Performance, Prediction and Optimization

Performance prediction and optimization is decidedly not a black art

Just going for a sail – getting a sailboat to move in the general direction that you want it to go – is not as hard as it is often made out to be. However, getting a boat to perform at absolute peak efficiency in a variety of wind and sea conditions is definitely hard, and predicting the potential performance of a new design (or the effects of modifications to an existing boat) over a range of conditions is even harder. This performance prediction and optimization process is decidedly NOT a black art, but doing it reliably and successfully does require a unique combination of hard science and practical experience.

This complexity should not be a surprise, as the physics of sailing involves the effects of (and the interactions between) all of the following, in constantly varying combinations: lift and drag forces in two different fluids that have very different densities and viscosities (air and water); surface waves that are generated by the boat’s motion on the free surface between those two fluids; different fluid flow velocities (boat speed and wind speed); different heel angles; different angles of attack (from upwind to downwind); and different sea conditions (smooth water to big waves).

Modeling, simulating, and quantifying all these forces and their interactions is an obvious challenge, and until the last century, performance prediction and optimization depended almost entirely on the skill, experience, and intuition of sailors and designers. Over time, various analytical tools have been developed to help guide the process, but to this day, none are entirely comprehensive or reliable.

One such tool is the ‘towing tank’, where scale models can be towed down a long, enclosed water channel, in which the forces on the models can be measured and then scaled up to full size. An early use of this tool in yacht racing was in the development of Sir Thomas Lipton’s Shamrock II for the 1901 America’s Cup. Her designer, George Watson, spent nine months tank testing eleven different candidate models for Shamrock II, but she was beaten by the American yacht Columbia, that was designed by Nathaniel Herreshoff without the benefit of any such testing. Afterward, Watson wryly remarked, “I really wish that Herreshoff had had a test tank available!” Use of a towing tank likely contributed to the success of the 1937 America’s Cup winner Ranger, but it was blamed (possibly unfairly) for the notable failures of the Cup candidates Valiant and Mariner in the early 1970′s. The use of much larger models since (scaled on the order of 1:3, rather than 1:13) has substantially improved accuracy, repeatability, and reliability, but this has also exponentially increased the cost of a comprehensive towing test program. I have learned a lot ‘in the tank’ about hull section shape and keel details (leading edge strakes, section thickness, fin/bulb volume distribution, etc). Still, even with the best of tank facilities and techniques, variations in experimental test results remain large enough so that a winner in the tank hardly guarantees a winner on the race course.

Another tool that has been developed much more recently is that of ‘computational fluid dynamics’ (CFD). These complex computer codes can predict three-dimensional fluid flow directions (streamlines) around various 3-d shapes (e.g. hulls, appendages, and sails) with impressive accuracy, and within limits, can also usefully quantify some the lift and drag forces resulting from these fluid flows. There are limits, however; the accuracy of CFD predictions can be heavily dependent on the size of the panels defining the test shapes, and the codes have difficulty with predicting the energy in surface waves and the drag from details such as complex underwater shapes and immersed transoms. CFD seems to work well in optimizing sail shapes, but for hull and fin details its predictions are best validated by either full size or large model testing before reliable conclusions can be drawn. Absent this validation step, some recent CFD-driven hull development programs have led to some unfortunate, and very expensive, failures. Like any good tool, CFD needs to be used appropriately and carefully.

A Velocity Prediction Program (VPP) is yet another performance optimization tool, and one that can be far less costly than tank testing or validated CFD. This family of computer codes is now widely available from a variety of sources, but all trace their DNA to pioneering research done at the Massachusetts Institute of Technology in the 1970′s. Conceptually, VPP’s model performance based on numerical representations of hull, appendage, and rig parameters such as effective sailing length, beam, draft, displacement, stability, wetted area, sail details, etc , all of which are supplied as program inputs. The code then calculates lift and drag forces generated by the hull and rig as defined, over a range of wind speeds and true wind angles, (points of sail), and it finds combinations of these forces that put the total drive, side force, and resistance forces in equilibrium. As output, a VPP then produces the equilibrium conditions that give the best performance at a predetermined set of true wind speeds and angles. The accuracy and reliability of any VPP depends not only on the quality of its underlying code, but also on the quality of the input data. Reliable outputs require representative inputs, and defining parameters such as effective sailing length over a range of speed and heel angles is a challenge. A few VPP’s predict absolute performance accurately, and most can reliably predict relative performance deltas between a control ‘base boat’ and a generically similar test boat whose parameters have been systematically modified. However, no VPP can be expected to resolve subtle differences in details such as keel leading edge shape, transom drag, or rudder type.

VPP output comparing the predicted performance of four different 'test' boats relative to that of a 'base' boat, on a 50%/50% Windward/Leeward course

VPP output comparing the predicted performance of four different ‘test’ boats relative to that of a ‘base’ boat, on a 50%/50% Windward/Leeward course

Since all of these analytical tools have their limitations, practical experience still plays a vital role in performance prediction and optimization. This experience can be gained most reliably via focused two boat testing at full scale, but this is surprisingly difficult to orchestrate, very time consuming, and a crushing bore for everyone involved except for the handful of gearheads initiating the tests. I like to think that one of my most significant contributions to Bill Koch’s 1992 America’s Cup win was in improving our two boat testing. When I started reviewing our test data, at least 2/3 of the tests run were of no use, due to bad line-ups, wind shifts, bad sail trim, etc. Towards the end, our testing was far better, and it was critical to our ultimate win. Over and over, two boat testing has proven to be decisive at the Volvo Race and America’s Cup level, but doing it well requires very careful coordination, two top crews and a big budget, so it is not practical for most design development projects.

The other main source of the practical experience that is so vital to reliable performance prediction and optimization is the race course. Distance racing is seldom very helpful, since too many variables come into play when boats scatter over a big expanse of ocean. However, a great deal can be learned from short course buoy racing, where boats spend a lot of time matched up in close proximity, much as in a good two boat test. I have spent a huge amount of time short course racing in my career, partly because I love the competition, but also because every sequence of every leg of every race can be analyzed as a useful science experiment. I could no doubt boast of some better race results if I spent more time looking up the course, and less time focused on the ‘experiments’, but I have learned an enormous amount in the process.

Following is an abbreviated list of issues (there are lots more!) that are not good candidates for cost effective towing tank, CFD, or VPP studies, but that have been resolved via practical experience in the short course racing ‘lab’:

Transom immersion and drag

Transom extensions added to the sterns of two lightweight boats (one my design) with short aft overhangs lit up the performance of both boats in moderate conditions. (An independent tank test has since confirmed that drag from an immersed transom can represent a significant portion of the total drag of even a moderate displacement boat). This experience led me to a focus on establishing a systematic approach to setting optimums for transom height, aft profile slope, and stern overhang length, all of which vary with different hull shapes and parameters.

Skeg hung vs free standing spade rudders

The before and after performance ‘bounce’ from changes from the former to the latter has been startling in every case: better speed, more precise handling, and no loss in ultimate control. Several examples each of the Tartan 41, Swan 44, and Seguin 44/46 were modified in this way, and all morphed from also-rans into race winners essentially overnight.

Stability vs sail area, rating, and ‘real world’ performance

It is well known that generating adequate stability relative to sail area is critical to sailing yacht performance, and any good rating rule measures stability and assesses its effect. (Notably, IRC does not measure it, and uses surrogates less successfully). In the later years of racing under IMS, a lot of racing teams went to unusually low stability, seeking the resulting lower, slower, ratings. In the end, the ratings did get slower, but the tippier boats became much harder to sail well. Elite crews sailing unusually ‘boxy’ hull shapes in smooth water may have gained a performance vs rating advantage and done well, but more typical crews, sailing more normal hull shapes in more typical conditions often found that with lower stability their actual performance slowed down more than their rated performance. The towing tank, CFD, and VPP’s all model performance in idealized, steady state conditions. The race course ‘lab’ seldom provides consistent, controlled conditions, but that is precisely why it better represents real world sailing so often.

A proprietary measure of stability relative to sail area, plotted against effective sailing length. The trend line shown is just a visual referenced, and not a mathematical fit to the data.

A proprietary measure of stability relative to sail area, plotted against effective sailing length. The trend line shown is just a visual referenced, and not a mathematical fit to the data.

Appendage profile areas s

In the same way, the keel and rudder profile areas that might be optimal in steady state, smooth water conditions are typically quite different from those that suit the real world. Tacking, gybing, pinching off starting lines, rough seas, ducking a starboard tacker in breeze, even downspeed maneuvers in tight marina quarters, all require more fin area than the steady state ideals that even the best analytical tools might suggest. My ‘research’ on the race course led to ‘fin sizing’ algorithms that establish reference areas for both keel and rudder, based on hull and rig parameters, and varying with boat size. Knowledge about who will be sailing the boat, how they plan to sail (race, day sail, or cruise), what the draft restrictions might be (deep or shoal), and what wind and sea conditions they might expect to see most often, will typically shift the target fin sizes a bit either side of this calculated baseline, but seldom very far; this routine has proven to be a remarkably reliable guide.

Reliable performance prediction and optimization requires not only a good understanding of sailing’s underlying physics, but also a keen awareness of the limitations of that science, and broad experience with the realities imposed by the real world.