Volkswagen engineers used automated optimization methods to reduce injuries in car-to-car-collisions
Volkswagen engineers used automated optimization methods to demonstrate how the design of a vehicle could be changed to reduce injuries in car-to-car-collisions. “Compatibility” measures the tendency of a vehicle to injure the occupants of another vehicle involved in a collision. It is an issue of growing concern. In the past, engineers studied compatibility by building a finite element model that included both vehicles and manually changed the design of the striking vehicle while measuring the effect on a crash dummy in the struck vehicle. Recently, Volkswagen engineers Martin Meywerk and Roel Kersten substantially improved on that method by using LMS OPTIMUS automated design optimization software with PAM-CRASH to explore the full range of design alternatives for the simplified front end of the struck vehicle
The optimization software drove a finite element based crash analysis code to construct a simplified polynomial model that described the crash-dummy loads in the struck vehicle as a function of the striking vehicle design variables. Using this approach, they showed that a better insight in the interaction between two vehicles can be gained. When a validated simplified model is available, a parameter study is finished after about three days of modeling and setup and 15 days of computer simulation, compared to about four months that would have been required using previous methods.
The crashworthiness performance of passenger vehicles has traditionally been evaluated on the results of a series of well-defined laboratory crash tests. These tests, by their nature, focus on evaluating and minimizing injuries to the occupants in the subject vehicle. However design modifications that minimize injuries in one vehicle have the potential of increasing injury levels in the other vehicle. The compatibility of a specific vehicle is controlled by its weight, geometry and structure. Important geometrical factors include the hood profile, sill height, and bumper height. Structural factors include frontal stiffness as determined from crash tests and engine location.
Applied modeling method
In this example, Volkswagen engineers examined two modified vehicles to see if the compatibility of the striking vehicle could be increased. Their focus was on the effects of changing the stiffness distribution of the front structure of the vehicle while calculating the dummy load in another vehicle struck from the side at 90 degrees and 50 km/hr. The engineers used a simplified finite element model of the front structure of the striking vehicle that was developed using PAM-CRASH crash analysis software. The three dimensional stiffness distribution of the front structure was represented by 72 beam elements with individually adjustable nonlinear force displacement characteristics.
The vehicle that was struck from the side was created with a newly developed modeling method for the vehicle side structure. This method has been developed and investigated in a Ph.D. thesis by Kersten. In this simplified model, a grid of beam elements replaced the internal vehicle structure. The outer panels were taken from a conventional finite element model. Spacing beams interconnected the grid of beam elements and external plates. The beam elements had axial and bending stiffness properties equivalent to the complex finite element structure they replaced. The stiffness characteristics, such as axial and bending stiffness over force-displacement bending moment angle curves, were obtained from finite element analysis of the part in question. The simplified side structure was connected to nodes in the middle of the vehicle through beam elements. These elements simulated the support of the side structure provided by the roof, floor panel construction, dashboard and vehicle rear section. The nodes in the middle of the vehicle and the nodes defining the vehicle-road surface contact were joined in a rigid body definition. The inertia of the vehicle without the modeled side structure was assigned to this rigid body. The door trim panel and the driver’s seat of the conventional FE-model were mounted in the simplified model. The crash safety was evaluated by assessing the load of a PAM-CRASH EuroSID-1 FE-dummy on the driver’s seat.
Automating the optimization process
The resulting model is capable of determining the compatibility of the design for any combination of design parameters. The challenge comes from the enormous number of potential design parameters and combinations: it takes about 10 hours to evaluate each combination by setting up a potential design, analyzing the model (analogous to physical testing) and starting all over again in an effort to solve the problems revealed by the analysis. The results of the analysis provide little or no direction on what is needed to resolve it. While the engineer can determine over time whether or not he or she is improving the design, it’s essentially impossible to ever optimize the aggressivity of the design, regardless of how many months are spent trying.
Over the last several years, in an effort to find a better approach, Volkswagen engineers have made increasing use of software packages designed to drive their crash simulation software and other analysis tools to an optimum solution without manual intervention. On this problem, they selected LMS OPTIMUS for two primary reasons, according to Meywerk. “OPTIMUS’ graphical interface makes it very easy to define the optimization procedure,” he said. “In addition, OPTIMUS was one of the first programs to support parallel processing. The fact that it works with the LSF queuing system makes it possible for us to harness the computing horsepower of idle workstations located throughout our facilities.
Generating a response surface model
Kersten defined five stiffness areas in the front crash model, which in the original design contained beams with identical stiffness. The stiffness in these areas was approximated with force-displacement curves with two constant force levels. The constant force levels were defined as the design variables of the optimization, resulting in a total of ten design variables. He defined the optimization goal as minimizing the average of nine dummy load values, which are normalized by the prescribed limiting value of the injury criteria. He also specified that the limiting values for the individual injury criteria as defined by government regulations could not be exceeded.
Meywerk and Kersten used design of experiments to reduce the number of design points that needed to be computed in order to determine an optimum value. They established an experimental design using the Latin Hypercube method with a total of 70 experiments. The experiments were used to define a response surface model, an arbitrary function that is used to approximate the shape of the exact response function. This approach is known as response surface methodology (RSM). A second order Taylor polynomial with 29 coefficients was used to define the model. The model approximated the shape of the exact objective function, making it possible to estimate the system response in far less time than would be required to evaluate every possible design alternative. A scatter plot which compared the results of the RSM model to actual simulation runs showed very little error, indicating that the RSM model accurately predicted the value of the objective function over the design space.
Iterating to an optimum solution
Volkswagen engineers then used the sequential quadratic programming optimization algorithm contained within OPTIMUS to identify the optimal value on the RSM. After the best design was found with this model, the design was re-analyzed with exact data. The best value within the prescribed design space provided a reduction in the total dummy load in the struck vehicle of 25%. This reduction was accomplished by lowering the stiffness of the outer structure, the longitudinal members and the upper structure of the front end to the lower boundary of the design space. The stiffness of the bumper area and the lower structure of the front end, on the other hand, were raised to the upper boundary of the design space. The high stiffness in the area of the bumper and the lower structure in the optimum can be explained by the seat-dummy interaction. The simulation shows that as a result of a higher stiffness in the area of the lower structure, the seat is struck harder, which results in a higher force acting through the back of the seat on the back plate of the dummy. The movement of the dummy away from the intruding structure is increased by this interaction. This reduces the upper body load. It should be emphasized that the interaction of the back plate and the seat is artificial in the sense that a human being has no back plate. That means that for Volkswagen the real goal of any side impact investigation, i.e. the protection of the human being, is not reached with the reduction of dummy load. It is, however, the first step in this direction.
“We aren’t waiting for government regulations to require us to design cars that have higher compatibility, but rather we are moving proactively to address the issue,” Meywerk said. “This study showed how the crash compatibility of1 vehicles can be studied. The key to a fast solution is using the latest optimization methods. It would have taken several months of analyst effort to reach a viable solution through the manual use of traditional crash analysis tools. But with LMS OPTIMUS we were able to set up the problem in only a few days and solve it in parallel on available workstations in about three weeks. This is clearly the ideal approach for any automotive OEM interested in addressing this increasingly important subject.”
Volkswagen engineers used automated optimization methods to demonstrate how the design of a vehicle could be changed to reduce injuries in car-to-car-collisions. “Compatibility” measures the tendency of a vehicle to injure the occupants of another vehicle involved in a collision. It is an issue of growing concern. In the past, engineers studied compatibility by building a finite element model that included both vehicles and manually changed the design of the striking vehicle while measuring the effect on a crash dummy in the struck vehicle. Recently, Volkswagen engineers Martin Meywerk and Roel Kersten substantially improved on that method by using LMS OPTIMUS automated design optimization software with PAM-CRASH to explore the full range of design alternatives for the simplified front end of the struck vehicleThe optimization software drove a finite element based crash analysis code to construct a simplified polynomial model that described the crash-dummy loads in the struck vehicle as a function of the striking vehicle design variables. Using this approach, they showed that a better insight in the interaction between two vehicles can be gained. When a validated simplified model is available, a parameter study is finished after about three days of modeling and setup and 15 days of computer simulation, compared to about four months that would have been required using previous methods.
The crashworthiness performance of passenger vehicles has traditionally been evaluated on the results of a series of well-defined laboratory crash tests. These tests, by their nature, focus on evaluating and minimizing injuries to the occupants in the subject vehicle. However design modifications that minimize injuries in one vehicle have the potential of increasing injury levels in the other vehicle. The compatibility of a specific vehicle is controlled by its weight, geometry and structure. Important geometrical factors include the hood profile, sill height, and bumper height. Structural factors include frontal stiffness as determined from crash tests and engine location.
Applied modeling method
In this example, Volkswagen engineers examined two modified vehicles to see if the compatibility of the striking vehicle could be increased. Their focus was on the effects of changing the stiffness distribution of the front structure of the vehicle while calculating the dummy load in another vehicle struck from the side at 90 degrees and 50 km/hr. The engineers used a simplified finite element model of the front structure of the striking vehicle that was developed using PAM-CRASH crash analysis software. The three dimensional stiffness distribution of the front structure was represented by 72 beam elements with individually adjustable nonlinear force displacement characteristics.
The vehicle that was struck from the side was created with a newly developed modeling method for the vehicle side structure. This method has been developed and investigated in a Ph.D. thesis by Kersten. In this simplified model, a grid of beam elements replaced the internal vehicle structure. The outer panels were taken from a conventional finite element model. Spacing beams interconnected the grid of beam elements and external plates. The beam elements had axial and bending stiffness properties equivalent to the complex finite element structure they replaced. The stiffness characteristics, such as axial and bending stiffness over force-displacement bending moment angle curves, were obtained from finite element analysis of the part in question. The simplified side structure was connected to nodes in the middle of the vehicle through beam elements. These elements simulated the support of the side structure provided by the roof, floor panel construction, dashboard and vehicle rear section. The nodes in the middle of the vehicle and the nodes defining the vehicle-road surface contact were joined in a rigid body definition. The inertia of the vehicle without the modeled side structure was assigned to this rigid body. The door trim panel and the driver’s seat of the conventional FE-model were mounted in the simplified model. The crash safety was evaluated by assessing the load of a PAM-CRASH EuroSID-1 FE-dummy on the driver’s seat.
Automating the optimization process
The resulting model is capable of determining the compatibility of the design for any combination of design parameters. The challenge comes from the enormous number of potential design parameters and combinations: it takes about 10 hours to evaluate each combination by setting up a potential design, analyzing the model (analogous to physical testing) and starting all over again in an effort to solve the problems revealed by the analysis. The results of the analysis provide little or no direction on what is needed to resolve it. While the engineer can determine over time whether or not he or she is improving the design, it’s essentially impossible to ever optimize the aggressivity of the design, regardless of how many months are spent trying.Over the last several years, in an effort to find a better approach, Volkswagen engineers have made increasing use of software packages designed to drive their crash simulation software and other analysis tools to an optimum solution without manual intervention. On this problem, they selected LMS OPTIMUS for two primary reasons, according to Meywerk. “OPTIMUS’ graphical interface makes it very easy to define the optimization procedure,” he said. “In addition, OPTIMUS was one of the first programs to support parallel processing. The fact that it works with the LSF queuing system makes it possible for us to harness the computing horsepower of idle workstations located throughout our facilities.
Generating a response surface model
Kersten defined five stiffness areas in the front crash model, which in the original design contained beams with identical stiffness. The stiffness in these areas was approximated with force-displacement curves with two constant force levels. The constant force levels were defined as the design variables of the optimization, resulting in a total of ten design variables. He defined the optimization goal as minimizing the average of nine dummy load values, which are normalized by the prescribed limiting value of the injury criteria. He also specified that the limiting values for the individual injury criteria as defined by government regulations could not be exceeded.
Meywerk and Kersten used design of experiments to reduce the number of design points that needed to be computed in order to determine an optimum value. They established an experimental design using the Latin Hypercube method with a total of 70 experiments. The experiments were used to define a response surface model, an arbitrary function that is used to approximate the shape of the exact response function. This approach is known as response surface methodology (RSM). A second order Taylor polynomial with 29 coefficients was used to define the model. The model approximated the shape of the exact objective function, making it possible to estimate the system response in far less time than would be required to evaluate every possible design alternative. A scatter plot which compared the results of the RSM model to actual simulation runs showed very little error, indicating that the RSM model accurately predicted the value of the objective function over the design space.
Iterating to an optimum solution
Volkswagen engineers then used the sequential quadratic programming optimization algorithm contained within OPTIMUS to identify the optimal value on the RSM. After the best design was found with this model, the design was re-analyzed with exact data. The best value within the prescribed design space provided a reduction in the total dummy load in the struck vehicle of 25%. This reduction was accomplished by lowering the stiffness of the outer structure, the longitudinal members and the upper structure of the front end to the lower boundary of the design space. The stiffness of the bumper area and the lower structure of the front end, on the other hand, were raised to the upper boundary of the design space. The high stiffness in the area of the bumper and the lower structure in the optimum can be explained by the seat-dummy interaction. The simulation shows that as a result of a higher stiffness in the area of the lower structure, the seat is struck harder, which results in a higher force acting through the back of the seat on the back plate of the dummy. The movement of the dummy away from the intruding structure is increased by this interaction. This reduces the upper body load. It should be emphasized that the interaction of the back plate and the seat is artificial in the sense that a human being has no back plate. That means that for Volkswagen the real goal of any side impact investigation, i.e. the protection of the human being, is not reached with the reduction of dummy load. It is, however, the first step in this direction.
“We aren’t waiting for government regulations to require us to design cars that have higher compatibility, but rather we are moving proactively to address the issue,” Meywerk said. “This study showed how the crash compatibility of1 vehicles can be studied. The key to a fast solution is using the latest optimization methods. It would have taken several months of analyst effort to reach a viable solution through the manual use of traditional crash analysis tools. But with LMS OPTIMUS we were able to set up the problem in only a few days and solve it in parallel on available workstations in about three weeks. This is clearly the ideal approach for any automotive OEM interested in addressing this increasingly important subject.”
