Neurowavelet packet analysis based on current signature for weld joint strength prediction in pulsed metal inert gas welding process

Abstract

The monitoring of welding process is crucial for the development of a real time quality control system for the pulsed metal inert gas welding (PMIGW) process. This work introduces an intelligent system for weld joint strength prediction in a PMIGW process based on the analysis of acquired current signal by wavelet packet transform. A thirteen-dimensional array of process features, i.e. six process parameters and seven wavelet packet features, are used to describe various welding conditions. These process features obtained from a set of experiments are employed as input vectors of an artificial neural network model to predict the corresponding weld joint strengths. The results, i.e. the prediction errors, show that the use of wavelet packet features gives much accurate prediction as compared to the use of the purely time domain features. IntroductionPulsed metal inert gas welding (PMIGW) is one of the major joining processes in modern manufacturing industries to produce high strength or high quality welds. One of the difficult tasks during arc welding is to maintain the consistency while making part after part. This is due to the fact that same process parameter settings do not guarantee the same weld quality. 1,2 As a result, there is a need for different technologies to precisely predict the weld quality, 1-4 more importantly the weld strength and the mechanical properties, 5-8 under varying welding operating conditions. Because of the complex nature of the process, artificial intelligence based methods are well suited for this purpose. Recently, different types of artificial neural networks and fuzzy logic systems have been developed for controlling the welding process and monitoring of weld quality. 3,4,[9][10][11][12][13][14][15][16][17][18][19] Among the various sensors used for weld quality monitoring, arc sensors are the most reliable, simple and competitive. 20,21 Moreover, a large number of researchers 22-30 have proposed arc sensing techniques for monitoring and control of various aspects of welding processes. Time domain features 3,4,31 of arc signals are less informative about the process, and are affected by the noises and disturbances. Furthermore, time frequency analysis with FFT is unsuitable, because the arc signals are time varying. Recently, wavelet transform has been applied in weld quality monitoring systems, 32,33 because this could eliminate most of the disadvantages of the FFT analysis. Wavelet transform has a good resolution in frequency and time domains synchronously; but there is no available published literature on its use to monitor arc welding processes. In this work, the acquired current signal was processed using wavelet packet transform analysis. Sensitivity analysis was also performed to find the best wavelet packet features. The best wavelet packet features and the process parameters were used in an ANN model for predicting the strength of the welded joint. Experimental Specimen preparationIn the present research, two mild steel specimens with dimensions of 125610068 mm were use...

Adaptive chirplet transform for sensitive and accurate monitoring of pulsed gas metal arc welding process

“…Recently however, the interest has shifted much to time-frequency analysis. Various forms of affine transforms have been used till date [13,14,[16][17][18][19]] to try to effectively analyze various nonstationary signals in the time-frequency domain. STFT, Gabor, Gabor-Wigner, wavelet, and chirplet transforms were the first to arrive.…”

“…So, significant efforts have been made in this regard to identify the weld joint strength on basis of various sensor outputs [13,14]. So a better method of analysis of the signals has been proposed to differentiate between a good quality and a bad quality weld.…”

“…Therefore, extraction of the various characteristics of these non-stationary signals is the main crux of the differentiating mechanism of the proposed method. Statistical signal processing techniques like wavelet analysis have been applied to this problem [14], but they could not successfully differentiate the signals on the basis of certain properties of the weld joint strength. This can be mainly attributed to the presence of high degree of nonlinearity in the sensor output of the signal.…”

“…In the previous studies [14], the approach used to study the weld strength joints in various processes of welding that has been used to predict the weld strength joints from the sensor outputs is the wavelet transformation technique. But certain shortcomings have been observed in them.…”

A novel hybrid intelligent optimization model for twin wire tandem co-pool high-speed submerged arc welding of steel plate

“…Therefore, the research of numerical indexes that can real-time, comprehensively and quantitatively represent the rationality of welding parameters and quality is the key to realize the selection of parameters by mathematical method. Arc energy signals (current, voltage) of welding process contains a wealth of real-time and dynamical information, it contains not only the information of welding power source performance but also the welding quality information such as welding arc stability and quality (Dickinson, et al, 1980, Eckelt, et al, 1989, Wu, et al, 2001, Absi Alfaro, et al, 2006, Pal, et al, 2008, Cullen, et al, 2008, Li, 2012. It is important method for welding quality monitoring to extract feature parameters quantitatively representing the rationality of welding parameters and quality.…”