January 28, 2020
Mechanical Support Optimization with Tight Simulation Budget
The automotive industry is looking for innovative solutions to decrease lead time and improve performance of their product. Lead time reduction has an especially huge impact on the cost, considering the size of manufacturing volume at stake.
In this article, we describe how to leverage simulation toolchain automation and data analysis in order to develop a process to improve design performances of a mechanical support. We show that within a very limited computational budget imposed by a tight business schedule, straightforward application of the Surrogate-Based Optimization technique affords a significant design choice time reduction and dramatically increases the efficiency of use of the simulation model.
Figure 1 shows a view of the numerical model used to simulate the physics at stake. The goal of the design process is to find a solution to assembly the PCB into the cover, so that the assembly is rigid enough with respect to the range of allowable vibration modal eigenvalues, and that the induced thermomechanical stress remains acceptable. The bind is made using plastic pins which number should be minimized, hence for each number of pins, we must find the pins positions which induce the highest modal eigenvalues with the lowest thermomechanical stress.
Fig. 1. Simulation model used in the optimization process
The goal of the study is summarized in the following table.
|Modes values||Above 2000 Hz|
|Thermomechanical stress||As low as possible|
Thermomechanical stress and structure modes values are evaluated by two FEA simulation models developed in Ansys. Fixed boundary conditions are set on the cover, and all the connections between the assembly’s parts must be carefully set up for the vibration model. The thermomechanical stress is induced by the temperature variation which must be sustained without any risk of failure.
Current process is usually achieved by the iterative manual actions and could take several weeks.
- There are many forbidden areas with non-trivial shapes where the pins cannot stand
- Two criteria are competing; hence the optimization problem is multi-objective
- It is hard to say whether the search is exhaustive when done by hand
Automation of the computational toolchain
For each new design, a new CAD model must be generated, and the mesh must be updated. This can be done manually, but it is very time-consuming and expensive. Therefore, geometry and mesh generation were automated in a pSeven workflow.
Figure 2 shows a view of such pSeven workflow. Scripting capabilities of SpaceClaim are used to generate the geometry with a given number of pins and its positions. The geometry is then provided to Ansys Workbench, where the mesh is updated, and the vibration and thermomechanical models are solved. The optimization is driven by pSeven after the user has set a simulation budget (maximum number of simulations pSeven can run).
Fig. 2. Automation of the tool chain in a pSeven workflow
An interactive view of the model is produced at runtime and allows to check that the geometry is well generated. Visualization of the PCB with 6 pins is shown in Figure 3. Forbidden areas (where the pins cannot stand) are handled with user-defined analytical constraints, which are regularly handled by pSeven. The user defines an analytical function which is positive only when the pin is inside a forbidden area. A design is said “feasible” only if that function is negative.
Fig. 3. Interactive view of the geomtry produced during workflow execution. Forbiden areas are handled by analytical functions defined by the user.
Trying to solve the straightforward multi-objective optimization of the PCB, we obtain the Pareto frontier for 3 pins configuration shown on Figure 4. That plot shows that, to some extent, the two objectives are competing. Although the search is efficient and provides insight in a relatively short time, none of the designs satisfy the constraint on modes values. Hence, we decide to optimize the PCB in two successive steps:
- We start with the optimization of the modal behavior, which provides the highest modes values for a given number of pins. The configuration with the fewest number of pins which achieves modes values specifications is selected for next step.
- Once the requirement on modes values is achieved, we proceed with minimization of the thermomechanical stress with a constraint imposed on modes values.
Fig. 4. Pareto frontier obtained by multi-objective Surrogate-Based Optimization.
Optimization results, modes only
A few configurations were optimized with different numbers of pins and a simulation budget set by the user. Results are gathered in the following table.
|Num. pins||Sim. budget||CPU time (h)||Mode 1 (Hz)||Mode 2 (Hz)||Mode 3 (Hz)|
Surrogate-Based Optimization performs well even for cases with low simulation budget. Especially for the 6 pins configuration, the best design is found with less than 16 points per variables. That outcome is very interesting in a context where development time is tight.
Moreover, the Surrogate-Based Optimization technique is global by nature. The search for the best design is exhaustive, which allows to efficiently eliminate infeasible configurations. In this study, we have shown that at least 5 pins are required so that mode 3 is above 2 kHz.
Most importantly, the time required to make a design choice is drastically reduced, compared to the manual design space exploration.
- The simulation budget for the optimization can be set by the user, following tight schedules. pSeven can find interesting designs even with a very low simulation budget.
- The search is made by pSeven and does not rely on the user. While the user may lack objectivity, pSeven objectively proposes new designs based on data analysis.
- The design space exploration is exhaustive, the best design cannot be missed if the user sets the proper simulation budget.
- The simulation toolchain is automated. The user does not have to run the simulation manually, pSeven automatically calls SpaceClaim and Ansys Workbench.
- That process can be re-used on a new industrial project after 1 man-day re-work of the pSeven workflow.
By Martin Pauthenet, Application Engineer, DATADVANCE and Ciprian Crisan, Thermomechanical Simulation Engineer, VITESCO
Vitesco Technologies, formerly (until September 2019) the Continental Powertrain Division, brings together the full spectrum of Continental’s drivetrain technology expertise. The company’s aim is to develop innovative, efficient electrification technologies for all types of vehicle. Its portfolio includes 48-volt electrification solutions, electric drives, and power electronics for hybrid and battery-electric vehicles. Furthermore, the product range counts electronic controls, sensors and actuators as well as solutions for exhaust after-treatment.