A computational inverse technique for material characterization of a functionally graded cylinder using a progressive neural network

Abstract

A computational inverse technique of neural network (NN) by means of elastic waves to material characterization of functionally graded material (FGM) cylinder is presented. The displacement responses on the outer surface are used as the inputs for the NN model. The outputs of the NN are the material property of FGM cylinder. The analytical–numerical method is used as the forward solver to calculate the displacement responses of FGM cylinder to an incident wave for the known material property. The NN model is trained using the results from the forward solver. Once trained by, the NN model can be used for on-line characterization of material property if the dynamic displacement responses on the outer surface of the cylinder can be obtained. The characterized material property is then used to calculate the displacement responses. The NN model would go through a progressive retraining process until the calculated displacement responses using the characterized result are sufficiently close to the actual responses. This procedure is examined for material characterization of an actual FGM cylinder composed of stainless steel and silicon nitride. It is found that the present procedure is very robust for determining the material property distribution in the thickness direction of FGM cylinders.