APPLICATION OF ARTIFICIAL NEURAL NETWORK IN PREDICTING THE ACTUAL MAXIMUM STRESS IN A TUNGSTEN INERT GAS (TIG) WELDMENT
Faculty
Department
Year of Publication
Keyword
upload
Publication Type
Abstract
Welding is a vital manufacturing process used in several industries, including aerospace, automotive, and construction. However, residual and induced stresses that develop during welding due to rapid heating and cooling cycles often affect the structural integrity of the weldment thereby reducing the integrity of the structure. The study investigates the application of Artificial Neural Networks (ANN) in predicting the actual maximum stress in Tungsten Inert Gas (TIG) weldments and develop a predictive model capable of accurately estimating the actual maximum stress in TIG welded joints based on key process parameters such as welding current, voltage, and gas flow rate.
Twenty (20) experimental runs as generated by the Central Composite Design (CCD) was used to carry out TIG welding on mild steel plates. A U niversal Stress Testing Machine was used to measure the actual maximum stress in the weldment and the result was recorded for each experimental run. This experimental result was then analyzed using ANN.
ANN trained the neural network using 70%(14) of the observations and used 15%(3) for network validation and another 15%(3) for network testing. The best validation performance value of 80.6689 was observed at epoch 5 with an overall performance value of 0.96864. The results revealed that the developed ANN model achieved high prediction accuracy with minimal error, confirming its capability to learn and represent the complex nonlinear relationship between the welding input parameters and the resulting actual maximum stress. ANN predicted response values was compared with the experimental result and it showed a meritorious correlation with the experimental result as well as the experimentation trend
Twenty (20) experimental runs as generated by the Central Composite Design (CCD) was used to carry out TIG welding on mild steel plates. A U niversal Stress Testing Machine was used to measure the actual maximum stress in the weldment and the result was recorded for each experimental run. This experimental result was then analyzed using ANN.
ANN trained the neural network using 70%(14) of the observations and used 15%(3) for network validation and another 15%(3) for network testing. The best validation performance value of 80.6689 was observed at epoch 5 with an overall performance value of 0.96864. The results revealed that the developed ANN model achieved high prediction accuracy with minimal error, confirming its capability to learn and represent the complex nonlinear relationship between the welding input parameters and the resulting actual maximum stress. ANN predicted response values was compared with the experimental result and it showed a meritorious correlation with the experimental result as well as the experimentation trend
Supervisor(s)
co-supervisor