Multi-objective optimization for 3D printed origami crash box cell based on artificial neural networks and NSGA-II

Authors:

Abstract
Composite structures are increasingly used in the automotive industry due to their lightweight and specific energy absorption capabilities, while 3D printing is also widely used in industry because of their high efficiency and high precision. Recently, the Origami crash box (OCB) has been proposed as an energy absorber for automobiles because of their low initial peak load and high average load. Experiments and theory have shown that the energy-absorbing effect of OCB will change significantly with the change in size. Since OCB is composed of multiple OCB cells, therefore, it is necessary to develop a model that can predict the energy absorption effect according to scale change of OCB cell. And this model is utilized to optimize the size to maximize its energy absorption capability while reducing the initial peak force. This paper explores the energy absorption effect of 3D printed OCB, which is made of carbon fiber-reinforced nylon in the same wight and thickness with a stable surface area of 14400mm2. The Artificial Neural Network (ANN) model which used Mean Squared Error (MSE) to measure its accuracy is established to predict high non-linear behavior of OCB cell at different size. And then the Non-dominated Sorting Genetic Algorithm (NSGA-II) in which initial Peak Crush Force (PCF) and Energy Absorption (EA) are used as optimization metrics, is applied to complete the multi-objective optimization. The utilized ANN model precisely predicts the variation of load capability with displacement in different size of OCB cell with an MSE as 0.046kN², energy error as 5.97J and PCF error as 0.17kN. A configuration of OCB is generated by NSGA-II shows superior performance than standard OCB cell. In terms of prediction, there is a 13.5% decrease in PCF, reducing it from 2.75 kN to 2.38 kN, while EA experiences a 7.8% increase, rising from 34.5 J to 37.2 J. In experimental results, PCF exhibits a 14% reduction, decreasing from 3.08 kN to 2.65 kN, while EA shows a 14.3% increase, climbing from 30.61 J to 35 J.

This publication has 18 references indexed in Scilit: