downtime can cost millions, break the schedule, and affect shipment. To ensure that production is continuous and maintenance downtime events are as short as possible, it is imperative that chemical plant equipment operates without faults.
For one of Russia's TOP 5 methanol manufacturers, Rubius has developed a gas turbine fault and wear prediction system. It forecasts a fault 1-3 months ahead, ensures timely maintenance, and cuts repair costs by 7 times. System implementation pays for itself after the very first fault is prevented.
The system is powered by a neural network trained on 626,000 repair data records and 4-year equipment parameters. How it works:
- analyzes 50+ equipment parameters parsed from SCADA software: shaft vibration and axial displacement, support bearing temperatures, inlet/outlet steam temperature, turbine speed, and others
- using this 1-3-month data, the system predicts faults and critical wear of the equipment
- calculates the mean time between failures (6,582 hours for a gas turbine)
The system prevents emergency stops and secures continuous operation of a chemical plant. The company has optimized the repair schedule: from now on, any work is performed if the equipment condition – rather than a formal procedure – requires it. Such an approach helps avoid neglected cases entailing high-cost equipment recovery.