The frequency-domain identification of gas turbine dynamics is discussed. Models are directly estimated from engine data and used to validate linearised thermodynamic models derived from the engine physics. This work is motivated by the problems previously encountered when using time-domain methods. A brief overview of frequency-domain techniques is presented and the design of appropriate multisine test signals is discussed. Practical results are presented for the modelling of the fuel feed to shaft speed dynamics of a twin-spool engine. The gathered data are analysed and the frequency response functions of the engine are estimated. The identification of parametric s -domain models is discussed in detail and a comparison made between the identified models and the linearised thermodynamic models. The influence of engine nonlinearities on the linear models is also examined.

This paper deals with the three most important sources of error in the practical identification of linear gas turbine models. These are noise, nonlinearities and unmodelled linear dynamics. Techniques are described which allow each of these sources of error to be studied and their influence to be assessed.

The linear multivariable modelling of an aircraft gas turbine is presented. A frequency-domain identification method is employed to estimate a family of models from engine data at a range of operating points. It is found that the fuel feed to shaft speed dynamics can be represented by second-order models. This matches the results obtained for the reduced-order linearised thermodynamic models derived from the engine physics. A direct comparison can thus be made between the estimated and thermodynamic models, which shows significant discrepancies between them for the engine tested. This work illustrates how modern system identification techniques can be used to verify engine models. The multivariable framework employed means that additional inputs and outputs can be easily incorporated into the model.

This paper examines the estimation of a global nonlinear gas turbine model using NARMAX techniques. Linear models estimated on small-signal data are first examined and the need for a global nonlinear model is established. A nonparametric analysis of the engine nonlinearity is then performed in the time and frequency domains. The information obtained from the linear modelling and nonlinear analysis is used to restrict the search space for nonlinear modelling. The nonlinear model is then validated using large-signal data and its superior performance illustrated by comparison with a linear model. This paper illustrates how periodic test signals, frequency domain analysis and identification techniques, and time-domain NARMAX modelling can be effectively combined to enhance the modelling of an aircraft gas turbine.

In this paper a feedforward neural network is used to model the fuel flow to shaft speed relationship of a Spey gas turbine engine. The performance of the estimated model is validated against a range of small and large signal engine tests. It is shown that the performance of the estimated models is superior to that of the estimated linear models.

This paper presents PID controller designs based on NARMAX and feedforward neural network models of a Spey gas turbine engine. Both models represent the dynamic relationship between the fuel flow and shaft speed. Due to the engine non-linearity, a single set of PID controller parameters is not sufficient to control the gas turbine throughout the operating range. Gain-scheduling PID controllers are therefore used in order to obtain optimum control. A comparison between the controller designs based on the two model representations is also made.

In this paper Nonlinear Model Predictive Control (NMPC) is applied to a gas turbine engine. Since the performance of model based control schemes is highly dependent on the accuracy of the process model, the estimation of global nonlinear gas turbine models using NARMAX and neural network is first examined. To solve the NMPC problem, the Newton-based Levenberg-Marquardt Approach (NLMA) with hard constraints and Sequential Quadratic Programming (SQP) with soft constraints are validated using a wide range of large random, small and ramp signal tests. It is shown that the control performance using SQP is slightly better than that of NLMA, and proposed methods are robust in the face of large disturbances and model uncertainties. The results presented illustrate the improvement in the control performance using both methods over against gain-scheduling PID controllers.