Prediction of exhaust gas temperature in GTE by multivariate regression analysis and anomaly detection

Statistical multivariate linear regression technique has been applied in predicting exhaust gas temperature (EGT) for a small gas turbine engine using three independent input variables. Data collected earlier over three years (YR) of operational cycle are used for modeling, training, testing and validation of the models. Regression coefficients, probability and significance, R-square and RMSE values are considered for quantitative comparison between regressed and measured EGT data. R² and RMSE values are observed to be highest for intermediate period (intermediate cycle in YR2) while the two values are substantially low for YR3 data (end cycle). These two values are 0.22 and 2.23 for 2010 data as compared to 0.93 and 11.4 for YR2 data. The results obtained through this work are indicative of an anomalous situation and support our earlier findings by ANN technique.