PREDICTIVE CHARACTERIZATION MODEL FOR IMPACT CUSHIONING CURVES - CONFIGURING THE PREDICTIVE CHARACTERIZATION MODEL

Engineers and designers utilize mechanical properties and material beh avior to assist in the design and manufacture of products. The materia l data obtained from standard tables tend to be general and may not co rrelate well with the actual material being used. To meet the design s pecifications, a larger number of iterative experimental tests than pl anned are usually conducted, This paper explores the use of neural net works as a predictive approach to characterize the impact cushioning c urves so as to reduce the number of experimental tests required, Key d esign considerations in configuring a neural network for optimal perfo rmance are also highlighted. This approach is able to predict the poin ts on the curves quite accurately but does have some limitations. To d evelop an effective predictive characterization model, the neural netw orks need to couple with appropriate algorithms so as to obtain a set of randomly distributed training data and generate the requisite point s for curve characterization. Two algorithms are developed and found t o be suitable for this purpose.