The needs
The design of the intake and exhaust manifold geometry and valve timing of an Internal Combustion (IC)engine is very critical as it affects the engine performance,fuel consumption and exhaust emissions.
This application shows how LMS OPTIMUS –the leading software in Design Space Exploration and Optimization – is used with RICARDO WAVE –the leading engine performance and gas dynamics simulation software – to optimize an IC engine ’s performance.
The design problemThe simulated engine is a 1.6L, four-cylinder,four-stroke, 16
valve, gasoline engine (Figure 1).The design variables that are examined are six geometry parameters of the four intake runners as well as four parameters that influence the intake and exhaust valve profiles and timing. Each intake runner is made of three parts and the length and diameter of each part is a design variable. The valve profiles are parameterized by the lift center angle and the lift duration. The design outputs of interest are the Total Volumetric Efficiency (VOLEFD) at different engine speeds. A global performance objective is to maximize the weighted summation of VOLEFD over the complete rpm range. Also, by differing the weighting factors for some engine speeds, it is possible to tune the performance of the engine to certain rpm regions.
The process
OPTIMUS is used to generate different design alternatives by assigning values to the set of design variables, automating the WAVE runs, and the computation of the objective function,towards an optimized design. Figure 2 shows the OPTIMUS analysis sequence.It specifies that WAVE is run using tut-16v-optimus.dat as input data file and that output results are extracted from file tut-16v- optimus.sum. All the process of updating the input file and retrieving the results is managed automatically by OPTIMUS.
The solution
OPTIMUS enables a number of different strategies in identifying the optimum. Direct application of optimization
methodologies on the analysis sequence, hybrid methodologies involving Design of Experiment (DOE) and Response Surface Modeling (RSM) along with optimization algorithms, design space exploration methods and others. Here two different strategies are examined in solving this problem:
- Explore the design space using DOE techniques and RSM, generate an approximate optimal solution based on the RSM and use this approximate solution as a starting point for an optimization based on the WAVE simulations.
- Perform the optimization directly on the WAVE simulations.
For the frst strategy, a classical DOE plan, the Full Factorial Design (210 experiments) is used. In order to save computation time (the analysis requires about 5 minutes per experiment on a HP J5000) a Quarter Fractional Factorial Design (257 experiments) is used instead. All the variables are still varied over two levels but highest level interactions are avoided. The above DOE is suffcient for the generation of a frst-order Taylor polynomial. As it appears in Figure 3 the simulated center point shows a large difference between VOLEFD calculated by the RSM and the actual value. In enhancing the RSM, a random-type DOE (Latin Hypercube) with 150 experiments is processed in order to add information to the factorial plan ‘inside’ the design space
