Development Of A Knowledge-Based Hybrid Methodology For Vehicle Side Impact Safety Design

Abstract

The present research work has been carried out to develop a unified knowledge-based hybrid methodology combining regression-based, lumped parameter and finite element analyses that can be implemented in the initial phase of vehicle design resulting in a superior side crash performance. As a first step, a regression-based model (RBM) is developed between the injury parameter Thoracic Trauma Index (TTI) of the rear SID and characteristic side impact dynamic response variables such as rear door velocity (final) and intrusion supplementing an existing RBM for front TTI prediction. In order to derive the rear TTI RBM, existing public domain vehicle crash test data provided by NHTSA has been used. A computer-based tool with a Graphical User Interface (GUI) has been developed for obtaining possible solution sets of response variables satisfying the regression relations for both front and rear TTI. As a next step in the formulation of the present hybrid methodology for vehicle side impact safety design, a new Lumped Parameter Model (LPM) representing NHTSA side impact is developed. The LPM developed consists of body sub-systems like B-pillar, front door, rear door and rocker (i.e. sill) on the struck side of the vehicle, MDB, and “rest of the vehicle” as lumped masses along with representative nonlinear springs between them. It has been envisaged that for the initial conceptual design to progress, the targets of dynamic response variables obtained from RBM should yield a set of spring characteristics broadly defining the required vehicle side structure. However, this is an inverse problem of dynamics which would require an inordinate amount of time to be solved iteratively. Hence a knowledge-based approach is adopted here to link the two sets of variables i.e., the dynamic response parameters (such as average door and B-pillar velocities, door intrusion, etc.) and the stiffness and strength characteristics of the springs present in LPM. In effect, this mapping is accomplished with the help of an artificial neural network (ANN) algorithm (referred to as ANN_RBM_LPM in the current work). To generate the required knowledge database for ANN_RBM_LPM, one thousand cases of LPM chosen with the help of the Latin Hypercube technique are run with varying spring characteristics. The goal of finding the desired design solutions describing vehicle geometry in an efficient manner is accomplished with the help of a second ANN algorithm which links sets of dynamic spring characteristics with sets of sectional properties of doors, B-pillar and rocker (referred as ANN_LPM_FEM in the current work). The implementation of this approach requires creation of a knowledge database containing paired sets of spring characteristics and sectional details just mentioned. The effectiveness of the hybrid methodology comprising both ANN_RBM_LPM and ANN_LPM_FEM is finally illustrated by improving the side impact performance of a Honda Accord finite element model. Thus, the unique knowledge-based hybrid approach developed here can be deployed in real world vehicle safety design for both new and existing vehicles leading to enormous saving of time and costly design iterations.