Simulation has been used from onset of CAE, in designing one of the most important aspects in vehicle design: Occupant safety system, to design safer vehicles. For over a decade I have been fortunate enough to work with some incredibly dedicated occupant safety system engineers in looking at the tools and processes to design and optimize safety systems such as airbags and seatbelt system. A favorite part of my role has been teaching new tools & methods, and the most important part of teaching is listening to the needs of the students. The primary need over time can be distilled as follows: “How do we have a true impact on the product design, instead of simply doing a lot of analysis?”
Safety systems must be effective based on vehicle crashworthiness characteristics and meet the regulation requirements and targeted consumer safety ratings. Regulations are continuing to be more complicated & stricter as well consumers have the better awareness of safety ratings. This generates the need for optimum safety system design for all vehicle classes.
Several important aspects must be considered during the design stage. A high level of nonlinearity is present, as the models need to capture the physics of advanced features like adaptive vents, multiple level output inflators, and adaptive seatbelts systems. Optimum and robust design should satisfy performance for various load cases. For example, when designing the frontal safety system, different combinations of scenarios should be considered for unbelted & belted occupants (5th, 50th and 95th percentile) subjected to frontal and oblique conditions. Simple and effective communication and presentation of results are also very important, as multiple teams of engineers (i.e. occupant safety and crashworthiness) from both OEM and Tier 1 suppliers are involved.
A process is made up of tools and methods. Any change in a tool or method is thus a change to the process – and one of the hardest things to accept for disciplined engineers is a process that is at risk for not working. Forward progress, however, can necessitate iteration and failure of processes and frankly an attitude of “try again” until we are comfortable. Mitigating the risks of inserting new tools and methods into a process can be accomplished by good discussion, experience, and iteration.
I have observed two important process improvements in occupant safety simulation field. First, both the crashworthiness and safety simulation are now done using the same solver, pre-processor, and post-processor so that they can communicate/interface well with each other. Second, the DOE and optimization have restart capability almost like the video games where you can pause and change a few things and continue playing. I will talk more about this very important restart feature as I noticed “changes” are often in early design phase of occupant safety system. Thus, process and tools should accommodate the real-life changes happening in designs.
To meet stringent performance targets with tight timelines, engineers can build accurate and robust models with reasonable run times such that multiple models can be analyzed for design of experiments (DOE), and then eventually towards optimization. These occupant/sled simulation models are typically built with the desired detail and accuracy based on the problem to be solved. On one hand it could be express (semi-rigid) sled model, where most of the vehicle surfaces are rigidly modeled, and only the seatbelt system, airbag and uses express dummies from Humanetics some portions of dummy are deformable. These express models typically run in 15-30 minutes. Or it can be a fully deformable occupant safety model, which is cut from a full vehicle simulation model and runs typically in 2-4 hours. Express models are typically validated based on the baseline or benchmark design performance. Shown below is an example RADIOSS express sled model of a frontal system built using Humanetics express dummy and a setup of Hyperstudy DOE for occupant safety system.
(Left Image: with Humanetics Express Dummies – © 2016 Humanetics Innovative Solutions)
So, what are these teams optimizing..? Well, they are minimizing the injuries for various load-cases while keeping the cost in check. Occupant injury criterion such as HIC (head injury criteria), Chest acceleration & deflection, neck injury criterion (Nij), femur force are typically the direct responses to control. In addition, there may be indirect responses like pelvis, head travel trajectory, contact forces and seat belt forces. Typical design variables are pre-tensioners’ activation time, airbag activation time, airbag shape, tether lengths, vent shape, vent activation, and load limiter characteristics. Often, multiple models simulating different configurations with some linked design variables across models are used for DOE and optimization studies. The current state-of-the-art tools make the process to set-up the DOE and optimization very simple and intuitive. They also provide a dashboard to review the results.
Traditionally, design exploration process involves a fairly large DOE followed by generation of the response surface. Optimization is then performed on this response surface, followed by a validation with the solver runs. Considering the nonlinearity level in the occupant models, it is very difficult to get a good correlation between the response surface and solver validation results with a reasonable size of a DOE matrix. A very large DOE matrix size may provide with a good response surface but the turnaround time becomes practically prohibitive.
The global response surface method in HyperStudy suggests to eliminate the independent DOE step and provides a practical and simple optimization process. This method explores the design near local as well global optima by optimizer directly interfacing with solver, without user intervention. Simple and intuitive this process avoids the need for extra steps (to fit and generate mathematical surface), to be done explicitly by the user. However, in the case of occupant safety system scenario where optimization process is continuously improving, it is recommended to adapt this as a two-step approach. First, do a relatively smaller size DOE which would give a peek on the response scatter and at the same time allows an opportunity to validate the models. Next, optimization with performed directly on the solver can be continued from the preliminary DOE results. The flow charts for these approaches’ are shown below for a typical occupant safety optimization study.
Problem formulation; identifying design variables and responses, is one of the most critical steps the engineer performs in design exploration and optimization. As they learn about the problem, they often need to make changes to it such as adding/removing design variables or responses. As a large amount of machine time and human effort could have been invested in these initial design phases, the most desired and efficient option for the optimizer is to be able to reuse the designs from the initial investigations. Implementation of this capability must be robust, flexible and yet easy-to-use. One such implementation exists in HyperStudy allowing engineers to reuse their previously explored designs.
Wow! If you had told me 10 years ago that we could get meaningful results that can truly affect the design in that sort of timeframe, I would have called you crazy. Not so crazy anymore…and it is a great feeling to work with our clients to have optimization and analysis work truly having day-to-day design impact. No more “check box analysis” – we are designing safer vehicles!
Go ahead, make your models robust, explore and optimize; it has become easier than ever. The ability to move designs forward through a rapid capture and optimization, taking advantage of previous iterations and learnings, make today an exciting time to be involved in safety system design. No more lost efforts, just more fuel for evolving great products. To summarize, I have three recommendations about DOE/Optimization of Occupant Safety System by simulation:
1. Don’t do very large DOE in beginning. You will want to make changes.
2. Don’t rely on mathematical response surface for extrapolation. Your models have enough non-linearity, let the tool run optimizer directly with the solver.
3. Take advantage of the flexibility of tools: Pause & Play. DOE of small matrix, optimize with the solver, make changes to the setup, and restart by letting tool include your previous design exploration learning.