Multi-Cylinder Adaptation of In-Cycle Predictive Combustion Models

Abstract

Adaptation of predictive combustion models for their use in in-cycle closed-loop combustion control of a multi-cylinder engine is studied in this article. Closed-loop combustion control can adjust the operation of the engine closer to the optimal point despite production tolerances, component variations, normal disturbances, ageing or fuel type. In the fastest loop, in-cycle closed-loop combustion control was proved to reduce normal variations around the operational point to increase the efficiency. However, these algorithms require highly accurate predictive models, whilst having low complexity for their implementation.Three models were used to exemplify the proposed adaptation methods: the pilot injection?s ignition delay, the pilot burned mass, and the main injection?s ignition delay. Different approaches for the adaptation of the models are studied to obtain the demanded accuracy under the implementation constraints. Non-linear adaptation techniques are necessary for the proposed models. This was compared to a linear formulation that reduced the complexity. A reduced multi-cylinder approach is presented as a method to reduce the total number of parameters while preserving the accuracy. A method to select the parameter for the reduction is also proposed. The sensitivity of the models and the robustness of the algorithms was studied. To reduce the complexity of the model implementation, the performance of Taylor?s expansions was studied.The methods were tested from experimental data obtained from a Scania D13 six-cylinder heavy-duty engine run with conventional diesel, rape methyl-ester (RME), and hydrotreated vegetable oil (HVO). The adaptation of the models proved to significantly improve the prediction accuracy for each of the cylinders. The average bias error is eliminated whilst the total error dispersion was halved. The results validated the reduced multi-cylinder adaptation as a method to reduce the total number of parameters and have similar prediction accuracy. Furthermore, the multi-cylinder adaptation was the most robust against measurement errors. For the ignition delay models, the sensitivity to the nominal point of linearization was under the required prediction accuracy for the in-cycle closed-loop control algorithms i.e. under the detection accuracy of 0.2CAD.