Multidisciplinary Optimization Success at OPEL with LMS OPTIMUS

The development of lightweight vehicles requires the use of innovative design concepts supported by powerful computer-aided optimization strategies. In this article, Dr. Axel Schumacher, project engineer in the OPEL International Technical Development Center in Ruesselsheim, Germany, highlights some of the issues raised during the definition of multidisciplinary optimization problems, and describes some of the applications that have already been successfully addressed.

“Before starting the deployment of a computer-aided optimization system we took a hard look at what exactly we really wanted to achieve”, said Axel. “We needed an interaction not only between the different simulation methods but also the different people in the development process. The optimization procedures have to help the engineers of different disciplines (e.g. static analysis and crash simulation) in their daily work. The system had to be open to many of our codes and procedures; it had to be usable by people of differing skill sets; it had to present a choice of DOE methods and be open to others of our design team when necessary; and it had to converge to an answer in a practical time scale using realistic compute resources. LMS OPTIMUS was found to address these needs, and is now used extensively within OPEL. With the software infrastructure in place, we could then begin to develop a process route map for solving multidisciplinary optimization problems.”

Firstly, a detailed definition of the optimization problem is required when working on a large-scale project in which several departments and disciplines are involved. The available variation possibilities for the structure (design variables), the requirements on the component goal and constraint functions, and the load cases (analysis model), need to be written into a specification list.
The choice of the design variables has substantial influence on the optimization of a component. The following questions are necessary: Which parameters of the component model can be changed? Which structural elements can be expected to have a large, medium, or small influence on the system behavior? What are the possibilities for the variation of the design variables in the optimization loop?

The requirements for the component are defined with the help of goal and constraint functions. The question then moves to: What are the objectives of the optimization? Which goal functions and constraint functions can be defined? Can we parameterize the goals and the constraints? Can we determine the sensitivities considering the continuity of the design constraints with respect to the variables?
Of course, it’s common for an optimization process to contain a structural finite element model. However for a complete multidisciplinary optimization, we also need to integrate procedures for flow calculation, fatigue, and production simulation... Also material data and the load cases have to be taken into account.

An automotive side rail must meet safety certification requirements for both front crash and static behavior. Crash performance is modeled using the explicit finite element code LS-DYNA, while the implicit finite element code MSC/NASTRAN is used for static analysis. Although the structure remains the same, each simulation has its own model description and input deck in the structural optimization loop. The design variables are four wall thicknesses of the frame side rail. The goal is to maximize the inertial energy (F) in the front crash. The constraints are then as follows (use of aluminum material):

The optimization results achieved by OPTIMUS are shown above. The first run had no constraint for G3, the second was a full optimization run. This demonstrates that the objective can be maximized while satisfying all the constraints (by using 41 DYNA- and 41 NASTRAN-Analyses).

In OPEL’s experience, working efficiently with applications involving large finite element models (>200,000 elements) is also possible with OPTIMUS. This is illustrated by the next example - roof crush resistance. In this case, the goal of the optimization is to raise the maximum resistance force without raising the structural mass. Additionally, the reduction of the resistance force after reaching its maximum must not be less than 20%. To reduce simulation time, the optimization used as few LS-DYNA runs as possible. Only the eight panels which were thought to have an influence on roof crush were elected as design variables: outer a-pillar; outer b-pillar; side panel, inner rear; reinforcement hinge pillar; reinforcement lower hinge pillar, b-pillar; inner a-pillar; roof frame side; and the front side panel. Based on only 23 crash simulations the optimization raised the maximum resistance force by 11% without raising the mass.

The basic idea in this exercise was to create a parameterized UniGraphics CAD-model that could be used within the optimization loop. The parameters describe the topologies, shapes, dimensions, and material properties of the components. The involved simulation programs produce response parameters which have to be defined as goal functions or constraints. The main work is the creation of the parameterized CAD-model. The models are generated based on several, so-called, sketches. The coordinates and the orientations of the sketches, and the sketches themselves, depend on the defined parameters. In advanced design processes, the definition of the parameters is made with the definition of the design variables in mind. Otherwise, the design variables have to be used to determine the CAD-parameters with the help of special equations - which can easily be defined in UniGraphics.

The number of parameters in a real CAD-model is very high. For example, we have not less than 200 parameters for the description of a simple radiator bracket. With the previously defined methodology, we only have to define 10 design variables. During the optimization process, the other parameters are constant or are functions of the 10 variables. By building a good parameterized model, it is possible to create different variants for different car lines by changing the key parameters - in our example the key parameters are related to bracket height and length. Here we can use the new model for the optimization by transferring the key parameters into the optimization process.

All the applications introduced so far have used a variety of computer simulations to predict system behavior. However, the optimization procedures provided by OPTIMUS are just as usable with test-based models of the system. An interesting example is the engine calibration based on results of the engine test bench. For example, engine torque is measured at different engine speeds, load, and air fuel ratio as a function of spark advance. First, OPEL engineers generated a design of experiments with OPTIMUS. They then carried-out the experiments on the engine test bench and measured the result torque. The results were imported into OPTIMUS and an engine map generated using response surface techniques. This RSM was then used in the optimization loops to obtain the best overall engine performance given the constraints. By automating the map generation and optimization process within OPTIMUS, OPEL was able to automatically perform 800 sub-optimization problems for the determination of the optimal spark on the whole engine map.