Built to win

Lausanne, Switzerland—America's Cup yacht races are fiercely competitive. Even after a race of several hours, just seconds can separate two teams at the finish line. Unlike aerospace applications, for which engineers must maintain vast safety margins, America's Cup competition demands the utmost performance, and a design team must perform extensive tests and simulations to ensure its racing yacht can pursue victory without slipping over the edge into disaster.

The America's Cup series held this past spring in Auckland, New Zealand, and won by Team Alinghi of Geneva, tells both sides of this story: victory and disaster. To help wring every drop of performance out of the latest technology, the Alinghi syndicate named the Federal Institute of Technology in Lausanne (l'Ecole Polytechnique Federale de Lausanne, or EPFL) as its official scientific advisor.

Jan-Anders Manson, Director of EPFL's Laboratory for Composites and Polymers and the EPFL-Alinghi Partnership Coordinator.

Such wasn't the fate of Team New Zealand, the host team and sitting champion, which suffered from a boom breaking in the first race, a mast snapping in the fourth race, and a spinnaker pole breaking in the final race. The defenders stunned their countrymen by failing to win a single race. Some observers speculated that inadequate testing may have led to these equipment failures.

I recently visited the Alinghi project team at EPFL to learn about the tests they conducted and the challenges they faced in helping the boat achieve its excellent performance. Although the competitive nature of the America's Cup prevents the researchers from revealing every detail, I was able to garner enough information to get a sense of the process they followed.

Alfio Quarteroni, chair of modeling and scientific computing for EPFL's Institute of Analysis and Scientific Computing.

Yacht designers obviously want to minimize resistance to water and air friction, as well as the production of waves, while maximizing propulsive force; just a 1% difference in hull resistance can mean a 30-s difference at the finish line. The amount of frictional resistance, for instance, depends largely on the hull area over which laminar flow (low friction) can be sustained before the flow becomes turbulent (high friction). The hull must be designed in order to achieve the best tradeoff between wave and viscous resistance. In addition, the shape of the keel (including the underwater bulb, its fins, and the "sail" that drops down to support the bulb) and even the shape of the rudder blade play important roles. For the Alinghi, Professor Alfio Quarteroni, chair of modeling and scientific computing for EPFL's Institute of Analysis and Scientific Computing, evaluated the hydrodynamic and aerodynamic flow of more than 100 different hull and keel configurations.

Quarteroni noted that equations for analyzing fluid flow around a geometry are well known, but applying them to an America's Cup design is complicated by the complex physical modeling of the boat, wave generation on the water surface, and fluid-structure interaction with the mast and sails, including the effects from a competitor sailing nearby.

Modeling the boat was perhaps Quarteroni's biggest challenge in the Alinghi project. His group broke the vessel down into roughly 200,000 surfaces of varying size in a nonuniform grid, and that process alone took several weeks the first time. This "breaking down" had to be done properly so the group could accurately simulate the complex flow field around the boat.

Simulation helped EPFL researchers study fluid flow around the Alinghi's underwater bulb and the "sail" that drops down to support it.

Clearly, Quarteroni needed some powerful tools for this job. His group did the bulk of their work with a commercial package called Fluent, which Quarteroni said is one of the most popular computational fluid dynamics (CFD) packages of this class. He needed additional software to approximate activity at the boat's water line—known as "free-surface simulations"—so he turned to code developed at Princeton University.

For problems that involved greater numbers of mesh cells, the group turned to two parallel central servers. The first is a Silicon Graphics Origin 3800, a supercomputer containing 128 R14000 processors from MIPS (Mt. View, CA), each running at 500 MHz, and with a total of 64 Gbytes of RAM. The second is the Swiss T1, a supercomputer developed at EPFL, which contains 64 Alpha ev6 processors each at 500 MHz and working with a total of 32 Gbytes of RAM.

The team ran all simulations using the Princeton code on the Origin 3800 with between 6 and 16 processors. To get a sense of the complexity of the computations, consider that the largest Reynolds-Averaged Navier-Stokes simulations using either Fluent or the Princeton code (involving 5 million mesh cells) required as much as 10 hours on a 16-processor configuration with either the Origin or Swiss T1 machine.

The raw computations aren't of much value, and getting visual displays that help researchers understand what's actually going on are crucial. The postprocessing tools available in Fluent created most of the computer visualizations, but the team postprocessed the free-surface simulations from the Princeton code using pV3 (Parallel Visual 3), a CFD visualization tool for parallel machines and workstation clusters developed at MIT (Cambridge, MA).

To accurately simulate the fluid flow around the immersed part of the boat, Quarteroni's team needed to know the location of the line at which the transition from laminar to turbulent flow occurs. This information was required to calibrate the numerical transition model on which the accurate prediction of frictional drag hinges. To experimentally determine this laminar-turbulent transition line on a full-scale training boat under different sailing conditions, a team led by Professor Peter Monkewitz at EPFL's Laboratory of Fluid Mechanics mounted a series of sensors on the underwater sail.

These sensors were Senflex foil hot-film skin friction sensors (Tao of Systems Integration, Williamsburg, VA), mounted in a salt-water resistant fashion such that the electrical connections could withstand the forces of the fast-flowing water. Each sensor was connected in a standard bridge configuration and slightly heated. The current required to keep the sensor temperature—and its resistance—constant provides a measure of the heat transfer, which is related to the wall shear stress or skin friction. The absence or presence of skin friction fluctuations indicate whether the flow is laminar or turbulent, respectively.

The main challenges were to arrange correctly the array of sensors on the sail to avoid interference between sensors, to separate boat motion from turbulence during data analysis, and to characterize the transition location in a statistical sense (as the transition location constantly moves around on a moving boat). In practice, it was also necessary to distinguish between the output of a malfunctioning sensor (for instance, one with corroded connections) and turbulence. For all this, Monkewitz used digital filtering techniques and statistical data analysis, taking advantage of the fact that the frequencies of turbulent fluctuations are generally high with respect to boat motion.

Another important source of drag is flow separation from the hull and the sails. This phenomenon is due to fluid particles that no longer follow the surface of the hull or sail and thereby create regions of low-pressure backwash that "hold" the boat back. One area particularly prone to separation is the trim tab that is located on the trailing edge of the "sail" between the keel and the bulb. This trim tab is angled to increase the lift of the keel sail, which counteracts the wind's tendency to push the boat sideways. When the trim tab is angled too much, however, the flow separates, the lift gain is lost, and drag increases. To avoid these problems and still push the trim tab angle to its limit, Monkewitz developed a simple real-time "separation sensor." This sensor connects to a warning light in the cockpit area, which allows the helmsman during a race to correct instantly a flow separation condition after accidentally turning the trim tab too far. The trick, Monkewitz added, was finding a sensor with the right geometry to give it maximum sensitivity.

After the researchers determined the optimum geometry for various vessel components, the boat builders had to create them. The material of choice for many components is "prepreg," which consists of a fabric of carbon fibers impregnated with epoxy resins. It comes in a pliable form, and it becomes hardened when you cure it under temperature. Applying pressure by curing it in an autoclave reduces the porosity in the material to levels below 1%, enhancing its mechanical qualities.

America's Cup rules, however, allow the use of such a pressure container only for some parts but not for the hull. The reason is cost; an autoclave big enough to hold hull pieces becomes extremely expensive, and this rule allows teams with smaller budgets to compete. Thus, the curing process for the hull uses a vacuum bag at 1 bar (roughly 1 atm, or 14.5 lbs/in.²).

This restriction doesn't typically exist in industry. With airplanes, for instance, you want the best strength, almost regardless of the expense. Thus, engineers in industry have considerable experience working with prepreg in an autoclave. Getting the best strength out of prepreg in a vacuum-bag environment, though, is a relatively small but growing research area, noted François Bonjour, an engineer at EPFL's Institute of Materials, Lab of Polymers and Composites. One key task for his team was to examine different high-performance prepregs and select the one that would produce the best results under a vacuum-bag environment. He also had to determine the optimum curing cycle (how much heat for how long) and design a manufacturing process that the boat builders could implement.

Because a porous prepreg has far less strength than a prepreg with a lower porosity, Bonjour had to find the process that led to minimum porosity. In the hull, two prepreg layers sandwich a honeycomb.

To measure porosity in the prepreg, Bonjour examined cross-sections of it with a BH2-UMA research microscope from Olympus (Melville, NY), which is connected to a Macintosh with a package called Image Grabber (Scion, Frederic, MD), and he placed the images on the EPFL intranet. He then called them up into a Windows machine running ImageTool, an academic image-analysis package developed at the University of Texas Health Science Center at San Antonio Dental School and distributed at no charge. Using that package, he calculated the ratio of the area of the porosities to the entire area. He also looked for porosities and glue meniscuses around the core wall.

Francois Bonjour, an engineer at EPFL's Institute of Materials, Lab of Polymers and Composites.

In evaluating materials and coming up with a curing cycle, Bonjour had to work closely with the boat building team at Décision S.A. (Fenil sur Corsier, Switzerland). In fact, many departments at EPFL worked closely together, feeding their results to each other in a continual simulation/materials/boat-building/on-water test cycle.

It's impressive that the researchers delivered their results in such a short time: Team Alinghi was initially organized in September of 2000, resulting in an amazingly short time for technology transfer. Teamwork on and off the water was essential to victory.