Investigation through Artificial Neural Networks on the Influence of Shot Peening on the Hardness of ASTM TX304HB Stainless Steel

The present research is aimed at obtaining and experimentally validating an artificial neural network to predict the hardness of TX304HB steel tubes subjected to shot peening. The experimental scope consisted of 228 tubes. Seven variables were considered as input parameters: rotation speed, line speed, material flow, air pressure, the size of the nozzle, and the internal diameter of the tubes; the experimental data demonstrated the need for considering the material bulk hardness as an input variable. One specimen from each tube was taken and subjected to Vickers microhardness tests at a depth of 40 μm from the interior circumference as well as a depth beyond the influence of the shot peening (bulk condition). The hardness was proven to follow Gaussian distribution. Therefore, a neural network was designed and tuned to provide the mean and standard deviation of the hardness for each of the combinations of input variables. The neural networks designed in this way were able to faithfully reproduce the experimental results. Several statistical parameters were determined to measure the goodness of the fitting. Thus, the correlation between experimental and predicted numerical values of mean hardness yields R2 = 0.7651 and a mean absolute percentage error of 1.547 % for the training data set and 0.7402 and 2.054 % for the test data set. The corresponding values for the prediction of the standard deviation are R2 = 0.4713 and 17.946 % for the training set and R2 = 0.6847 and 17.071 % for the test set.