Collaborative Multidisciplinary Design Optimization in the Automotive Industry is a white paper just released by ESTECO detailing how Ford achieved “streamlined, multi-user design process management by expanding its MDO approach at [the] enterprise level with ESTECO’s collaborative web-based environment, SOMO.”
The accomplishments of Ford and ESTECO documented in this paper reinforce our long-held view that institutional adoption of design space exploration, design optimization and process integration is a crucial goal for engineering organizations working to establish these as strategic competencies, not just tactical. Too often implemented at only the department or workgroup level, the technologies and their attendant work processes need to be recognized and given backing as enterprise capabilities to have their greatest impact on engineering’s ability to advance corporate strategic objectives. See Design space exploration: Institutionalizing the practice and Ford adopts ESTECO’s SOMO to institutionalize enterprise MDO.
The new paper centers on two case studies. The first documents how Ford combined “multi-domain, simultaneous analysis and optimization with advanced mathematical tools that enable consistent computational resource savings,” while the second “looks into the extension of the collaborative optimization approach to the ONE Ford core platform system and targets a wider design scope.” From the white paper:
DOE and RSM-based design optimization meets NCAP/IIHS requirements
When designing a vehicle frame, performance targets and requirements are collected from many sources and translated into design features and attributes by engineering teams. The overall objective of this project was to minimize the weight of the frame while meeting key attribute requirements, more specifically: attaining safety NCAP 5-star and IIHS top rating and hitting durability performance targets.
To achieve the different performance targets (stiffness, NVH quality, durability and crashworthiness), the group employed a variety of CAE tools (Nastran, LS-Dyna to name a few) to model the diversity of car body behavior. Among the target behaviors examined, stiffness—both static and dynamic—primarily involved linear calculations. NVH analysis, on the other hand, entailed complex multiphysics problems that considered the physical interaction of frame components and whole-body vibration. Crashworthiness, which is nonlinear to different degrees, depending on whether it is front, rear or side impact, presented the most difficulty. “For each attribute, engineers build the optimization workflow, create a specific DOE, train relevant response models and then publish the results to the central library, making them available to the MDO expert. He/she is now able to compose the comprehensive MDO workflow based both on his/her multidisciplinary expertise and the attribute experts’ knowledge made available in the central repository,” says Yan Fu, Technical Leader of Business Strategy and Engineering Optimization at Ford.
In order to bring all the data together into a “best-performance” body structure design, the team first employed the most suitable Design of Experiments (DOE) techniques and trained approximation models (Response Surface Models—RSM) that predicted the responses of the disciplines considered. In fact, when using computationally expensive simulation like the one involved in this case study, it becomes convenient to rely on a hybrid strategy involving both simulation codes and response surface models. In fact, RSMs enable the use of evaluation-intensive optimization algorithms, better suited for finding global optima but that require many evaluations (e.g., genetic algorithms).
In this case, the final workflow consisted of 49 variables, some of which were common to some of the attributes, and 12 constraints. By considering all the different performance targets at once, the optimization run enabled engineers to perform a tradeoff analysis of conflicting performance responses. As Yan Fu explains, “The MDO expert doesn’t have to re-run the single-discipline workflows, nor understand all specific details related to them. He/she combines the different domains using the ESTECO collaborative design optimization environment: without SOMO, he/she would have had to knock on everybody’s door, get their models and understand their requirements, review the whole workflow together with all attribute engineers…it would have taken a lot of time.”
Before building the top-level MDO workflow, attribute engineers confirm the accuracy of the response models by performing the FE validations of their respective domain, checking that the response models match the simulations predictions. After the validation step, the MDO workflow is ready to be run on the company High Performance Computing (HPC) system, where the design evaluators are managed efficiently by the queue manager built-in to the SOMO architecture. The RSM-based workflow took about 5 minutes to calculate more than 20K designs before proceeding to the FE result validation step involving the best designs, which took about a day to be completed.
“When it’s time to present our results to top management, we not only have the comparison between baseline and optimized designs, but now we can also rely on an effective web-based tool that enables a quick decision-making process across attributes. SOMO’s insightful charts help highlight how much weight we are saving, what are the tradeoffs between considered disciplines, how sensitive the designs are, and much more. The analytics tools automatically identify design candidates and help evaluate alternative options, allowing us to address not yet emerged feasibility issues and have second choices. Thanks to the design history archived, the manufacturing aspect can be evaluated and considered for the final choice. Most important of all, the backup solutions are archived and versioned, immediately ready and retrievable in case of need,” says Yan Fu.
Getting real with direct optimization
The second case study opens up to a more pervasive application of the collaborative MDO approach by combining its mechanisms with Ford’s core platform philosophy.
The design team wanted to perform a direct optimization analysis involving the core platform for a big pickup truck, which comprises three different truck sizes. The main objective was again to minimize the weight of the truck platform while respecting the constraints relating to safety, NVH and durability. When breaking the problem down into the different domains involved, the complexity of the project becomes clear: the final workflow included 7 models for a total of 113 variables, some of which were different for the various frame, box and cab models considered, others of which were in common. Additionally, 34 performance outputs were evaluated in order to predict the behavior in terms of stress for the durability analysis, intrusion and VPI to ensure the safety standards and the bending and torsion modes to meet the comfort level objectives.
Running the test for the single-domain workflows for safety and durability on the 32-CPU HPC system was fast, taking only three hours. The execution of the whole vehicle optimization took 8 days. Regarding the implementation of the MDO approach to the case, Yan Fu says, “In the past, it would have taken us a month or more to collect all the models, formulate the MDO problem and build the workflow. Now, with SOMO tightly integrated with Ford private cloud computing and IT security systems, attribute engineers can do their own work and publish the latest models so the MDO team can build the top-level MDO process right away. This is something new that has revolutionized our way of working.”
Thanks to the execution summary dashboard, the values of design variables, constraint and objective trends were monitored runtime and the team could follow the evolution toward the optimal solutions. The final results of the MDO workflow execution were then analyzed using the post-processing tools available directly through the web interface. Yan Fu adds: “The SOMO broken constraint tool helps us draw attention to where the problem is and guides the designer by indicating where the effort should be focused in order to solve it. On the other hand, the parallel chart helps select the optimal configuration among the design candidates across attributes.” Engineers were thus able to see the incidence and number of designs violating the constraints using the Broken Constraint Chart and together with the Parallel Coordinates Chart were able to spot patterns in variable behavior and the correlations between the variables. The outcome of this sophisticated, web-enabled analysis was the choice of the final design that resulted in a significant mass reduction of 1.36% compared to the baseline configuration, valid for all three pickup truck frames considered.