Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
105 CHAPTER 3 SURVEY OF MODELING BEST PRACTICES INTRODUCTION Identifying model building best practices provides a means for the roadside safety computational mechanics community to capture its best practices in easily retrieved form so that both new and experienced users can develop models that are highly likely to run without errors. Unfortunately, most papers and reports that present the results of finite element simulations rarely reveal all the details of the simulations so there is relatively little to be found in the literature with specific recommendations for, say, minimum time steps or the maximum change in total energy. Likewise, most papers and reports on verification and validation do not address this subject because it is highly specific to the application. For example, parameter variations that may be perfectly acceptable in roadside safety may have little relevance to computational fluid dynamics problems. Since there is little in the literature to help define reasonable parameter variations, a survey of practitioners in the art of roadside safety computer simulations was conducted. The original survey form and the tabulated responses are included in Appendix D. The survey asked practitioners what types of techniques they use and what range of variation they considered acceptable when performing typical roadside safety simulations. The results of the survey will be discussed for each parameter in the following sections; more information about the survey can be found in Appendix D. Several other solid mechanics communities appear to be in the process of developing best practices guides although none appear to be very far along at this point. ASME PTC-10, for example, is planning a series of documents that expand on the general framework of the ASME V&V guide. These documents will constitute a series of best practices guides on a variety of topics including model building, incorporating uncertainty, calculating metrics, etc. The aircraft seat committee of SAE is also in the process of developing a best practices guide to complement the FAA Circular Advisory. This guide will describe typical model procedures like minimum time steps, the use of mass scaling, element dimensions and type among other model building details. VERIFICATION Definition Fifty seven percent of practitioners agreed with the definition of verification given in the survey. Geometry Generation Survey responses show that geometries for roadside hardware models are generated from drawings and based on previously developed successful models. The vehicle models, on the
106 other hand, are mostly obtained from NCAC library at George Washington University. Other public domain sources are sometimes used to obtain vehicle models. Purchasing proprietary models as well as obtaining models from the automotive industry was not common among the surveyed practitioners. One out of three practitioners that participated in the survey sometimes chooses to build their own vehicle model. The single unit truck, reduced C2500 pickup truck and detailed and reduced Geo Metro were selected as the most frequently used models from the NCAC. All four of these vehicle models, however, were often used after modifications by practitioners. The remainder of the vehicles available at the NCAC vehicle library seems to be rarely used by survey respondents. Practitioners were also asked whether they were using any updated vehicle models. Based on the responses, the updated version of Geo Metro model by Politecnico di Milano was most frequently used followed by the F800 single unit truck and the C2500 pickup truck model. Practitioners agree that articulated suspensions, rolling tires and coarser meshes are highly desirable model features at least in the area of roadside safety. Detailed geometry, failing tires and vehicle component failure are considered as medium important modeling features, unless a particular crash scenario requires a more detailed model. Also, responses from practitioners varied when they were asked about whether they used detailed or kinematic equivalent systems to model complex articulated systems, such as suspensions and steering systems. 32 percent of the surveyed responded that they mostly used detailed modeling, whereas 40 percent of the surveyed said that they mostly used kinematic equivalent systems. Most practitioners used approximation when modeling bolts (62%), rivets (93%), welds (87%), soil (62%) and road surface (90%). Some practitioners, on the other hand, choose to use detailed bolt and soil models, probably in response to specific project needs. 58 percent of practitioners agreed that material properties of models are mostly obtained from laboratory experiments. Sometimes they obtain material properties from material specifications and other successful finite element simulation studies. According to the survey, material failure is frequently determined from maximum effective plastic strain followed by element erosion (i.e., 88 percent and 73 percent, respectively). Some of the practitioners choose not to specify any failure in their models. Failure parameters in models are mostly obtained from experimental data and previously used successful models. Results show that 61 percent of practitioners use strain rate sensitive material models. Welded connections are sometimes modeled as merged nodes or as tied contacts (i.e., 47 percent), however, welds are most often (i.e., 52 percent) modeled as spot welds with failure. Modeling welds with tied contacts with failure is not very common. Bolted connections, as opposed to welded connections, are never represented with tied contact or tied contact with failure. Springs are sometimes or rarely used to model bolted connections (i.e., 37 percent). According to survey results, bolted connections are mostly modeled with spot welds with failure and sometimes with merged nodes if failure is not expected. On the other hand, a significant
107 number of practitioners (i.e., 38 percent) choose not to use merged nodes to represent bolted connections. Finally, when practitioners were asked about how they model post-soil interaction, the survey determined that the majority (i.e., 52 and 48 percent) never used Eulerian solid meshing and fixed nodes, respectively. Some practitioners favored the use of nonlinear springs (i.e., 37 percent) to model soils. 31 percent of the participants typically use explicit geometric models of the post and soil continuum and an equal percentage who never use them. Mesh Sensitivity and Quality Determination Practitioners were asked about the largest shell element dimensions they use in modeling steel components. The average smallest element dimension when used in a contact region was 23 mm. Outside of the contact region but expected to deform significantly it was 44 mm. Outside of the contact region and where no deformation was expected, it was 93 mm. When the question asked was about the largest solid element dimension to be used in modeling a wood post, the average solid element dimension resulted was 21 mm in case the post was expected to fracture, 33 mm in case the post was expected not to fracture but displace in the soil and 56 mm for posts which did not experience any deformation. These results are shown in Table 1 and Table 2. Table 1. Average shell element dimensions used in simulation studies Contact Region Not in contact region but expected significant deformation Not in contact nor deformation region Shell Element (mm) 23 44 93 Table 2. Average solid element dimensions used in simulation studies Fracture Region Not expected to fracture but displace in the soil Not expected to deform nor displace significantly Solid Element (mm) 21 33 56 Practitioners generally (i.e., 38 percent) try to keep the maximum warpage angle of the elements in the model smaller than 5 degrees. Also, some practitioners (i.e., 28 percent) accepted warpage angles as large as 10 degrees. The majority of practitioners (i.e., 48 percent) try to keep the percentage of elements with the biggest warpage angle in the model between zero and five percent. Also, a large number of practitioners (i.e., 45 percent) prefer an aspect ratio smaller than 3 to 1 followed by 5 to 1 (i.e., 23 percent) and 2 to 1 (i.e., 19 percent) when creating a meshing. Contact Stability Issues Practitioners usually choose (i.e., 38 percent) to refine the mesh in the contact area when contact instabilities occur during a simulation. A larger number of practitioners (i.e., 52 percent)
108 choose to modify the contact parameters like the penalty factors when they experience contract problems. There is not a clear answer from the practitioners on whether changing the contact type would solve the contact instability problem. Energy Balance and Comparisons Survey results show that as long as the variation in total energy and added mass are less than five percent practitioners (i.e., 52 percent) are generally not concerned about the energy balance and mass increase. Similarly, as long as the ratio between the hourglass energy and total energy is less than 10 percent practitioners (i.e., 42 percent) are generally not concerned. Time Step Issues Regarding Element Size, Mesh Density and Mass Scaling The majority of practitioners (i.e., 77 percent) prefer an initial time step in the range 1.0E-06 sec to 5.0E-06 sec. Very few (i.e., 20 percent) choose to accept initial time steps less than 1.0E-06 sec. To control the time step, all practitioners control both element dimensions and use the minimum time step option (i.e., mass scaling). Most practitioners (i.e., 60 percent) generally used some mass scaling in their models to keep the time step at a desirable level and prevent run times from becoming excessive. The majority of practitioners (i.e., 75 percent) agreed that they make sure the overall mesh is adequate for the contacts and the expected deformations. They use mass scaling to prevent the time step from getting too small. VALIDATION Almost all practitioners (i.e., 91 percent) agreed with the definition of validation given in the survey and taken from ASM V&V10-2006. When they were asked about the order of importance in validating a finite element simulation with a physical test, the results were very distinct. Other than the qualitative comparison of vehicle damage, which was chosen to be not so important, all other statements were deemed to be very important. Qualitative comparisons of barrier damage were favored most by the practitioners (i.e., 48 percent) followed by the quantitative comparison of displacement/rotation time histories (i.e., 39 percent) and qualitative comparison of crash sequence (i.e., 36 percent). Qualitative comparisons of velocity time histories, trajectories and quantitative comparison of acceleration time histories came next in the list (i.e., 32, 30 and 26 percent, respectively). Practitioners agreed that they always use qualitative comparisons, such as hardware deformations, location/number of failed components, visual methods and phenomological response to compare simulation results to the physical tests. They also agreed that they mostly use velocity-time histories, always use displacements and rotations and sometimes use stress/strain relationship to make quantitative comparisons. Responses on acceleration time histories were distributed evenly among always (i.e., 32 percent), mostly (i.e., 29 percent) and sometimes (i.e., 23 percent) , while responses on energy balance comparisons were sometimes or rarely performed.
109 TRAP parameters like OIV and ORA were the most common metrics used by the practitioners (i.e., 64 percent) as shown in Appendix D. TRAP was by far the most commonly used metric in publications as discussed earlier in the literature review. Other metrics, such as Geerâs MPC metrics, the NARD Validation Manual metrics and the ANOVA metrics were rarely used by the practitioners surveyed. When asked about whether or not practitioners used a standard procedure for model verification and validation the most common answer was no (i.e., 59 percent). Similarly, when practitioners were asked about whether they used different validation methods according to specific roadside hardware studied, a majority answered ânoâ (i.e., 60 percent) . The majority of practitioners (i.e., 67 percent) used LS-PrePost to filter their simulation data followed by those who used SAE J211 (i.e., 47 percent). Some practitioners (i.e., 30 percent) reported using TRAP or other methods to filter simulation data. The majority of practitioners (i.e., 81 percent) used LSDYNA as their analysis code for crash simulation. Finally, almost all practitioners (i.e., 94 percent) validate components or subassemblies of larger models whenever possible before validating the overall model.
110
