SCHIFFSTECHNIK | SHIP
SCHIFFSTECHNIK | SHIP TECHNOLOGY Route optimization: Myth or reality? A critical view at fuel savings from routing: The key components of routing, namely meteomarine forecasts and hydrodynamic response model, have much larger uncertainty than the alleged savings. Volker Bertram argues in favour of more validation for hydrodynamic models Ship routing denotes generally the optimization of a ship’s course and speed with respect to some objective. First developments date back at least to the 1970s, and have proliferated since then, with many variations on the key elements of any routing application: optimization algorithms, objectives (such as minimum fuel consumption), meteo-marine models (weather forecast), and ship performance models. Which estimates to believe? None! We will consider in the following only routing for energy efficiency objectives where ambient waves have a significant impact. Claims to how much fuel routing saves differ widely, from almost zero to 10 %. This should not surprise. Serious estimates should always come as a bandwidth, as the honest answer on saving potentials of any measure will be the timehonoured »it depends«. Numbers given by vendors or selected (positive) cases are »All models are wrong, but some are useful« invariably higher than numbers by thirdparty consultants. But which of the estimates should we then believe? I believe the correct answer to this question is – none. The uncertainty in the results of any optimization depends largely on the accuracy of the used functions for the objective. Next, we will discuss the key elements of routing with focus on the uncertainties or errors associated with each element. The meteo-marine forecasting predicts the ambient conditions (wind, wind sea, swell, currents). These ambient conditions drive the ship’s response, in our case the added fuel consumption. Obviously, errors in the ambient conditions (input signal) will propagate to errors in response (output signal). Error sources are: (1) Wind is highly nonlinear and thus difficult to predict over longer periods; beyond 3 days, the predictions become noticeably fuzzy. (2) For wind sea, the problem is rather that we approximate actual distributions of energy over wave frequency and direction by © Dataloy 52 HANSA – International Maritime Journal 10 | 2024
SCHIFFSTECHNIK | SHIP TECHNOLOGY standard spectra with just two parameters: significant wave height and main direction. The standard spectra were derived fitting many measured spectra (histograms) with a smooth curve. Taking a standard spectrum (e.g. the Pierson-Moscowitz spectrum) as an approximation for actual instances of wave distributions (histograms) comes always with an error. This is akin to some regulations using a standard weight of 75 kg per passenger, where at least in my case this is flatteringly far away from the truth. Some meteomarine forecasts neglect swell which can lead to significant errors. For currents, the spatial resolution of meteomarine measurements is often not fine enough for accurate predictions. The ship performance model predicts the »performance« (e.g. power requirement or fuel consumption) as function of various input variables describing ambient and operational parameters. Such performance models are needed for a variety of applications: trim optimization, performance monitoring, routing, etc. While it makes sense to reuse knowledge for setting up such models the requirements for models differ according to their intended purpose. Routing generally requires the inclusion of higher waves in the performance model. The associated challenges are higher than for trim optimization and hull management. The ship performance model may be built using various approaches, first-principles simulations (e.g. CFD), machine learning, or a mix of both. »All models are wrong, but some are useful«. Measured input data introduce errors and uncertainties, but so do models. There is no approach that can determine added power requirements in medium-to-large seaways, especially if there is a significant contribution from waves that are relatively short compared to the ship length (say less than half a ship length). Our collective knowledge of uncertainties and errors for added resistance in waves (as a simplified proxy for added power requirements in waves) is at best sketchy. Model tests with controlled conditions for waves and the simplest case of ship in head waves show already significant scatter if performed repeatedly in the same model basin. The scatter increases when comparing different model basins. For computational methods, validation workshops give some indication on significant scatter even if the same type of code or even same codes with different grids and parameters are used. Simple semi-empirical approaches may feature errors of 400 %-1,000 %, making them hopeless. In conclusion for the ship performance model, assuming errors of 30 % in the computation of added power requirements in natural seaways is rather optimistic. The published fuel savings from routing are not real: There is no possibility to have two identical ships sailing at the same time on the shortest route and the determined optimum route to measure the difference in fuel consumption. Instead, any such published fuel saving refers to the difference according to the ship performance model. And while the shortest route is clearly determined, the computed optimum route depends also on the other elements of routing optimization, foremost the meteo-marine weather forecast. The various elements in routing, especially the uncertainties for the natural seaway and the ship performance model prediction for added power requirements in natural seaways, come with large errors, at least in the order of 30 %. Routing savings are given in the range of 0–10 %, but let’s say often by more serious sources at 3 %, i.e. an order of magnitude smaller than the error in the function to be optimized. If errors cancel out partially, or if errors are systematically added in same magnitude, the optimization process may still find an almost optimum solution (in our case the best route), but the associated savings will be wrong, possibly lower, possibly higher than what the ship performance model gives as result. In any case, the results are highly doubtful, or »myth«. Continue using routing It would be interesting to see the effect of variations on the results (sensitivity analyses): What different routing solution products predict for a given route as best route and as saving, and what a given routing solution product gives for variation of input within margins of uncertainty for the input. In general, more validation and quantification of error margins, especially for the ship performance model, would be helpful to assess saving potential of routing software seriously, and to get guidance on where current software and models can be improved. For example, one could take a given route, record speed profile, hindcast meteo-marine data, and predict consumed fuel in blind tests to get an idea how well or not a given software or ship performance model performs. In conclusion, please continue using routing, even if we cannot quantify savings. Routing will save fuel, just how much we don’t know, and likely less than vendors claim. This is an adapted and abridged version of a presentation published at this year’s HIPER (High-Performance Marine Vehicles – »Technologies for the Ship of the Future«) conference, for which HANSA is the exclusive media partner. More information here: http://www.hiper-conf.info © ABB HANSA – International Maritime Journal 10 | 2024 53
