MultiObjective Optimization in Engine Design Using Genetic Algorithms to Improve Engine Performance
The use of genetic algorithms (GAs) is enjoying ever-increasing popularity among engineers and designers as a way of optimizing their products and processes. One of the major advantages of such techniques is that they allow true multiobjective optimization (where the objectives are kept separate throughout the optimization process, rather than being collapsed into a single, weighted objective function from the beginning).
The engine chosen for this work was an automotive 4-stroke, 4 cylinder, 2.2 liter engine with a bore of 93mm, a stroke of 81mm and a compression ratio of 8.9. There are two valves per cylinder and fuel is injected into the intake port. The engine runs in speed mode, where the engine speed is specified and the engine torque is calculated. Combustion is modeled by a Wiebe function (with spark trimming simulated by the 50% burn point parameter), in-cylinder heat transfer is modeled using the Woschni model and in-cylinder wall temperatures are fixed at specified values for WOT and 5000 rpm.
Thus, at the completion of the optimization process, the engineer is presented with a Pareto Frontier (or trade-off curve) of results, from which he can pick the point at which he would like to operate.
Moreover, they are very applicable to practical engineering problems, as they allow the use of discrete variables rather than the continuous variables required by single-objective gradient methods. The inherent criticism of such methods (that they need many data points in order to create a useful trade-off curve) is largely by-passed when they are applied to programs such as GT-Power which can run an iteration orders of magnitude faster than the more CPU-intensive CAE tools. Here we show the results of using the modeFRONTIER optimization and process-integration software to an engine design case, where the multiple objectives were the minimization of both the brake specific fuel consumption and the NOX production.