Original scientific paper
https://doi.org/10.21278/brod77405
Optimization of marine propeller characteristics for maximum open water efficiency using an ANN-GA tool trained on experimental data
Carlo Giorgio Grlj
; University of Zagreb, Faculty of Mechanical Engineering and Naval Architecture, Ivana Lučića 5, Zagreb 10000, Croatia
*
Nastia Degiuli
; University of Zagreb, Faculty of Mechanical Engineering and Naval Architecture, Ivana Lučića 5, Zagreb 10000, Croatia
Ivana Martić
; University of Zagreb, Faculty of Mechanical Engineering and Naval Architecture, Ivana Lučića 5, Zagreb 10000, Croatia
Marta Pedišić Buča
; Jadranbrod d.d., Avenija Većeslava Holjevca 20, Zagreb, 10000, Croatia
* Corresponding author.
Abstract
This study proposes a numerical approach for identifying the propeller characteristics that achieve maximum open water efficiency for a specific ship, considering its propulsion characteristics and defined operating conditions. The proposed method combines an artificial neural network (ANN) with an optimization procedure based on the genetic algorithm. The ANN is trained using experimentally obtained open water characteristics of 143 propellers, enabling accurate prediction of thrust and torque coefficients as well as the open water efficiency as functions of propeller geometric parameters. The optimal ANN achieved an R2 of 0.95 and RMSE of 0.20 on the validation set. Once trained, the ANN is integrated into the optimization procedure to explore the design space and identify the optimal propeller, while satisfying the imposed constraints. The approach is validated on several benchmark ships. The obtained results show good agreement with those from literature, despite the relatively small training dataset used in the present work. The obtained open water efficiencies are higher than those of the original propellers for all ships considered. It is demonstrated that the required propulsion characteristics used as input parameters can be obtained from different sources, including numerical simulations, experimental data, and empirical prediction methods such as the approach proposed by Holtrop and Mennen. For practical implementation, a standalone application was developed in MATLAB, integrating the trained ANN and genetic algorithm (GA) optimization procedure into a user-friendly environment.
Keywords
Preliminary propeller design; open water characteristics; artificial neural network; surrogate model; Bayesian regularization; optimization; genetic algorithm
Hrčak ID:
351207
URI
Publication date:
1.10.2026.
Visits: 0 *