OPTIMIZATION OF ALUMINUM-CFRP HYBRID CRASHWORTHY COLUMN IN THE CASE OF LATERAL LOAD

• Main Author: Adiwiguna, Tias
• Format: Theses
• Language: Indonesia
• Subjects: Teknik (Rekayasa, enjinering dan kegiatan berkaitan)
• Online Access: https://digilib.itb.ac.id/gdl/view/57104
• Institution: Institut Teknologi Bandung

This research was motivated by the lack of research on optimization of crashworthy columns in lateral loading with purpose to determining the optimal hybrid column configuration and method, knowing the effectiveness of the optimal model against numerical and experimental models and knowing the factors that influence the results of the comparison of crashworthiness parameters of optimal, numerical and model models experiment. This research uses numerical studies, optimization with ANN, NSGA-II and MOORA, as well as experimental methods for validation. The research began with the creation of a basic numerical model in the form of a circular column with 6063-T6 aluminum material and Carbon Fiber Reinforced Plastics fiber direction [-45/45/45/-45] with a thickness of 1.5 mm each. The basic model was then validated by experimental methods using a 25 kN loadcell and the configuration of the basic model was then used as the basis for forming a dataset of 150 data with various cross-sectional geometries of circles, squares, hexagons, octagons, tophats and double tophats for ANN optimization. Based on the ANN results, the thickness of aluminum and the composite layer has the most influence on the weight of the model. ANN produces weight and bias equations which are then used for optimization of NSGA-II and MOORA. The optimal numerical model was obtained with an octagonal shape, 3 mm thick aluminum and 8 layers of composite with fiber direction [51/36/24/29/29/24/36/51] and resulted in a significant improvement from the basic model, where there was an increase in specific moment by 115.3%, increase in specific force by 115.3%, increase in SEA by 90.75% and increase in critical rotation angle by 61.36%, but this is also followed by a consequence of an almost fourfold increase in mass. The numerical model is then validated with an acceptable error value. Errors may occur due to the use of loadcell (25 kN) which is still far from the value of the force that occurs (8 kN) causing the experimental data to be less accurate In addition, the optimal model of machine learning has also been verified with a numerical model, and also produces an acceptable error (below 10%). Several factors that influence the error are the use of loadcells that are too large and the internal setting factors of the numerical modeling media that are not considered by machine learning.