CFD data based neural network functions for predicting hydrodynamic performance of a low-pitch marine cycloidal propeller

An experimental blade-controlled platform for the design of smart cross-flow propeller

“…Recent studies [9,15] has implemented more complex models to continue to estimate as well as possible cycloidal propeller performance. Lastly improvements on modern CFD solvers are accurate validation resources and a new interesting area of work for optimization purpose [2,10,11,17].…”

Research on the hull form optimization using the surrogate models

“…Miao and Wan (2019) also used the kriging model to predict the total resistance of S60 ship. Bakhtiari and Ghassemi (2020) used the feedforward neural networks to calculate the hydrodynamic coefficients of thrust and torque of the marine cycloidal propeller. Kim et al (2020) selected the artificial neural network to predict the ice resistance of the ship using different variables.…”

Predicting Dry Gas Seals Reliability With Machine Learning Techniques Developed From Scarce Data

“…Table 1.2 has the authors evaluated and the following information regarding their respective research: source of data used to compose the dataset; how the dataset was created; whether use classic machine learning techniques or deep learning techniques were used. Currently, there is a great focus on research to create representative predictive models of different types of physical phenomena, being created from experimental data [33,34,37,40,42,43,[45][46][47] or from numerical simulations [31,32,35,36,38,39,41,44,48]. From the previously mentioned, the studies [38,39] use CFD models to create datasets to be used to train the models, where the input variables are conditions that determine the operation of the evaluated equipment (e.g., inlet pressure and temperature).…”

“…Currently, there is a great focus on research to create representative predictive models of different types of physical phenomena, being created from experimental data [33,34,37,40,42,43,[45][46][47] or from numerical simulations [31,32,35,36,38,39,41,44,48]. From the previously mentioned, the studies [38,39] use CFD models to create datasets to be used to train the models, where the input variables are conditions that determine the operation of the evaluated equipment (e.g., inlet pressure and temperature). In [32,35], numerical data are generated to predict thermodynamic properties of gas mixtures, similar to the physical phenomenon present inside sealing systems.…”

“…When evaluating the models used in the studies, it is possible to verify the use of more conventional ML techniques, such as: Polynomial [31,36,37], Support Vector Machine [35,37,39,43,[45][46][47], Decision Tree [37], Elastic Net [43], Gaussian Process [43], Random Forest [34,43,46], Gradient Boosting [43,45], Extreme Gradient Boosting [45,46] and K-Nearest Neighbors [45,47]. However, in these studies, Deep Learning techniques were also evaluated, from simple ones such as Multilayer Perceptron [37,38,[44][45][46], to more complex Artificial Neural Networks (ANN) [32-34, 40-43, 45, 47, 48], obtaining errors of similar magnitude to the more conventional models previously mentioned. Furthermore, it is important to realize that there is a great focus of research on the Support Vector Machines model and Deep Learning techniques, and there are not many studies with simpler models and possibly sufficient to represent the desired physical phenomena in their respective studies.…”