The use of optimal estimation techniques in the analysis of gas turbines

This thesis discusses several methods that can be used to analyse gas turbines, based on an optimal estimation algorithm called the Kalman Filter. These techniques overcome the difficulties of more 'traditional' analysis methods, which can give misleading results because they do not explicitly consider the possibility of measurement error. An enhancement to the Kalman Filter (the 'Concentrator') is presented, which overcomes the Kalman Filter's tendency to 'smear' the effects of genuine changes in a small number of component changes and/or sensor biasses over the whole set of changes and biasses being considered. To complement this, methods of optimising some of the statistical inputs to the Kalman Filter in order to improve the ability of the 'Concentrator' to carry out the required analysis are discussed. These are based on analytical methods developed to determine the sensitivity of the Kalman Filter to its inputs. Techniques are also presented for determining the gas-path measurements in a gas turbine that are needed to enable the required analysis of component changes and/or sensor biasses to be performed, including determination of both possible measurement redundancy and the ability of a set of measurements to successfully differentiate between all the component changes and sensor biasses being sought. A recursive algorithm for time series analysis (the Smoothing/Trending Algorithm) is also presented. This produces, for each point in a time series, best estimates of the underlying levels and trends (rates of change of level) of the process generating the observations. A method of combining the 'Concentrator' and the Smoothing/Trending Algorithm is also presented, which reduces the effects of sensor noise on the analysis of component changes and sensor biasses from time series data. Many types of prime movers and process plant could be effectively analysed using the methods described in this thesis.