Artificial neural network modelling of cold-crack resistance of high strength low alloy steel 950A

• Published in: Journal of engineering (Stevenage, England) Vol. 2019; no. 2; pp. 447 - 454
• Main Authors: Manivelmuralidaran, Velumani, Senthilkumar, Krishnasamy
• Format: Journal Article
• Language: English
• Published: The Institution of Engineering and Technology 01.02.2019; Wiley
• Subjects: alloy steel; ANN; arc welding; artificial neural network model; artificial neural network modelling; cold cracking resistance; cold-crack resistance; cracks; current 950.0 A; experimental data; gas metal arc welding process; given input parameters; heat input; high strength low; identified process parameters; impact strength; neural nets; oxide particle content; oxide particles content; plate welding; plates (structures); predicted observed values; preheating temperature; propagation model; Research Article; sensitivity analysis; weld metal; welded joints; welding; welds; impact strength; ANN; sensitivity analysis; plate welding; gas metal arc welding process; heat input; predicted observed values; welds; welded joints; identified process parameters; artificial neural network model; cracks; current 950.0 A; preheating temperature; oxide particle content; oxide particles content; welding; cold-crack resistance; high strength low; alloy steel; plates (structures); weld metal; arc welding; cold cracking resistance; propagation model; given input parameters; artificial neural network modelling; neural nets; experimental data
• ISSN: 2051-3305; 2051-3305
• DOI: 10.1049/joe.2018.5277

The objective of the study is to predict the cold cracking resistance of high strength low alloy 950A welded joints using an artificial neural network (ANN) model. A bead on plate welding is carried out using the gas metal arc welding process. The identified process parameters for the ANN are preheating temperature, oxide particle content, and heat input. The impact strength of the weld metal is considered as the output parameter. A feed-forward back propagation model with ten neurons in the hidden layer is developed to predict the impact strength of the weld metal. The neural network model is created, trained, and tested with a set of experimental data. The proposed model correctly predicted the impact strength of the given input parameters. The predicted value of the impact strength is in agreement with the experimental data. The error percentage between the predicted and observed values is <5% and the root mean square error value is 2.2%. Sensitivity analysis is performed to identify the significance of input parameters. It is evident that the preheating temperature contributes 50.04%, oxide particles content contributes 37.15%, and heat input contributes 12.81% to impact strength.