Abstract
The tire industry is moving away from the design and development of new tires by the process of trial and error, i.e. design, build, test and review. The ready availability of powerful computer hardware and software makes it possible to focus more on the design of the tire before a prototype is built. The ability to analyse the performance of the tire in a virtual environment can enable its design to be optimised to the product specification while achieving considerable savings in time and cost associated with the iterative process of building and testing prototypes.
Modelling the behaviour of tires in a virtual environment can be performed in different ways including mathematical modelling and finite element analysis. Conducting any of the aforementioned approaches requires the availability of a predefined and detailed configuration of the tire. It then requires that the modelling process for the new tire design be repeated iteratively until a design that satisfies the desired performance requirements is obtained. This process can be very time consuming and expensive.
To tackle this problem of high development cost, an Artificial Intelligence (AI) based method is introduced in this paper, which can lead to a quicker and logical achievement of the desired tire properties. In this method, several configurations of vehicle tire are first used to obtain tire performance characteristics using validated Finite Element models. The tire performance characteristics obtained in this stage along with the tire design input parameters enables a group of tire input and output data to be generated.
The next stage is based on utilizing artificial neural network (ANN) to learn the table of tire properties. The use of ANNs has been recognized as a powerful Artificial Intelligence method for modelling the nonlinear relations between the data which are captured from either natural or inanimate events. To use ANNs, there is a need to separate the data into inputs and outputs. Inputs are the design parameters of the tire which cause the differences in the performance characteristics of different tires. In this paper, some of the design parameters considered as ANN model input include ply thickness, number of carcass plies, number of belt plies, cord end density, belt ply angle etc. ANN outputs are tire performance characteristics that may be of interest. These are mostly obtained from the FE analysis including radial stiffness, lateral stiffness, cornering stiffness etc.
The validity of this relation between input and output is examined with different tire configurations and it is shown to be a reliable and very fast approach to predicting the properties of new generated tire configurations without using FE analysis. Once a tire configuration which satisfies the desired performance requirements has been achieved, a full FE analysis can then be carried out to confirm the validity of the design. In this way, parametric studies are carried out using the ANN model and expensive FE analysis is only carried out once the appropriate tire design configuration that will produce the desired performance characteristics has been determined. The benefits of this technique are considerable savings in design cycle time and costs.
Keywords: artificial intelligence, design parameters, modelling, neural network, tire configuration.