This thesis proposes an integrated approach combining AI-based fuel modeling for fuel consumption reduction and onboard carbon capture (OCC) technology as practical strategies for carbon mitigation in LNG-fueled vessels under increasingly stringent gr...
This thesis proposes an integrated approach combining AI-based fuel modeling for fuel consumption reduction and onboard carbon capture (OCC) technology as practical strategies for carbon mitigation in LNG-fueled vessels under increasingly stringent greenhouse gas regulations in the maritime sector.
In Chapter 2, sensor data collected from an LNG carrier are utilized. The ship’s engine system is modeled using Aspen HYSYS, and variables that directly or indirectly influence engine performance are identified. Among 254 sensor variables, 60 engine-related variables and two key performance indicators (KPIs) representing engine efficiency are selected, resulting in a dataset consisting of 62 variables. Since real ship data inherently contain noise arising from data acquisition and operating conditions, data preprocessing is performed to mitigate its impact on prediction performance. A Savitzky–Golay filter is applied, leading to more than a twofold improvement in prediction accuracy.
For prediction, a Long Short-Term Memory (LSTM) model, which is well suited for time-series data, is employed to forecast one of the two engine efficiency KPIs. As a result, a high prediction performance with an R2 value exceeding 0.95 is achieved on the test dataset. Furthermore, key variables influencing the prediction model are identified using a permutation-based feature importance method. Variables closely related to the KPI, such as fuel inlet flow rate, are selected, confirming the physical plausibility of the model results. Finally, the robustness of the prediction model and feature selection is validated using liquid-fueled vessel data, where the prediction accuracy remains above an R2 value of 0.95. In addition, when a portion of the dataset is replaced with data generated from HYSYS simulations, the prediction performance remains above an R2 value of 0.93, demonstrating that process simulation data can be effectively utilized as a complementary data source when access to real ship data is limited.
In Chapter 3, the applicability of OCC is evaluated through CO₂ adsorption–desorption experiments and process simulations using a novel sorbent, amine-infused resin (AIR). The experiments employ a gas mixture representative of LNG vessel exhaust, consisting of 5% CO2 balanced with N2, and assess both the adsorption performance of the sorbent and its performance variation over repeated operating cycles. The effects of amine type, regeneration temperature, and operating pressure are investigated.
DEA-infused resin and MEA-infused resin are compared, and the DEA-infused resin exhibits a higher CO2 loading. Moreover, unlike the MEA-infused resin, no pronounced performance degradation is observed over repeated adsorption–desorption cycles. Stepwise regeneration experiments further confirm the feasibility of low-temperature regeneration. Among regeneration temperatures of 50 ℃, 60 ℃, and 90 ℃, the best adsorption performance is achieved at 50 ℃. In contrast, a reduction in adsorption pressure leads to a significant decrease in capture performance, indicating that pressure is a critical design parameter for OCC systems. Finally, an adsorption model is developed using Aspen Adsorption. The model is evaluated in terms of CO2 concentration at the reactor outlet and CO2 uptake within the column, and comparison with experimental results shows excellent agreement, with an R2 value exceeding 0.99.
In summary, this thesis demonstrates, through both experiments and modeling, that a combination of data-driven operational efficiency enhancement and AIR-based OCC technology with low-temperature regeneration capability can function as an effective strategy for reducing carbon emissions from LNG-fueled vessels.