Title: Effectiveness of frequency domain parameters in decision tree-based classification of GMAW defects

Authors: Thomas P. Rajan; K. Sundareswaran; P.R. Venkateswaran

Addresses: Mar Athanasius College of Engineering, Kothamangalam, Kerala, 686666, India ' National Institute of Technology, Tiruchirappalli, Tamil Nadu, 620015, India ' Bharat Heavy Electricals Limited, Bengaluru, Karnataka, 560012, India

Abstract: Capability of frequency domain parameters obtained through Fourier analysis of weld signature (current and voltage signals) in decision tree-based real-time weld defect classification is investigated. GMAW under standard weld conditions and with disturbance induced conditions are conducted. Three types of disturbances are introduced - fluctuations in arc length, air flow disturbances and surface contaminants. Experimental trials are done under each category and the real-time weld signature is measured using hall effect sensors through Labview and MATLAB-based instrumentation setup. Using discrete Fourier analysis of the signatures, four spectral parameters are proposed, which has potential to categorise the weld condition. These parameters are used as predictors to create a binary splitting decision tree capable of classifying the weld condition. Classification accuracies above 90% are obtained during training and testing phases, which indicate the effectiveness and advantages of using frequency domain parameters. Compared to time domain statistical approach, very few predictors are needed in frequency domain, to obtain comparable classification accuracy. The proposed system is simple and computationally less demanding, enabling real-time prediction and classification of GMAW disturbances. [Submitted 7 December 2021; Accepted 20 July 2022]

Keywords: gas metal arc welding; GMAW; defect classification; discrete Fourier transform analysis; frequency spectral parameters; decision tree classifier.

DOI: 10.1504/IJMR.2023.132839

International Journal of Manufacturing Research, 2023 Vol.18 No.3, pp.251 - 265

Received: 07 Dec 2021
Accepted: 20 Jul 2022
Published online: 11 Aug 2023 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article