Structure Property Correlation of Friction Stir Welded Al-Ce-Si-Mg Aluminium Alloy and Optimization of Process Parameters using Soft Computing Techniques

Abstract

The demand for Aluminium alloys for uses as structural material is growing day by day, due to their distinct benefit of high strength to weight ratio. However, these alloys show a great challenge during welding by conventional methods due to the physical properties being dissimilar to steel and other materials and the property of improved hardness. Solid state welding method offers an alternative to conventional welding methods and leads to the improved joint efficiency due to microstructural alteration. Researchers around the world are carrying out wide-ranging experiments on one such process known as Friction Stir welding or FSW to join the materials effectively in solid state. In the present research study, the evolution of microstructure at the weld zone during friction stir welding of Al-Ce-Si-Mg aluminium alloy (Al-10Mg-8Ce-3.5Si and Al-5Mg-8Ce-3.5Si), as well as the effect of Friction Stir Welding (FSW) on joint strength was carried out. The microstructural study of the FSW joint has been carried out using Scanning electron microscopy as per ASTM E112-12(2012). An EDAX analysis as per ASTM F1375-92(2012) and an Optical emission spectrometry as per ASTM E1251-11 have been carried out to ascertain the chemical composition. An x-ray diffraction has been carried out as per ASTM F2024 - 10(2016) to ascertain the phases present in the alloy. The tensile testing has been done as per ASTM E8-04 and Vickers hardness test as per ASTM E92–17. It is very difficult to identify the process variables to obtain the desired joint strength by conducting numerous individual experiments. Therefore, to analyse the welding process variables, the orthogonal array technique (OA) type Taguchi design of experiment helps in arriving at the best possible solution. Design of Experiments were adopted to find out the influence of various input process parameters on mechanical properties such as ultimate tensile strength (UTS), hardness and grain size of the joint and to predict the UTS of the joint. The Taguchi experiments showed that the tool pin shape, speed of tool rotation and speed of welding have a bearing on the quality of the FSW joints in aluminium alloys. It was observed that the grain size at Nugget Zone (NZ) is dependent on the speed of tool rotation, speed of welding, tool pin shape and composition of the aluminium alloy. The grain size at the bottom of the NZ was found to be decreasing as compared to the top of the NZ. It was observed that highest hardness was found at NZ. Minimum hardness was obtained at HAZ and all the tensile specimens fractured at HAZ. Optimal joint strength was obtained for a speed of tool rotation of 1000 rpm, speed of welding of 20 mm/min, tool of triangular pin shape and 10% (wt%) of Mg (Al-10Mg-8Ce-3.5Si) aluminium alloy. The Taguchi orthogonal array-based design has shown that the Tool pin shape has greater significance in increasing the joint strength, followed by speed of welding, Material composition and speed of tool rotation. A speed of tool rotation 1000rpm, speed of welding 20 mm/min, Triangular Profile Tool (TPP) tool and Al-10Mg-8Ce-3.5Si aluminium alloy were obtained as the optimum variables of the process. The percentage contribution of each of the input process variables on optimum output quality characteristics was also found out and found to be lying well within the confidence interval of 95% suggested by the Taguchi design. Further work is carried out to predict the model for aluminium alloy joints fabricated using friction stir welding using Artificial neural network (ANN) technique. The Multilayer perceptron neural network (MLP) with error back propagation learning algorithm is selected as it can predict the ultimate tensile strength, percentage of elongation and hardness of the joint for given rotation speed welding speed, tool pin profile and composition of the material. The validation of the predicted model is performed by conducting validation experiments. The prediction is done by the model, and later it is analysed to assist the suitability of the ANN prediction model. The present work has shown that the prediction results with ANN are more superior to those predicted using statistical methods like Taguchi Techniques.