Multiple Nonlinear Neuro-Regression Modeling And Design Optimization Of The Rocket’s Different Sub-Units


Dinç M., Sargın T., Aydın L., Gültürk E.

1st International Congress on Artificial Intelligence and Data Science, İzmir, Turkey, 26 - 28 November 2021, pp.330-336

  • Publication Type: Conference Paper / Full Text
  • City: İzmir
  • Country: Turkey
  • Page Numbers: pp.330-336

Abstract

Rockets are widely used in aerospace and defense with their speed and improved mechanical properties. The skills and mechanical properties of rockets can be developed depending on the design methods, production process and structural features. Lots of design parameters in the rocket modeling process has an impact on the structural features In this study, the effects of fourteen different major design variables on the output parameters in rocket modeling process were investigated. This study was carried out in two stages, simulation and design-optimization. In the first part, scenarios were determined by using the Design of Experiment (DoE) approach to collect data and these scenarios were realized through the OpenRocket simulator. MacroRecorder app was used to speed up to process applied tries on the OpenRocket and the outputs such as maximum velocity, apogee were recorded. In the second part, different mathematical models were created to define the phenomena by using the multiple nonlinear regression analysis with combining neuro-regression method. The coefficient of determination (R2 ), adjusted coefficient of determination (R2 adjusted) also R 2 training and R2 testing values were calculated for each model, to see how well the models define the phenomena. As a design-oriented solution, the values of the process parameters that provide the maximum speed values are optimized based on stochastic optimization algorithms (Differential Evolution Algorithm, DE). The results show that modeling and optimization are important in achieving higher efficiency in the rocket modeling process.