Issue: Volume: 25 Issue: 3 (March 2002)

Virtual Testing

By Pamela Waterman

Would you drive a car whose components have been designed without the aid of physical testing? Actually, you are probably already doing so. The Big Three automotive manufacturers have put years of effort into the testing of virtual assemblies, and for good reason. Global competition has forced them to slash vehicle development time from five years to two, and they're achieving such reductions largely by cutting the number of physical prototypes they have to build for testing.

General Motors calls its program "moving from Road to Lab to Math," emphasizing virtual testing for everything from tracing the path a glass window follows as it opens and closes to determining the behavior of a vehicle in a simulated crash. Daimler-Chrysler has been employing virtual testing since 1999 in six different vehicle programs; fields of interest include suspension fatigue issues and fluid flow performance. And Ford, taking a somewhat different focus, is pursuing a Virtual Manufacturing Strategy that increasingly relies on simulation for detailed planning in such areas as plant layout, ergonomics, assembly sequencing, mold-flow, formability, and welding.

The reason virtual testing works is because software and hardware developers have greatly improved the correlation of physical and computer-generated systems so that simulated results are more believable. This concept, used for years in the nuclear and aerospace industries, has expanded to include component-level, assembly, and full-system testing in a number of other industries, including aviation, building design, heavy-equipment manufacturing, even entertainment animatronics, with automakers leading the way.

Companies intent on speeding up their design process are investing in more virtual test software than ever before, a market trend tracked by analysts at Daratech (Cambridge, MA), the IT marketing research and technology assessment company. The firm's figures show a 10 percent growth in user spending on virtual prototyping products in 2001 (over 2000) in spite of the recession. The research firm expects a modest 8 percent growth for 2002, with an average growth of 15 percent per year through 2007. Even at these rates, this growth outpaces that of other CAD/CAM and CAE applications, and virtual testing is a significant factor in this area.

"Virtual prototyping has its roots in the CAE and Nastran codes of the 1960s, with aerospace companies doing performance prediction and simulation in cases where you couldn't feasibly work on the real thing," notes Bruce Jenkins, Daratech executive vice president. "For example, analyzing the potential stress and strain on a rocket by actually building and launching one just wasn't cost effective." Other early application areas included modeling accelerated aging tests of nuclear reactor vessels and determining high stress regions on aircraft turbine blades. In both cases, the simulations permitted making subtle design changes before casting even a preliminary part.

"What is changing is how this technology is used," Jenkins says. "For a long time, users viewed virtual prototyping and simulation primarily as a defensive tool. If a product began failing in the field, a simulation could let them see what was happening and pinpoint where the design needed to be beefed up. Now, this is increasingly being seen as an offensive tool, driven by the competitive need to compress time-to-market and time-to-revenue from new products, as well as to improve product quality."
GM uses simulation to optimize vehicle performance in the event of a crash. The company's math-based systems successfully run computer models through virtual tests to complement and, in many cases, reduce the role of physical prototypes.

Physical prototype fabrication and testing, inherently a one-of-a-kind handcrafted process, puts roadblocks in the efforts to shrink product development cycles, especially in the automotive world. In particular, performance testing, although improved for years through the replacement of many road tests with lab tests, is one area where virtual test scenarios are playing an increasingly important role. Today, computer models of vehicles, subsystems, and components are tested virtually against one or more input variables, allowing engineers to analyze the results and decide where to improve the original design. Much of this happens before one tool is cast or a single piece of sheet metal is cut. Designers still build prototypes, but the number is greatly reduced.

The key to making all this work is to ensure the accuracy of the simulated models, and that includes using reliable information about how external factors (the loading conditions) affect their behavior. For example, to characterize new shock absorber designs, engineers must model tire and road interactions, the expected range of vibration frequencies, and even seemingly minor modifications that affect mass and inertia calculations.

Sudhakar Medepalli, manager of chassis analysis for Daimler-Chrysler, comments there are three ways to obtain such load data for the fatigue/durability design of any component. In the first approach, one could rely entirely on traditional road-test (physically measured) data, but this is expensive, time-consuming, and non-predictive. A second option would be to create a completely virtual model; however, this method is computationally intensive and actually requires more testing to verify the results than a physical approach.

The third choice is a hybrid approach currently used at Daimler-Chrysler. Here, one measures a minimum set of physical data, then uses these values as inputs to a computer model, in this case built with Adams software from Mechanical Dynamics Inc. (MDI), which calculates the remaining load points. All subsets of the mathematical model are then fed into finite element analysis calculations. The beauty of the hybrid approach is that it combines the best of both the physical and virtual load-generation techniques with fast turnaround times, simultaneous calculations of various components, and the ability to correlate values.
In fatigue tests of its shock absorbers, Daimler-Chrysler finds that virtual test results (green curves) correlate directly to physical lab test data (red curve, top graph).

Until recently, a potential downside was the necessity of having good models for such components as shock absorbers, engine mounts, bumpers, and elastomers-items that have traditionally defied accurate characterization. Daimler-Chrysler now obtains much of this information from MTS Systems, a major mechanical testing and simulation equipment supplier that has successfully moved into the realm of virtual testing.

Daimler-Chrysler has applied this hybrid method to more than 25 load-data acquisitions, including shock-absorber fatigue testing. The company used wheel-force transducers for the physical portion and Adams for the virtual parts. For each data-acquisition channel eliminated from physical measurement, this approach saves the company approximately $1500 to $2000. It also reduces the instrumentation and data acquisition time, and results in a more efficient utilization of the acquisition vehicle. The company is also using heuristics (experience-based results) to extend the method as a predictive tool.

General Motors has made a corporate-wide commitment to virtual testing as part of its "Math-Based Synthesis-Driven Vehicle Development Process," a mouthful of a phrase that describes the company's fundamental quest for a virtual proving grounds. Steve Rohde, technical director in the Engineering Process & Math Strategy Group, notes that GM has been in the business of mathematical modeling for a long time, even in the '40s and '50s. Given the title of his department, it's no surprise that GM actively uses math-based tools to make decisions on form, function, sales, and marketing, to reduce development time and cost, improve product quality, and even enhance innovation. In the past year, with virtual testing and increased compute power, GM has achieved a 300 percent increase in the total number of crash simulations completed, and reduced the wait times for results by factors of 6 to 60.

However, as GM moves toward an approach in which math-based analysis is supported by "hardware" (physical prototype) testing, rather than the other way around, the main question in Rohde's mind is, how good is the math relative to the hardware? "We'll always be building some physical prototypes, but instead of building to see what's wrong, we build to confirm the math models. We're moving from 'road' to 'lab' to 'math,' determining which vehicle specifications must be evaluated in hardware, which ones still need some hardware combined with math modeling, and which ones can do without development hardware but still need math validations."
In this virtual test of a V6 crankshaft, Virtual Lab from LMS shows (clockwise from top left) a 3D model of the crankshaft, vibration of the car body subframe, the engine vibration pattern, and the supporting data.

As an example of how increased virtual evaluations have already reduced road and lab tests, Rohde points out that GM performed 80 front and rear impact tests on the 1991 Chevrolet Caprice to meet requirements, but only 23 on the 1999 Chevrolet Impala for the same set of requirements. Although the number of tests performed varies widely with the complexity of the vehicle, these two vehicles were very similar in design complexity, so the reduced number of tests directly reflects the successful impact of the mathematical approach.

Ford is focusing on a "Virtual Manufacturing Strategy." The plan emphasizes the need for a seamless process that takes virtual products from the design phase all the way through launch. Issues that traditionally need to be addressed in manufacturing, in order of increasing complexity, include geometric compatibility, kinematics/sequencing ergonomics, equipment control, dynamic behavior, and product and process integrity. Possible virtual manufacturing tools to address these areas fall into four categories: process design, process verification, process validation, and operations.

Within these four stages, models of virtual tools could be developed for geometry creation, assembly sequencing, discrete event simulation, ergonomics, die design, fixture design, assembly-line design, off-line programming, and overall manufacturing station design. More difficult to model are attributes that control such manufacturing techniques as sheet-metal formability and mold-flow design.

Ford has found that, currently, few software tools support process integrity, and there is limited integration among them. However, it seems that virtual manufacturing data could support facility and tooling reuse as well as vehicle component and part reuse.

Looking to the future, GM's Rohde notes that "if you have a perfect math model, you can start looking at variations due to assembly, sheet-metal thickness, and even individual drivers, really pushing the envelope of virtual testing to learn about the impact of manufacturing on many possible performance variations. Eventually, you will be able to simulate these performance factors, not just for one vehicle, but for the entire fleet."

Pamela Waterman is a freelance writer specializing in CAD/CAM/CAE.

Several companies are putting extensive effort into providing the tools for users to conduct virtual testing and at the same time correlate results with complementary physical testing. Among these vendors are LMS International, Mechanical Dynamics, MTS Systems, nCODE, and PTC.

LMS International's Virtual.Lab is the company's integrated test environment for functional performance engineering. It speeds up the design process by offering the ability to use validated models of existing components to make predictions about new ones. Fatigue-life predictions that used to take a couple of weeks can now be completed in a day. Virtual.Lab incorporates products and configurations that are initially targeted to the automotive domain in such areas as engine acoustic radiation and vehicle noise, vibration, and handling analysis. It can also be used for generalized noise and vibration analysis.

Mechanical Dynamics Inc. (MDI) offers its Virtual Prototyping Maturity Model for companies looking to perform more functional testing and less physical prototyping. The service provides a roadmap for assessing the level of return a company currently receives from virtual testing, the ROI one should expect to receive, and ways to help achieve that vision.

MDI also has teamed up with testing specialists MTS Systems and with fatigue and durability experts nCode International to form the SmartSim Community in an effort to develop joint solutions with open standards. This group is committed to providing the vehicle development world with integrated simulation tools, and has already produced MTS's Empirical Dynamics Model (EDM) methods and Virtual Test Lab (VTL) tools.

EDM software combines input (such as displacements and forces) from tests run on a physical component with physical test equipment, and uses that information to calculate the load behavior in a neural network model. That data in turn becomes input for an analyst to use in FE analyses or in dynamic (motion) simulations. Rather than obtaining one plot per test in the lab, it generates a file that covers a broad behavioral spectrum.

MTS's VTL system is a set of files that models the actual test equipment (rather than the loads). Output data from a VTL system plays a large role in the predictive evaluation of fatigue, vehicle dynamics, handling, ride/comfort, and noise/vibration, by serving as input for MDI's Adams mechanical testing software. With the VTL-modeled test equipment, designers apply loads to simulated parts and assemblies, and set the latter in motion within Adams to "see" predicted behavior. -P.M.

LMS International ·
Mechanical Dynamics Inc. ·
MTS Systems ·
nCode International ·
SmartSim ·