Propulsive Property Prediction Utilizing Neural Networks with Varying Geometry Inputs
2nd International Conference on Aeronautical Sciences, Engineering and Technology, ICASET 2025, Muscat, Oman, 4 - 06 November 2025, pp.279-287, (Full Text)
- Publication Type: Conference Paper / Full Text
- Doi Number: 10.1007/978-981-95-5907-7_25
- City: Muscat
- Country: Oman
- Page Numbers: pp.279-287
- Keywords: Machine learning, Neural networks, Propeller design and optimization, Scaled conjugate gradient
- Middle East Technical University Affiliated: Yes
Abstract
In this study, artificial neural network (ANN) applications were used to predict propulsive properties of small propellers utilizing experimental data and detailed geometry information of the propellers. Scaled conjugate gradient (SCG) algorithm was used in training of the ANN. The amount of geometry information of the propellers were varied as input to the algorithm and it was shown that more geometry information results in better prediction such that the mean relative error of efficiency prediction was reduced from 9.21% to 4.36%. Input layer has diameter, pitch, RPM, advance ratio (J), chord and twist distributions. Hidden layer has neurons number varying from 1 to 100. Output layer has thrust coefficient, power coefficient and efficiency. High coefficient of determination (R2) values were obtained around 0.99 and low percent errors obtained around 1.8% which suggests that ANN applications are useful tools in predicting propeller parameters and designing for propeller driven air vehicles.