Prediction of Nonlinear Stiffness of Automotive Bushings by Artificial Neural Network Models Trained by Data from Finite
- PDF / 544,375 Bytes
- 13 Pages / 595 x 842 pts (A4) Page_size
- 4 Downloads / 262 Views
ght © 2020 KSAE/ 11820 pISSN 12299138/ eISSN 19763832
PREDICTION OF NONLINEAR STIFFNESS OF AUTOMOTIVE BUSHINGS BY ARTIFICIAL NEURAL NETWORK MODELS TRAINED BY DATA FROM FINITE ELEMENT ANALYSIS Yeon-Woo Jung1) and Heung-Kyu Kim2)* Graduate School of Automotive Engineering, Kookmin University, Seoul 02707, Korea 2) Department of Automotive Engineering, Kookmin University, Seoul 02707, Korea
1)
(Received 19 November 2019; Revised 18 February 2020; Accepted 3 March 2020) ABSTRACTDue to the nonlinear behavior of rubber for bushings, the prediction of mechanical properties of the bushing requires nonlinear finite element analysis (FEA) techniques and a lot of computation time. Therefore, we propose a method to efficiently predict the stiffness of bushings using an Artificial Neural Network (ANN) model trained by data from FEA. First, FEA was performed for the designed 3D and 2D bushing models. Based on the relationship between the bushing shape design variables and the stiffness values predicted by the FEA, we trained the Multilayer Perceptron (MLP) and the Convolutional Neural Network (CNN) models among the ANN models. Given the shape design variables of the bushing model, the stiffness values were predicted by the MLP model. Given the image of the bushing model, the stiffness values were predicted by the CNN model. The stiffness prediction results showed that both models can be used to predict the stiffness of the bushings, and that the CNN model is slightly more accurate than the MLP model. In particular, it is expected that designers can easily estimate stiffness values by taking advantage of the CNN model which can use photographic images of real parts as inputs. KEY WORDS : Bushing, Rubber, Finite Element Analysis, Stiffness, Artificial Neural Network, Multilayer Perceptron, Convolutional Neural Network
1. INTRODUCTION
investigated deformation behavior such as torsional deflection, axial deflection, radial deflection and tilting deflection of rubber bushings. Their study was applied to cylindrical bushings with a constant cross section, and the predicted results were compared with experimental results. However, the bushings used in real vehicles are designed with very complex geometries to meet a number of design goals, such as the ride comfort of the vehicle. If the shape of the bushing is not so simple, it is almost impossible to predict the mechanical properties of the bushing by analytical methods. In other words, numerical methods such as Finite Element Analysis (FEA) must be used. In general, FEA can calculate the mechanical behavior of a part with arbitrary geometry because of the advantages of numerical methods (Busfield and Davies, 2001; Kadlowec et al., 2003; Kadlowec et al., 2009). By using FEA, the mechanical properties of a part can be predicted and incorporated into the design before the part is actually manufactured. However, as mentioned earlier, the vehicle bushing is not only composed of rubber with nonlinear behavior, but also very complicated in shape. Therefore, even with FEA, a large amo
Data Loading...
