Top > Application Examples > Tool for Fast, Cost-Efficient Analysis

Tool for Fast, Cost-Efficient Analysis

Searching for the ideal combination of design variables and objective functions requires high level of engineering knowledge and experience; this is quite a task when there is limited time. Software Cradle’s Extension Option (Optimization) known as EOopti can cut down unnecessary trial and error and identify the most fitting variable and object combinations for each design case, all in a short amount of time. Shigeki Katsumura, Assistant Manager of the Cradle Engineering Department, explains why and how this tool was developed.

The advancement of computer processing speed has made it much more common that product engineers apply CAE (Computer Aided Engineering) in their daily design work. As a developer of fluid analysis software, Cradle is an important contributor to not only promoting this trend but also advancing it. In particular, even as the use of CAE has become more common, there is an increasing realization that CAE has even greater potential. The ultimate desire is for the CAE analysis to encompass our experiences, knowledge, and decision-making abilities, such that the final model contains the highest level of thinking.

Optimization involves trial and error, whether developing a vehicle with the least air resistance, allocating parts, or modeling heat sinks. Thermal fluid analysis can sometimes take significant amount of time to calculate the solution depending on model shape. The ideal situation would be to identify the optimized modeling and part allocation from only a few samples. However, in reality engineers must often review hundreds and thousands of samples, before determining the best solution, which consumes time and money.

That was what inspired us to develop an optimization tool that can reduce examination tasks and identify the design variables most fitting for the customer’s purpose. After a thorough search of available optimization methods, we decided to implement MOGA (Multi Objective Generic Algorithm), which was introduced by Tohoku University’s Obayashi Laboratory, in the EOopti available from Version 10 onward.

What Challenges Did You Face during Development?

When I was first assigned to develop EOopti, I’d had very little experience working with optimization problems. My first struggle was to come up with a concept for the GUI (Graphic User Interface) specification. To overcome the challenge, our team worked closely to investigate prototypes, discuss how they could be improved and implement those changes; we repeated this process a number of times before reaching the final design.

The biggest issue for me was that I needed to develop a deep understanding of the optimization algorithm to create highly extensible GUI. Naturally I had to do these things concurrently – namely, developing an understanding of the algorithm, implementing the algorithm, determining the GUI specification and implementing GUI.

It was challenging, but between my research and consulting with other developers who provided much useful insight, the vision for how the application should function gradually became clear.

EOopti will be displayed as shown in Figure 2. A response surface can be generated using the Kriging method. The condition wizard is used to input the necessary data to execute the multi-purpose optimization tool which implements MOGA. Optimized results will be added to the graph tree as new labels, and each graph can be viewed on a draw window.

The Kriging method is an interpolation technique to estimate the value at a given point based on the known values in the neighborhood of the point. Ideally we would have a mathematical expression for the optimizing target, i.e. each objective function value, but working with non-linear substances like fluid makes the process extremely complicated. As an alternative approximation to each objective function, EOopti applies the Kriging method, based on the sampling point, to generate the response surface distribution and then runs MOGA to find the optimized solution.

The process to determine the optimum solution can be divided into the following three phases:

**1. **Define the design variables and objective functions

**2. **Generate the sampling point (design variable value)

**3. **Specify the objective function value based on the design variable value from phase 2.

Phases 1 and 2 can be carried out in EOopti, whereas phase 3 requires values from additional information, including data from the analysis results.

An example of a heat sink can be used to illustrate the process. The heat sink dissipates heat inside an LED lamp as shown in Figure 4. Optimization in this case involves the following design variables: the number of fins, fin height, thickness of the heat sink shaft, and the outer diameter of the fin and shaft sections. The objective function values are based on the LED temperature and volume. The range of each design variable is shown in Figure 5.

1) Define the design variables and objective functions

Setting EOopti conditions can be done using the condition wizard (Figure 6). On the first page of the condition wizard, engineers input values for the design variables, objective functions and sampling point. In this case, the design variable is 5, the objective function is 2, and the sampling point is 50. These values simply mean that analysis results are necessary for 50 cases, and the results will be used to generate the response surface using the Kriging method.

- General Purpose Structured Mesh Thermal-fluid Analysis Software
- More Details

- 3D Structured Mesh Thermal-fluid Analysis System for Electronics Cooling
- More Details

- General Purpose Unstructured Mesh Thermal-fluid Analysis System
- More Details