Affiliations: [a] Department of Mechanical Engineering, SNS College of Engineering, Coimbatore, Tamil Nadu, India | [b] Senior Assistant Professor, SASTRA University, Tanjore, India
Correspondence:
[*]
Corresponding author: R. Sudhakaran, Department of Mechanical Engineering, SNS College of Engineering, Coimbatore – 641107, Tamil Nadu, India. Tel.: +91 422 6465201 (O); Fax: +91 422 6465202; E-mail: absudha@yahoo.com
Abstract: The quality of stainless steel weld is highly influenced by delta ferrite content expressed in terms of ferrite number. The quantity of delta ferrite content formed is controlled by the process parameters. This paper discusses the development of artificial neural network model for predicting the ferrite number in 202 grade stainless steel gas tungsten arc welded plates (GTAW). The process parameters chosen for study are welding gun angle, welding speed, plate length, welding current and shielding gas flow rate. The experiments were conducted based on design of experiments – fractional factorial with 125 runs. Using the experimental data, feed forward, back propagation neural models were developed and trained using Levenberg Marquardt algorithm. The training, learning, performance and transfer functions used are trainlm, learningdm, mean square error and tansig respectively. Five networks were developed with five neurons in the input layer, 1 neuron in the output layer and different nodes in the hidden layer. They are 5-3-1, 5-5-1, 5-10-1, 5-11-1 and 5-15-1. It was found that the artificial neural network (ANN) model based on network 5-5-1 predicted ferrite number more accurately than other networks. The prediction helps in identifying the recommended combination of process parameters to achieve a desired ferrite number in GTAW of stainless steel 202 grade plates.
Keywords: Gas tungsten arc welding, ferrite number, design of experiments, fractional factorial, artificial neural networks
