Aufrufe
vor 2 Jahren

HANSA 01-2021

  • Text
  • Hansaplus
  • Bord
  • Schifffahrt
  • Container
  • Ships
  • Shipping
  • Marine
  • Hamburg
  • Hull
  • Hansa
  • Maritime
Hull Performance & Coating · Svitzer · Yacht »Soaring« · Schifffahrtsaktien & Börsen · Harren & Partner · LNG in der Schulte-Gruppe · Berenberg Bank · Schiffsinspektionen

SCHiFFStECHNiK | SHiP

SCHiFFStECHNiK | SHiP tECHNoloGY A hybrid hull performance prediction Finnish technology company Wärtsilä Voyage developed a new model for predicting hull performance that could improve accuracy while keeping costs low enough to make the system viable fleetwide The topic »fusion of high-frequency navigational data and noon-reported data to predict hull condition« was presented and discussed during recent Hull Performance & insight Conference 2020. accurately predicting the impact of hull condition on vessel performance allows ship operators to better understand and reduce their fuel consumption. But typical methods of predicting hull performance are far from ideal. our paper explained how a new model for predicting hull performance had been developed and verified that could improve accuracy while keeping costs low enough to make the system viable fleetwide. Hull performance predictions are typically based either on high-frequency, sensor-based data or low-frequency noon report data. Methods using sensor data are more accurate but are expensive because they must be tailored to each vessel and sensors require calibration. Noon report-based methods can be more easily implemented but are less accurate due to averaging of inevitable changes in speed, weather and other factors between reporting intervals. Wärtsilä Voyage’s third way combines high-frequency data available through standardised navigation equipment (such as ECdiS) with noon reports and data from third-party sources. By applying machine learning to these data sources, this approach achieves data quality similar to high-frequency data logging, but with low capital and operational costs comparable with noon report-based logging. Aggregation error Hull performance prediction can suffer from errors in measuring, modelling, implementation or statistical analysis. one significant statistical source of error is introduced when basing hull performance model calculations on noon report data. Noon report data provides average readings of various variables affecting vessel performance, including speed, fuel consumption and weather. But these factors are rarely consistent across a day. The more they fluctuate, the less accurate the averaged readings are and the greater the statistical error that is introduced into the readings and any subsequent calculations. as an example, the charts on the left in the figure shows timeline for speed and consumption for non-constant speed operation. Blue dots mark the high-frequency data. The average and standard deviation are plotted as orange lines. The speed-consumption graph is given on the right, plotting speed and aggregated consumption in orange. The aggregated consumption lies significantly above the high-frequency data points. This error is known as aggregation error. When variables have changed often throughout the aggregation period, this can lead to a deviation of as much as 30% from the true consumption. A timeline for speed and consumption. Blue dots mark the high-frequency data. The average and standard deviation are plotted as orange lines © Wärtsilä Voyage 48 HaNSa – international Maritime Journal 01 | 2021

SCHiFFStECHNiK | SHiP tECHNoloGY Data to hand This aggregation error can be reduced by more frequent reporting. auto-logging fuel consumption and speed is one option. But retrofitting such systems is expensive and for existing vessel the business case is usually weak. another option is reading and reporting consumption on each significant speed or weather change. This is unlikely to be practical for crew. The problem can however be overcome by exploiting sensors and systems already present on all vessels. While recording high-frequency fuel flow data is expensive, other data – including position, speed through water and wind - is usually available through the vessel’s ECdiS at high frequency and low cost. often rPM is available through NMEa. and speed through water can be substituted by speed over ground (from GPS positions) when combined with current hindcast data. in a previous paper, Wärtsilä Voyage presented a sensor fusion model that used speed over ground, forecast data, and fuel consumption from noon reporting to create a »virtual flow meter«. This model, known as the Fuel Flow Model (FFM), was verified and it proved to be capable of removing the aggregation error from pure noon data. The FFM is a »grey box« model. While a black box model would purely use data to »learn« the relationships of various inputs such as speed and weather to fuel consumption, a white box model is powered by standardized formulas that include coeffcients depending on vessel characteristics. one problem with the white box is that it cannot account for vessel specifics such as hull fouling over time. Conversely, black box models are unable to predict outcomes in operational conditions, which have not been covered in the training data if e.g., weather or operational profile change to completely new conditions, this makes black box model predictions unreliable. as a grey box model, FFM combines the robustness of a white box approach with the data driven vessel specifics tuning of a black box model. input data includes high-frequency readings of speed over ground, heading, depth, etc. (through ECdiS), different hindcast databases for weather parameters, and draft and fuel consumption measurements from noon reports. The FFM is upsampling the noon aggregated data (containing aggregation error) to high frequency data (without aggregation error) Testing the model Hull performance is a major aspect for ship owners and managers a numerical study was used to verify FFM. This involves generating realistic, high-frequency draft and speed readings and then computing consumption for a generic test vessel using a simplified formula based on speed, draft and other vessel characteristics. The model was then applied to consumption data aggregated to noon periods as well as speed and draft. The modelled high frequency consumption output was compared to the original generic input data. The accordance of the model output with the »true« data was shown to be very close after a short learning period, indicating that FFM can predict high frequency (flow meter like) fuel flow which is free from the aggregation error. in a second phase of testing, FFM was also able to remove large parts of error added to the input data, representing potential human error when inputting noon report data. This proved that the model is robust enough to remove significant error from the noon report data. in real life, the performance of the vessel changes over time due to hull fouling. This needs to be accounted for by the modelling, adjusting vessel coeffcients as performance changes. in the third phase of testing the model was exposed to generic data for a vessel with performance changing in time. Six scenarios were tested, including three with added (human noon reporting) errors. The model was able to predict the performance change well in all scenarios. Outlook The fuel readings performed by the crew on a noon-to-noon basis are usually imprecise due to human error. Furthermore, speed consumption analysis based on noon aggregated data suffers from systematical statistical error. The Fuel Flow Model has been proved to be able to remove both types of errors. Thus, it provides a practical approach to hull performance prediction that can be applied on any vessel. The Fuel Flow Model presented here is a simpler version of the one underlying Wärtsilä Voyage’s Fleet operations Solution. in a business setting this approach has proved to be robust and reliable. it is being used to predict hull fouling and to check the quality of fuel consumption reporting. authors: Daniel Schmode, Head of Solution advisory Matti Antola, data Scientist Wärtsilä Voyage © HaNSa archive HaNSa – international Maritime Journal 01 | 2021 49

HANSA Magazine

HANSA Magazine

Hansa News Headlines