Engine systems must continuously increase their thermal efficiencies and lower their emissions in real operation. To meet these demands, engine systems are increasingly improving their transient performance through control technology. Conventional engine control systems depend on control maps obtained from huge numbers of experiments, which is necessarily limited by the available number of man-hours. These time-consuming control maps are now being replaced by control inputs derived from on-board models. By calculating optimized control inputs in real time using various information, model-based control increases the robustness of advanced combustion technologies such as premixed charge compression ignition and homogeneous charge compression ignition, which use auto-ignition and combustion of air-fuel mixtures. Models also incur relatively low computational loads because the specifications of the engine control unit are lower than those of current smartphones. This article develops a simple diesel combustion model with model-based control of the multiple fuel injections. The model employs the discretized cycle concept based on fundamental thermodynamic equations and comprises simple fuel injection and chemical reaction models. Our control concept aims mainly to decrease the fuel consumption by increasing the thermal efficiency and reduce the combustion noise in real-world operation. The model predicts the peak in-cylinder gas pressure and its timing that minimize the combustion noise and maximize the thermal efficiency, respectively. In an experimental validation of the model, the computed and measured in-cylinder pressures were well matched at each phase under various parameter settings. In addition, the calculation time of the model is sufficiently short for on-board applications. In future, the proposed model will be extended to the design and installation of controllers for engine systems. The control concept and associated problems of this task are also described in this article.

We developed a feed-forward controller for a conventional diesel combustion engine with triple fuel injection and experimentally evaluated its performance. A combustion model that discretizes an engine cycle into a number of representative points to achieve a light calculation load is embedded into the controller; this model predicts the in-cylinder gas-pressure-peak timing with information about the operating condition obtained from the engine control unit. The controller calculates the optimal main-fuel-injection timing to control the in-cylinder gas-pressure peak using the prediction result as a controller with a single input and output. The controller's performance was evaluated by experiments using a four-cylinder diesel engine under changing the target value of the in-cylinder gas-pressure-peak timing during a target-following test and the performance was also evaluated under changing the exhaust gas recirculation ratio at the constant target value of the in-cylinder gas-pressure-peak timing for the disturbance-response test. It was found that the controller could calculate the optimal main-injection timing over a cycle and maintain the targeted in-cylinder gas-pressure-peak timing even when the target value or exhaust gas recirculation changed. The combustion model was also shown to be fast enough at predicting diesel combustion for onboard control.

In this study, a multiple-input multiple-output feedforward controller for use with multiple-point fuel injection systems is applied to a diesel engine using an original control-oriented model. The target-tracking performance of this multiple-input multiple-output feedforward controller was then tested in terms of how well the controller adjusts the fuel delivery ratios to the pilot injection, pre-injection, main fuel injection, and the main fuel injection timing, and controls the in-cylinder peak pressure and its timing. Control experiments are conducted at different engine outputs and speeds while changing the targets of the in-cylinder peak pressure and its timing. The controller is able to track the varying targets within an acceptable error included in the original model. The calculation time, which is almost double that of a single-input single-output controller, is sufficiently fast to derive the applicable inputs.

No abstract

No abstract