Inverse problem approach using photoelastic analysis and artificial neural networks in tandem
An ANN was trained to determine the applied loads on a mechanical component using photoelastic fringe orders. Two different beams, both fabricated from birefringent material, were used. One beam was subjected to pure bending and the other to four non-symmetrical and random loads. Completely random loads were applied to the beams and relative retardation (N) values were collected using a digital polariscope. These values were used as inputs to the ANN and loads were calculated. For the case of pure bending the average error between calculated loads and actual loads was <4% and for the un-symmetrical loading test it was <8%.
Inverse Problem Approach using Photoelastic Analysis and Artificial Neural Networks in Tandem