Trim optimization is an approach considered by the industry to improve the energy efficiency of ships, having a potential in both reducing operational costs and to decrease the emissions of the ship. Potential fuel consumption reduction by trim optimization is 0.5 to 3 % and to up to 7% in extreme cases. Stolt Tankers, a shipping company active in the chemical tanker market, wants to increase the energy efficiency of their ships in operation by trim optimization. For a limited number of ships, trim tables from model scale towing tests are available, but these are not used and the accuracy of these tables are unknown. The objective of this research was to develop a method to decrease fuel consumption by trim optimization, by a dynamic fuel consumption estimation model based on available operational data, that can be integrated in the voyage management system of Stolt Tankers. A dynamic fuel consumption estimation model has been developed, using mainly the noon report data of the C-38 ship class, consisting of six sister vessels of 38 000 DWT chemical tankers. Quality of noon report data is an issue: human error in observing and recording data causes noise and a mismatch exists between the snapshot of conditions on one hand, and the 24 hr averaged sailing speed and recorded shaft power or fuel consumption on the other hand. A data pre-processing framework has been developed and applied to integrate and transform the data, clean and filter the data. The model is able to extract the effects of speed through water, mean draft, trim, sea water temperature, wind force, sea state and swell state and their relative direction to the ship and days since last hull cleaning and propeller polishing. The model has shown to perform optimal using the regression model of Lutzen and Kristensen (2013), combined with a multiple layer feed-forward neural network, consisting of 1 hidden layer with 15 neurons. The model is able to estimate the shaft power with an average accuracy of 6.58 % for a random test set of the noon report data. It is able to extract the effect of trim on shaft power and to consider the effect of weather and fouling conditions. Model results show that about 1 to 2% of shaft power per 0.50 m can be saved, with trim by bow being the optimal trim. Based on sea trial results, it is concluded that the model performs most accurate for conditions that are represented by a high quantity of historical data. The model can be applied for speeds between 12 and 14 knots and for mean draft conditions of 9.5 m and more. Within this range, the effect of trim and weather conditions are followed with reasonable accuracy. It is confirmed by the sea trial that trim by bow is the optimal trim, with a much stronger magnitude of the effect of trim on shaft power. A difference of 6 to 8% in shaft power was found for a change in trim of 0.50 m.

Chemicals within a chemical tanker need heating to keep the cargo above a desired temperature. This research focuses on the prediction of the fuel consumption used to heat the cargo. There are three methods discussed in this research: a physics-based model (PBM), a data-driven model (DDM) and a hybrid model (HM). The PBM is a model based on the physical relations defined in the theory of heat transfer. The DDM is a model based on data gathered by a chemical shipping company (Stolt Tankers). Here an algorithm is trained and tested to acquire the model. The last method is a hybrid model. This HM is the combination of both the PBM and the DDM to use the best of both worlds. The PBM uses ordinary differential equations to model the fuel consumption and uses a numerical approximation. This method is first compared with a simulation in COMSOL Multiphysics and then scaled to an existing vessel. The PBM is able to predict the fuel consumption over a trip where the deviation in fuel consumption is 0.043 tons of fuel per day (5.6% deviation). The temperature deviation averaged over all tanks is 0.68 ∘𝐶. The DDM compares two algorithms: the linear algorithm based on the least squares method and an algorithm based on support vector regression (SVR). The model using SVR acquires the most accurate result, with a mean absolute error (MAE) of 0.0014 ± 1.7880 ∗ 10−4 and a mean absolute percentage error (MAPE) of 5.18616 ± 1.05567. When the PBM and the DDM are combined into one model the HM is formed. The HM is able to predict the fuel consumption of the vessel with a MAE of 0.0001441 ± 0.00010 and a MAPE of 0.8929 ± 1.0955. The PBM could also be used to calculate the reduction in fuel consumption when insulation is applied. Different types of insulation are simulated, with different thicknesses and thermal conductivity properties. Insulation of the cargo tanks could potentially reduce the cargo heating consumption by 64.8%. This reduction makes it worth looking into the insulation of vessels to reduce the fuel consumption and emissions. The DDM and HM are also tested to see if these models are able to predict values outside the range for the trained data. The DDM is not able to do so, whilst the HM is able to calculate scenarios where the average cargo tank temperature is outside the range for trained data.