Prediction of hardness in friction stir processing by artificial neural networks

Abstract

This research focuses on the use of Artificial Neural Network (ANN) for the prediction of the microhardness of friction stir processed aluminium based metal matrix composite (AA6061+Al₂O₃). Different specimens were obtained by using rotating speeds of 1100, 1210, 1320 and 1430 rpm and travelling speeds of 36, 48, 60, 72 mm/min. The microhardness value (HV) of the processed surface of each of the samples was measured and the data collected from the specimens was used as learning data for ANN. Higher rotational speed and lower transversal speeds resulted in higher hardness value since processing at higher tool rotational speed causes high material flow and good resistance to the tool pin profile. The increase of tool rotational speed also leads to the decrease of spindle torque due to ease of flow of material. A uniform increase in microhardness was observed up to 1320 RPM and a subsequent decrease on any further increments of tool rotational speed. Subsequently, the highest values of microhardness were observed with a square mandrel at 1320 RPM and 36 mm/min.  The calculated results were satisfactorily compliant with the measured data and the ANN model was successful in predicting the microhardness.