The challenge
An upstream oil and gas operator runs gas-turbine-driven compressors to re-pressurise gas at a production field. Financial and production planning needed degradation monitoring and remaining-lifetime estimation, but nobody knew whether the historical sensor data contained enough information to detect and model degradation. Unplanned failure of these assets carries capital-scale maintenance consequences.
At a glance
DataPI historian (42 tags, 2007–2014), PLC (142 tags, 2004–2009)
MethodsARIMA, piecewise OLS trend fitting, indicator derivation
PartnerUniversity fluid-mechanics department
What we did
Instead of waiting for a richer dataset, SWICON worked with a university fluid-mechanics department to convert non-quantifiable engineering expertise into mathematical form. Thermodynamic degradation indicators (polytropic exponent, pressure ratio, efficiency) were derived from first principles and computed from historian and PLC time series spanning 2004–2014. Trends were fitted piecewise, segmented by maintenance events, to separate recoverable degradation (washing cycles) from non-recoverable long-term wear.
The result
The proof of concept confirmed that turbine degradation is detectable from the available sensor data. Two degradation indicators, a power indicator and a temperature indicator, were derived and validated, showing both recoverable and non-recoverable degradation in the time series. The client got a green light for the full degradation modelling project. This is the reusable answer to the objection that the data is not good enough for AI.
