Using a neural network model to evaluate the mechanical and tribological properties of vermicular cast iron based on hardness

Abstract

Measuring the tensile strength, wear resistance, and impact strength of metals, particularly cast iron, is complex and more expensive than performing hardness tests. In the present study, owing to the ease of specimen preparation and low cost, the Hardness (HB) test was used to approximately predict Wear Rate (WR), Impact Energy (IE), and tensile strength (TS). The relation between Mg% and HB, tensile strength, WR, and IE was examined by using three experimental groups of compacted graphite cast iron (CGI) treated with a nodulizer (Fe–Si–Mg) alloy at different carbon equivalents (CEs) of 3.5, 4.0, and 4.5 %. The produced CGI exhibited HB, TS, WR, and IE of 191–226 HB, 402–455 MPa, 30.1–23.8 mg/cm2, and 22–15 J, respectively. The good results were taken at a CE of 4.5 % and Mg content of 0.0118–0.0155 %. the regression analysis and artificial neural network model (ANNs) were used in the hardness test, and the results indicated the possibility of predicting IE, WR, tensile strength, and high accuracy Mg% of the produced CGI. It could be observed that, the neural network algorithm model has a high prediction precision for determining the Mg% content and the properties of the prepared CGI based on hardness. In the case of CE = 4, the MSE calculated for the predicted and measured data taken from the used ANNs model is 3.7 E−8, 20.33, 0.3084, and 0.099 for Mg%, TS, WR, and IE, respectively.