Model Reduction of Diesel Mean Value Engine Models 2015-01-1248
In the literature, a wide range of Mean Value Engine Models are used in the simulation and control of reciprocating engines. These models are often underpinned by a number of implicit assumptions, which determine the model structure and system states. Systematic model reduction approaches have been developed to avoid these assumptions, where high order models are reduced using singular perturbation techniques, eliminating states operating on irrelevant time-scales. While this framework allows the elimination of states based on sufficiently small perturbation parameters, a systematic method of identifying non-dimensional perturbation parameters has not yet been proposed. The development of a rigorous method to identify non-dimensional time scales present in the model is a natural and powerful extension to the existing approach.
In this work, starting from a calibrated, high-order physics based non-linear mean value model of a diesel engine, non-dimensional analysis is used to identify system states on relevant time-scales. Singular perturbation techniques are then used to isolate the identified states. The resulting boundary layer is then approximated before the impact of linearising the state and output equations is investigated. The result is a collection of engine models which trade off computational complexity for modelling accuracy, appropriate for use in detailed simulations and for model based controllers. The accuracy of the reduced models in the relevant time scales is validated through a combination of extensive simulation examples and engine test data.