APPLICATION OF ANN AND NATURE-INSPIRED ALGORITHM IN DESIGN OPTIMIZATION OF THE FRAME OF A LARGE-SCALE 3D CONCRETE PRINTER

Abstract

This article presents the application of artificial neural networks (ANNs) in computing and optimizing frame structure in a 3D concrete printer. An ANN (with 6 inputs, 30 neurons in hidden layer and 4 outputs) was developed to predict the system natural frequency f1 and displacement values uX, uY, uZ of nozzle according to 6 input parameters. The survey showed that the ANN has good predictive performance with the prediction errors uX, uY, prestige less than 3% and the errors of uZ, f1 less than 5%. This ANN model was embedded into the genetic algorithm (GA) to optimize the frame structure. The results of searching for an optimization solution indicate that although utilizing an ANN-GA combination took over 77 seconds, using GA directly interacting with the Ansys APDL-Matlab code took about 240 ÷ 300 minutes. Particularly, the optimization process was performed on a discrete spatial domain based on the dimensions of the standard steel box, so the optimal design solution can be used directly in completing the mechanical design of the 3D concrete printer.