Experimental Investigation to analyse the Mechanical Properties of Weld Strength Factor Performed by TIG Welding

Abstract

The study presents, mild steel plate, cut with dimensions of 60 mm x 40 mm, then welded with 100% argon gas by the TIG welding using design of experiment, Response Surface Methodology (RSM) and Artificial Neural Network optimization techniques. Welding current, gas flow rate, and voltage, have been selected as the process parameters during the TIG welding process. The effects of these process parameters on the weld strength factor were identified using analytical and computational intelligence techniques. The design of experiment, and Artificial Neural Networ optimization techniques were used to optimize the effect on Weld Strength Factor of the welded joints. An orthogonal array of the central composite design was prepared by the design of experiment (DOE) methodology in which experiments were performed duly as per this orthogonal array obtained. The 210.00A, 22.66 V, and 20.00 gas flow rate optimum setting of input parameters provides the better results for the weld strength factor. This solution was selected by design expert as the optimal solution having a desirability value of 0.880. The study reveals the successful use of artificial neural networks in predicting the weld strength for tungsten inert gas welding of mild steel plates. The mean square error was used to measure the performance of the network in each run. The mean square performance index for the network is a quadratic function. The input data are randomly divided into three sets. 70% are used to train the network, 15% are used to validate the network performance and 15% are used for the test. The validation of the network model produced a correlation value of 94.0% with a mean square error of 1.040E-4. the testing of the network model produced a correlation of 97.7% with mean square error 1.003E-5. The performance plot showed that the model developed was learning, which is expected of a very good network. The artificial network model produced predicted values for the weld strength of which the predicted values and the experimental values of the responses, closely fit and are in reasonable agreement with a high coefficient of correlation.