ISSN 2360-7955
Abstract: Residual stresses are those stresses that remain in a solid material even in the absence of external loading or thermal gradients. Residual stresses form a balanced force system within an object, as all forces and moments acting on one plane through the entire object must sum to zero. At cold temperatures, the stress after welding is not completely released because of the fast-cooling speed, which results in cracks. The central composite design model patterned the experimental matrix. The tungsten inert gas (TIG) welding kit was employed to weld the plates after chamfering their edges. 100 mild steel coupons, each measuring 80 x 40 x 10 mm, were used for the experiments. The experiment was conducted 20 times, with 5 specimens for each run. A 10 mm thick mild steel plate was chosen as the material for the experiment. This study is applying artificial neural networks to optimize and predict the residual stress of machined heat affected zone of mild steel welds using current, voltage and gas flow rate as the input variables. The Artificial neural network (ANN) was used to optimize and predict the residual stress of the weld specimen. 70% of the data was used for training, 15 % was used for validating and the remaining 15% for the actual test. It was observed from the analysis that it had 3 input neurons, 10 hidden neurons and 1 output neuron with gradient of 95.8669 at epoch 12 and validation check of 6 out of 6. It was also observed from the model summary statistics that a robust R2 value of 86.4% was obtained, with an adjusted R2 of 85.6%.
Keywords: modelling, residual stress, weld specimen, artificial neural network