大学生考试网 让学习变简单
当前位置:首页 >> 交通运输 >>



Parametric Design/Analysis with MSC/PATRAN - A New Capability
James G. Crose, Douglas A. Marx, Mark Kranz, Paul Olson and Carl Ball The MacNeal-Schwendler Corporation 2975 Redhill Avenue Costa Mesa, California 92626 (714) 444-5050 greg.crose@macsch.com

This paper describes a new capability of MSC/PATRAN to provide for automated parametric analysis in support of complex design processes. Computational resources are available today that can efficiently permit at least an order of magnitude more analysis support to the design process than was available only a few years ago. MSC has been participating in DARPA’s RaDEO (Rapid Design Exploration and Optimization) project with Ford Motor Company and the Rocketdyne and St. Louis divisions of Boeing Aircraft Company. That project has developed a new computer program to facilitate robust design processes that involve orders of magnitude more analytical simulation than is typically applied in design. The Robust Design Computational System (RDCS) computer program provides for automation of design processes such as parametric design scanning, application of Taguchi concepts, optimization and probabilistic analysis. It (RDCS) depends on the automation of multi-disciplinary parametric math models that simulate the behavior of the object being designed. It is this requirement for automation that is addressed in the paper. MSC is supporting RDCS by creating a powerful parameterization capability with the MSC/PATRAN pre and post-processor. The present result of this effort is a modification of MSC/PATRAN that permits the use of named variables to replace the usual fixed numerical values of the modeling parameters. These variable names are captured in the session file along with a default value. In addition, the values of these parameters different than the prescribed default can be provided by an external file that can be produced by another code such as the RDCS code referred to above. The goal of the MSC/PATRAN parameterization project is to make it possible for the user to use names and default values for variables (parameters) in every entry point on every form that can be accessed for modeling purposes. This goal has now been met for a large fraction of all the MSC/PATRAN forms. Similarly, we have provided for the definition and output of named response parameters such as maxima and minima of stress, strain, displacements and complex functions of results ( e.g.,Von Mises stresses or other measures of failure). These responses are directed to an output file for use by other codes such as the RDCS code referred to above. Since a MSC/PATRAN session file can be re-run in a batch mode, including running the analysis preference, the parameterized version of this file can also be executed in batch mode. This makes it possible to simulate the response of an unlimited number of design variations and capture the responses as a function of the parametric variables and do so in batch mode without user intervention.


This paper presents a description of this new capability that can be added to MSC/PATRAN and shows numerous examples of its use. Also, the coupling with RDCS is demonstrated. Introduction Detailed Finite Element Analysis (FEA) is mostly used in the mode of design certification or for trouble shooting. Although desirable as an aid to conceptual design and at other stages in the design process, the cost and time to accomplish it are frequently prohibitive. The present paper addresses one step in the evolution of design analysis that is necessary to position complex analytical simulation further towards the beginning stages of a products development. The payoff for this can be tremendous. The fact that FEA is frequently used in an after the fact trouble shooting mode suggests that it could have been used earlier to prevent the trouble! The cost of correcting a design flaw was reflected in a study done by the German automotive industry (Reference 1) with the following result: ? ? ? ? ? Suppose that a vehicle design has a serious flaw. When this flaw is detected at the conceptual design stage the cost for fixing it will be, for example, one unit. If this flaw is detected during the detailed design and analysis phase the cost will be 10 units. If the problem occurs when building prototypes, 100 units will be required. Finally, if the flaw is detected during production, the cost will increase to 1000 units.

It appears that processes that improve the quality and lower the cost of complex analytical simulation has great value to industry. Detailed FEA can play an important role in such processes. There are three major developments required to achieve the short time and low cost of implementing detailed FEA at earlier stages of design: ? ? ? Computational efficiency. Design scan software. Automated parametric math models.

Improvements in computational efficiency are taking place quite well without our intervention. New developments in CPU speed, lower cost of memory and mass storage and speedy networks provide the base upon which we can accomplish the other necessary developments. The Robust Design Computational System (RDCS) is one example of design scan software. It is also being reported on at this conference. It and the developments reported here have been partially funded by DARPA as part of their multi-year Rapid Design Exploration and Optimization (RaDEO) Program that is nearing its end. RDCS is capable of producing orders of magnitude more design explorations via automated parametric math models. These math models


can have any number of components called functional models. Each functional model represents one part of a products evaluation or performance simulation. They can be independent or dependent on other functional models. RDCS permits the construction of a master math model with all the dependencies accounted for and can execute that math model over a network of many computers and achieve parallelism to the extent permissible by the computer network and functional model dependencies. Basically, RDCS iterates through a large number of design instances. Each design instance is a realization of the effect of setting the values of all the variable design parameters as input to the math model. Upon completion of the math model execution, output response parameters are collected from the functional models and associated with the input parameter values. By so doing, any one of the available design tools in RDCS can be applied systematically. For example, design space can be explored and defined by sampling parameter values within defined limits, Taguchi methods can be applied to address robustness, sensitivity coefficients can be defined for each of the parameters, optimization can be accomplished iteratively, and probabilistic methods can be applied to define probability of failure. All of these functions can be applied through a simple and easy to understand graphical user interface. However, the entire process depends on automation and parameterization of the math model. It is this requirement that dictated the developments discussed in this paper. It was desired to provide a means for automating and parameterizing finite element based product simulations. MSC/NASTRAN is one example of a finite element code that can be a desirable part of any comprehensive math model representing a product’s function and for evaluating its survivability in its intended environments. The overall capabilities of MSC/PATRAN as a pre and post processor to MSC/NASTRAN and many other finite element codes, makes it an ideal vehicle for achieving automation of parametric FEA simulations for use with RDCS and other design scan managers. Making the key design parameters variable quantities that are then implemented in the modeling process does this. The model, its material constituency, its boundary conditions and its loading can all be variable, thereby representing an unlimited number of design alternatives. This can be done for static and dynamic structural analysis, thermal analysis, aerodynamic analysis and any of the many other computational processes served by MSC/PATRAN. The important contribution is the parameterization of a complete analytical simulation, not just geometric descriptions as are available in many CAD programs. Problem Definition It was desired to be able to use named variables as substitutes for quantified values as entries into any form presented to the modeler while using MSC/PATRAN. This would enable him to describe a range of models, each one defined by setting the values for all the variables. It was also desired to be able to re-model and re-analyse in batch mode (non-interactive) for subsequent variable definitions. The session file was adopted as the medium for re-modeling. Also, there must be a default value for each variable in order that the modeling process can be accomplished in the first place. Finally, there must be a way to define current values of design variables without user interaction. It was decided that an external file could provide these values and could be created by another program such as RDCS. The MSC/PATRAN code had a feature called “global variables” that had much of the functional capability desired. Unfortunately, many


of the sub-forms lacked the ability to take full advantage of “global variables.” The problem of making the variable name and its default value and a reference to an external file persist in the session file for re-running required considerable programming effort to accomplish. This has now been accomplished for the majority of the graphical user interface. We have also developed a new adaptive form for the GUI that is used for the purpose of defining input variables and their default values as well as manipulation of results to define meaningful results parameters for output to an external results file. It is now possible to conduct re-modeling, analysis processing and results definition entirely in batch mode by submitting the session file for execution and providing an input file with the new input parameter values. Description of the MSC/PATRAN Parameterization Feature A new toolbox implemented as an adaptive form accessed from the main menu bar and called “Parametric Modeling” has been developed. Figure 1 shows the toolbox/form in its variables definition mode. This form is used to define a variable name and its default value to be used during the modeling process. Variable types can be integer, real, or string and arrays of those types. This particular view indicates that several variables have already been defined. They are height, width, enf_disp (a vector), mesh_density, pressure and thickness. A new variable can be established by simply typing in a name, setting its type and giving it a current value. An optional description can also be added. New variables can also be defined using other variables and any of the extensive collection of available functions in MSC/PATRAN. The context of their usage must of course be consistent. Variable names can then be used in any other MSC/PATRAN form and are denoted by enclosing the name in the back tick symbol, as was the case for global variables. Upon completion of a modeling step, it can be seen in the session file that the variable name is provided along with its default value and a reference to an external file where subsequent different values can be set. When re-played, MSC/PATRAN will look for that file, read its contents to determine if there is a new value for the variable and if so, use that value instead of the default. Note that the modeling process must be accomplished in a sequential fashion. Later re-definition of a parameter value will not retroactively result in the imagined changes. Therefore, proper use of a variable depends on the sequence of modeling operations. Figure 2 shows the “Parametric Modeling Toolbox” in the mode of creating response variables. This view of the form is available after an analysis has been completed with the default value of all the input parameters. The purpose of the form in this context is to search through and manipulate the results of analysis to define meaningful metrics for export to an external file. Examples of such metrics might be maximum Von Mises stress in one previously defined group or material region or maximum strain in the fiber direction of a particular composite material, or any number of other criteria that can be developed based on the results coming back from the analysis code. Figure 2 shows that the user has already captured the maximum displacement, and maximum Von Mises stress from the results. Note that the values determined are also shown. The user has just completed finding the maximum Von Mises stress at a node from the stress tensor at that node with that result coming from the static subcase. Output variable types include nodal or element scalar, vector or tensor quantities. Results may be sorted by maximum, minimum, absolute maximum or absolute minimum value. Sorting


may be based on element or node ID, all entities in a graphics window, all in a user defined group, material ID or property ID. Results can be sorted to designated ply for composite materials. Mass properties can also be output with this form. The form changes based on selections made and only those parts of the overall analysis results applicable to those choices are allowed as further choices. Thus, the user is made aware of all the factors affecting his specification of outputs.

Figure 1. Variable Creation


Figure 2. Response Variable Creation

Application and demonstrations of the parametric capability The parameterization modification of MSC/PATRAN has been applied to a number of example problems. These include design of a composite tubular strut, modeling of a tire and design of a composite wing structure. The following sections present a summary of these examples.


Cure Cycle Shrinkage Stresses in a Composite Strut This problem consists of a square cross-section hollow composite strut designed for use in a space structure as illustrated in Figure 3. The strut consists of nine graphite epoxy plies. The longitudinal plies are high modulus and the +/- 45 plies and the 90 plies are low modulus. They are laid up at 0,90,0,-45,0,+45,0,90,0. The problem was studied previously by Stanton and Mack (Reference 2) in 1986. Due to the geometry and ply orientation distribution, the strut twists and develops residual stress due to cure cycle shrinkage. Since the weight critical design requires maximum zero degree reinforcement while also providing strength and stiffness in the torsion and hoop modes, the imbalance due to +/- 45 plies must be present. However, the resulting twist and residual stresses were unacceptable as reported by Stanton and Mack. We wondered if there might be a better design possible by perturbing the orientation of the 90 and 45 plies. MSC/PATRAN was used to model this structure for MSC/NASTRAN solution with the angles to the reinforcements as variable parameters. The parameterized session file was prepared as an RDCS math model. A design scan was completed with the RDCS code. RDCS produced the design surfaces shown in Figures 4 and 5. The constraint limits of the “90” plies from 80 to 100 and the “+/- 45” plies from +45 to +60 and -45 to –60 were chosen to represent the minimum required strength/stiffness in hoop and torsion to satisfy design requirements. It is seen from these figures that a lay-up of 0,95,0,60,0,-60,0,85,0 produces a design that greatly mitigates the amount of twist and residual stress while maintaining the minimum weight design and satisfying the torsional and hoop stiffness/strength constraints. This type of design investigation can be accomplished quite quickly with a parameterized model used with the RDCS code. In one day, we were able to make a substantial improvement in the design. In the original study, several man-months were expended on testing and computation. The ability to perturb design parameters in a good simulation up front in a design process has been shown to be extremely valuable. Rubber Tire Model A project was conducted to parameterize the design of a passenger car tire. The resulting model is shown in Figure 6. The 40,000 degree of freedom model of a 30 degree sector consists of parametric descriptions of the tread, two radial ply composites, two bias steel belt composite, wire bead, rim cushion, bead wrap, filler, inner liner and chafer. The mesh strategy was also parameterized to preserve high quality modeling while making large changes in the geometry of the constituent materials and the overall tire size. The loading was also parameterized to allow for variable inflation pressure, vehicle speed, and vehicle weight. A loaded patch shape was assumed for the purpose of demonstration. Figure 7 shows some of the more extreme possible permutations of the base line design. Each model is based on exactly the same session file, but with changes in the variable parameter values. The passenger tire model was also used as a math model in RDCS where a design scan and design sensitivity studies were completed. Figure 8 shows the results of a design scan on vehicle speed as it effects maximum displacement, maximum Von Mises stress, maximum shear stress and maximum principal strain. Figure 9 is the same type of design scan but for inflation pressure instead of vehicle speed. Figure 10 shows the result of a sensitivity study where the sensitivity of the analytical results on the design parameters: velocity; inflation pressure; outer ply thickness; buttress radius and; tread modulus


were obtained from use of the RDCS code. Once a parametric model and associated simulation analysis are available, the RDCS code can accomplish orders of magnitude more design studies of a given object than would be possible without parametric models.

Figure 3. Composite Space Structure Strut

Figure 4. Twist Due to Cure Shrinkage


Figure 5. Residual Stress Due to Cure Shrinkage

Figure 6. Rubber Tire Model

Default Passenger

Truck Tire

Racing Tire


Figure 7. Permutation of Rubber Tire Designs

Figure 8. Sensitivity Analysis Results

Figure 9. Design Scan Results for Velocity

Figure 10. Design Scan Results for Inflation Pressure


Composite Aircraft Wing Study An aircraft wing in the shape and size of the F22 aircraft was modeled in MSC/PATRAN for static analysis by MSC/NASTRAN. The base line model is shown in Figure 11. The model consists of a composite skin, spars, ribs, leading edge flaps, trailing edge flap, aileron and wing tip. 59 variable input parameters were used which give external control to the overall geometry, meshing, loading, composite construction and material properties. It has 14,000 degrees of freedom in its base line configuration. The flexibility of this model is illustrated in Figure 12, where it can be seen that a single MSC/PATRAN session file can be manipulated with the input parameters to produce a complete engineering model of wings shaped and sized as an S37 forward sweep wing, a B2 wing with 28,000 degrees of freedom and even a propeller configuration with twist. The wing is loaded with pressure using linearized supersonic shock analysis, which allows the loading to be variable with velocity, altitude and angle of attack.

Figure 11. F22 Style Wing Model


~ 14,000 DOF



~ 28,000 DOF

Figure 12. Permutations of Wing Designs


Conclusions MSC is providing a parameterization capability for use with MSC/PATRAN which will dramatically improve the ability to create automated parametric math models for use with design scan management software such as RDCS. The linkage with such software packages is very simple. It only requires the preparation of an input file providing the parameter values for each design instance, a script for executing MSC/PATRAN and ability to process the parameter output file that MSC/PATRAN produces. Once a parametric session file is available, the analysis preference associated with it runs under control of MSC/PATRAN and the user does not have to intervene in the execution or post-processing as all that is controlled via the parameterized session file. The parameterization capability in MSC/PATRAN enables the use of preferences such as MSC/NASTRAN, ABAQUS and other solvers in a parametric mode without having to further address solver parameterization. The parametric session file can be used productively by someone not schooled in how to operate MSC/PATRAN or the associated solver. Communication can be entirely via design terminology that relates specifically to the design problem at hand. The session file can be incorporated within a design process, which includes a design scan manager such as RDCS or manipulated one case at a time by editing the parametric input file. That file can include an easily


understood, parametric description of the design. After execution of MSC/PATRAN, the parametric output file can then be accessed to discover the results of the changes to the input file. Acknowledgments This work was performed pursuant to a cooperative agreement with the Rocketdyne Division of Boeing Corporation and the Ford Motor Corporation with partial funding from the Defense Advanced Research Projects Agency managed by Wright-Patterson Air Force Base. It is part of the Robust Design Computational System (RDCS) program which is part of DARPA’s Rapid Design and Optimization (RaDEO) Program. The work was also supported by MSC internal funding.

References 1. Sippel, H., E. L. Stanton, and J. G. Crose, “The Usage of Numerical Optimization in the Development Process,” Presented at the 1998 International FEM Congress in Baden-Baden, Germany, The MacNeal-Schwendler Corporation. 2. Stanton, E. L. and T. E. Mack “A Case Study of Cure Cycle Shrinkage Deformations,” ASME J. of Engineering for Industry, Vol. 109, February 1987.


网站首页 | 网站地图
All rights reserved Powered by 大学生考试网 9299.net