In recent years, on-line weld monitoring is the potential area of research. In this work, torch current deviation prediction systems are developed with Artificial Neural Networks to produce welds free from Lack of Penetration. Lack of penetration is deliberately introduced by varying the torch current. Thermographs are acquired during welding and hotspots are extracted using Euclidean Distance based segmentation and are quantitatively characterized using the second order central moments. Exemplars are then created with central moments as input parameters and deviation in torch current as the output parameter. Radial Basis Networks (RBN) and Generalized Regressive Neural Networks (GRNN) are then trained and tested to assess the suitability for torch current prediction. GRNN outperforms RBN in predicting the torch current deviation with 98.95 % accuracy.
CITATION STYLE
Nandhitha, N. M. (2016). Artificial neural network based prediction techniques for torch current deviation to produce defect-free welds in GTAW using IR thermography. In Smart Innovation, Systems and Technologies (Vol. 43, pp. 137–142). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-81-322-2538-6_14
Mendeley helps you to discover research relevant for your work.