Airbag fabric material modeling of nylon and polyester fabrics using a very simple neural network architecture

Abstract

The material properties of engineering fabrics that are used to manufacture airbags can not be modeled easily by the available nonlinear elastic–plastic shell elements. A nonlinear membrane element that incorporates an elaborate tissue material model has been widely used by the auto industry for the airbag simulation studies. This model is highly computation intensive and does not differentiate between the various physical properties of the fabrics like fiber denier, the polymer fiber, and weave pattern. This paper introduces a new modeling technique that uses artificial neural networks. Experimental permeability data for fabrics under biaxial strain conditions were obtained through a blister-inflation technique and were used to train the proposed network architecture. In this training environment, various properties of the fabric can be incorporated and the network can be trained to generalize relative to the environment. Once trained, the cause–effect pattern is assimilated by the network with approprate weights to produce a desired output. Fabrics tested in this study included nylon 66 fabrics with three different fabric deniers: 420, 630, & 840 and two types of weave, and two 650-denier polyster fabrics having different calendering effects. The predictions obtained from this neural network model agreed very well with the experimental data. This indicates that neural nets can be considered as a serious design tool use in determining permeability and biaxial stress–strain relationships for textile fabrics used in airbags. © 1996 John Wiley & Sons, Inc.