Construction and application of numerical diagram for high-skew propeller based on machine learning

The field of machine learning has experienced rapid growth, and it has introduced a new methodology for constructing propeller diagrams. To meet the high demand for designing high-skew propellers, a series of high-skew propeller schemes are generated, utilizing the INSEAN E1619 as the parent propeller. The Computational Fluid Dynamics (CFD) method was validated using the E1619 test results and was subsequently employed to perform virtual open water tests for all the series schemes. This effort produced 819 open water performance data of 42 propellers. The study trained and validated the traditional multivariate polynomial regression model and five conventional machine learning regression models based on the CFD calculation data. The analysis of the model prediction accuracy indicated that the Support Vector Machine (SVM) model had the least error among them for the digital expression of diagram hydrodynamic data. The prediction error of K-T, K-Q, and ? decreased by over 20% compared to the LM model. The study subsequently developed a high-skew propeller diagram design program using the SVM regression model and applied it to a specific underwater vehicle's propeller design. The design results demonstrated that, compared to the B-series propeller, the design scheme provided by this numerical diagram had a comparable efficiency and a 6% smaller optimum diameter under unlimited diameter and a 7% higher efficiency under limited diameter for this case. Consequently, the developed numerical diagram in this paper provides a new tool for the propulsion performance evaluation and parameter selection of the propulsion system in the preliminary design stage for the high-skew propeller.