Gas Turbine Bearing Temperature Monitoring via Regression Modelling

Abstract

This paper focuses on using Regression technique (MLR) towards finding solution to incidence of high compressor bearing temperature on one of the units at Geregu power plant in Ajaokuta, Nigeria. Monitoring of parameters related to the bearing temperature was carried out to find out causes for the high bearing temperature fault and came up with successful diagnosis by interrelating the gas turbine current lube oil test results of parameters like the kinematic viscosities, % concentration of additives and flash point with reference and standard VG46 lube oil data published in literature. Using statistical tools like the Pearson correlation and co-variant metrics for the five-years, the oil viscosities at 100^oC and 40^oC were selected as the input of the MLR model based on their Pearson coefficients of (-98.08%) and (-99.68%) respectively relative to the compressor bearing temperature. The MLR model for the bearing temperature prediction gave a root mean square error of 0.121 and coefficient of determination (R²) of 99.71%. The model predicts that by the 2^nd quarter of 2025, the bearing temperature would have reached the alarm point (90⁰C) from the current value of 85⁰C and that by the 1^st quarter of 2027, the bearing temperature would have reached the trip point (120⁰C). Conclusion reached is that a well formulated data driven model can reliably forecast bearing temperature and together with sensors aid in gas turbine condition monitoring. Likewise, it is concluded that shearing due to the consistent high temperature operation of the gas turbine lube oil is responsible for the depletion of the Zinc (-23.9%) and Magnesium(-26%) additives leading to the decay in the viscosity and consequent bearing temperature increment. Recommendation made is to either replenish oil with anti-wear additives or completely replace the oil to minimize the bearing wear rate and thus the bearing temperature.