Prediction of the mechanical behavior of Passenger airbag module by design factors based on deep neural network
Topic(s) : Industrial applications
Co-authors :
Gyuwon KIM (KOREA, REPUBLIC OF), Jae Hyun AHN , Se-Min LEE (KOREA, REPUBLIC OF), Hyun-Ji RHO (KOREA, REPUBLIC OF), Hak-Sung KIMAbstract :
Airbags play an important role in the safety of occupants during collisions of automotive. Therefore, it is important to understand the mechanical behavior of the airbag and ensure that the airbag is deployed at the appropriate time during collisions of automotive. However, it is difficult to predict the behavior of the airbag according to the dimensions of passenger airbag (PAB) module components because that components interact with each other.
Recently, finite element analysis (FEA) is widely used to predict airbag deployment [1]. However, FEA consumes a lot of time and cost from modeling to analysis of simulation results, the development of a more economical method of predicting airbag deployment is required in automotive industry. In this work, a deep neural network (DNN) was proposed to predict the behavior of PAB module according to the deployment of PAB door. The DNN model was trained based on a simulation database through the 3D PAB equivalent FEA models developed to reduce simulation time. In the FEA model of the PAB module, the strain rate dependent mechanical properties were substituted for the mechanical properties of Polypropylene (PP) and thermoplastic olefin (TPO) which are the materials that make up the PAB module. The 3D PAB equivalent FEA model was developed by comparing the 3D Cockpit PAB model with key data such as the angular velocity of the deployment of airbag doors and the housing deformation. In addition, through the PAB equivalent model, the effects of design factors as shown in Fig. 1 on the physical behavior of the airbag module such as the stress distribution and the symmetrical deployment of airbag doors were analyzed. The design factors and the mechanical behavior of the PAB door were used as the training input and output of the DNN model, respectively. For the input data, the failure lines and the hinge dimension of PAB were used. Also, the output layer consists of the coordinates of crash pad and housing, and stress distribution. To optimize the DNN, the hyper-parameters of the network such as the number of nodes and layers were determined using the Keras hyperband optimization algorithm [2]. As a result, the mechanical behavior of PAB module according to the design factors of the main parts of the PAB door could be successfully predicted by the DNN with high accuracy in a short time.
Recently, finite element analysis (FEA) is widely used to predict airbag deployment [1]. However, FEA consumes a lot of time and cost from modeling to analysis of simulation results, the development of a more economical method of predicting airbag deployment is required in automotive industry. In this work, a deep neural network (DNN) was proposed to predict the behavior of PAB module according to the deployment of PAB door. The DNN model was trained based on a simulation database through the 3D PAB equivalent FEA models developed to reduce simulation time. In the FEA model of the PAB module, the strain rate dependent mechanical properties were substituted for the mechanical properties of Polypropylene (PP) and thermoplastic olefin (TPO) which are the materials that make up the PAB module. The 3D PAB equivalent FEA model was developed by comparing the 3D Cockpit PAB model with key data such as the angular velocity of the deployment of airbag doors and the housing deformation. In addition, through the PAB equivalent model, the effects of design factors as shown in Fig. 1 on the physical behavior of the airbag module such as the stress distribution and the symmetrical deployment of airbag doors were analyzed. The design factors and the mechanical behavior of the PAB door were used as the training input and output of the DNN model, respectively. For the input data, the failure lines and the hinge dimension of PAB were used. Also, the output layer consists of the coordinates of crash pad and housing, and stress distribution. To optimize the DNN, the hyper-parameters of the network such as the number of nodes and layers were determined using the Keras hyperband optimization algorithm [2]. As a result, the mechanical behavior of PAB module according to the design factors of the main parts of the PAB door could be successfully predicted by the DNN with high accuracy in a short time.