An efficient knowledge-based hybrid methodology for vehicle crash safety design
Indian Institute of Science
Last modified: October 1, 2007
A complex engineering system such as an automobile or an aircraft is highly multidisciplinary in nature. Multidisciplinary design optimization which is currently being vigorously pursued tries to find a rational trade-off spanning at least some of the disciplines which influence a major sub-system of a product. However, without a robust initial design, an optimization procedure will fail to provide the most desirable solution. The task of arriving at a suitable initial design is complicated by the fact that it represents an inverse solution for a problem with output responses (i.e. performance parameters) being given as targets and values of input parameters being unknown. The input parameters are typically the design variables such as geometric dimensions and material properties of parts. The current talk will enumerate a knowledge-based technique pertaining to the crash safety design of a vehicle which requires nonlinear numerical models for assessing structural crashworthiness and occupant safety. Keeping in mind that design of complex systems is evolutionary, system-level safety targets are at first converted into target values of dynamic response parameters using test data-based regression models. For further progression of design, specifications are obtained for body stiffness and strength parameters defined in a spring-mass idealization of a vehicle in a given crash test mode using an artificial neural network which has been trained with a large number of forward dynamic solutions. Thus, instead of following a conventional trial-and-error procedure, the present technique can efficiently provide a rational initial solution in a single step using a scientific knowledge database.