Development of a welding residual stress predictor using a function-replacing hybrid system

Kumanan, S.; Kumar, Rishi; Dhas, J. Edwin Raja

doi:10.1007/s00170-005-0297-1

Cited by 8 publications

(1 citation statement)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Genetic algorithms are being attempted in guiding the weight updating process in the neural network technique [36,37] and scopes to build fast, accurate predictive networks. GA is effectively applied with neural networks to predict machining forces [38] , predict weld residual stress [39] and optimization of frictional vibration joining process [40] . PSO is used instead of backpropagation algorithm in neural network for the prediction of tool life [41] .…”

Section: Introductionmentioning

confidence: 99%

Prediction of weld quality using intelligent decision making tools

Dhas¹,

Kumanan²,

Jesuthanam³

2012

AIR

View full text Add to dashboard Cite

Decision-making process in manufacturing environment is increasingly difficult due to the rapid changes in design anddemand of quality products. To make decision making process online, effective and efficient artificial intelligent tools likeneural networks are being attempted. This paper proposes the development of neural network models for prediction ofweld quality in Submerged Arc Welding (SAW). Experiments are designed according to Taguchi’s principles andmathematical equations are developed using multiple regression model. Proposed neural network models are developedusing experimental data, supported with the data generated by regression model. The performances of the developedmodels are compared in terms of computational speed and prediction accuracy. It is found that Neural Network trainedwith Particle Swarm Optimization (NNPSO) performs better than Neural Network trained with Back Propagation (BPNN)algorithm, Radial Basis Functional Neural Network (RBFNN) and Neural Network trained with Genetic Algorithm(NNGA). The developed scheme for weld quality prediction is flexible, competent, and accurate than existing models andit scopes better online monitoring system. Finally the developed models are validated. The proposed and developedtechnique finds a good scope and a better future in the relevant field where human can avoid unwanted risks duringoperations with the deployment of robots.

show abstract