One of the concerns for several industries like automotive, aircrafts, and marine is to improve the surface of a manufactured component to make it corrosion resistant.
Cladding is one of the most common methods that is used for realizing such needs as a means of increasing the durability of products. Weld beads of poor quality tend to decrease the effectiveness of cladded surfaces. Choosing the right process parameters is, therefore, essential to ensure weld bead quality.
In this work, geometrical features of a weld bead, such as reinforcement height, depth of penetration, width of weld bead, reinforcement form factor and penetration shape factor, are tried to predict using artificial neural networks in order to examine the effects of weld current, travelling speed and heat input. Moreover, the number of neurons and hidden layers in different structured neural networks are compared to see how they affect prediction performance.
For this, nine beads-on-plate experiments are carried out using gas metal arc welding with two replications. Low carbon steel is used as a base plate and austenitic stainless steel is used as the electrode.
Findings demonstrate that prediction efficiency increases with increasing nodes in the lower number of hidden layers, while increasing neurons in a higher number of layers increases prediction error. On the whole, appreciable prediction accuracy is achieved.
This article is shared by Rajat Kumar Paul, Avishek Mukherjee and Santanu Das of Department of Mechanical Engineering, Kalyani Government Engineering College, West Bengal