Elsevier

Measurement

Volume 158, 1 July 2020, 107683
Measurement

Quality monitoring of aluminum alloy DPMIG welding based on broadband mode decomposition and MMC-FCH

https://doi.org/10.1016/j.measurement.2020.107683Get rights and content

Highlights

  • Broadband mode decomposition is proposed for decomposing broadband signals with “sharp corners”.

  • An associative dictionary library consisting of typical broadband signals and narrowband signals is constructed.

  • ACROA is adopted for determining the optimized parameter by using the smoothness function as the objective function.

  • BMD can effectively monitor the quality of aluminum alloy DPMIG welding combined with MMC-FCH.

Abstract

In double pulse metal inert gas (DPMIG) welding, the input broadband electrical signals are often affected by strong noise, which will decrease the quality monitoring accuracy. Therefore, a suitable method should be applied to extract features from the signals. However, due to the Gibbs phenomenon and the interpolation of extreme points, former methods such as variational mode decomposition (VMD) and ensemble empirical mode decomposition (EEMD) will generate unavoidable error. Therefore, broadband mode decomposition (BMD) method is newly proposed in this paper by constructing an associative dictionary library consisting of typical broadband and narrowband signals. Therefore, the drawbacks of the former methods can be avoided by searching in the dictionary. Analysis results indicate that by combining with flexible convex hulls (MMC-FCH), BMD is more accurate in extracting broadband components. Meanwhile, the mean accuracy of quality monitoring can be increased from 92.19% (VMD) and 93.75% (EEMD) to 100% by applying BMD.

Introduction

Metal inert gas (MIG) welding has been widely applied in mechanical field, due to the low cost and high efficiency. Extracting the feature information of various arc welding signals is often applied for evaluating the stability and monitor the quality of MIG welding. There are many ways for the measurement of welding quality, such as electrical signals, spectroscopy signals and sound signals. Zhao et al. studied the effect of current parameters on the heat transfer behavior of MIG welding, results showed the current pulsing parameters strongly affects the welding quality [1]. Madhulika et al. showed that optical microscopy image can be effectively applied for the surface property evaluation of welding joints [2]. Pal et al. applied arc sound and welding temperature for the MIG welding quality evaluation and showed a good performance in the prediction of metal deposition [3]. Among them, using electrical signals to monitor the MIG welding quality is a useful method. The effective information in the electrical signals can reflect the stability and the quality of arc welding, as the energy injected into the seam is mostly depend on the voltage and current inputs. As early as 1980, Dickinson et al. studied the welding feature monitoring approach by using electrical signals, which proved that the electrical signals can reliably reflect the welding quality [4]. Quinn et al. studied the welding quality evaluation method by extracting the distribution of short circuit transition period. The detected electrical signals are affected by electrical wiring and electromagnetic interference, which will generate strong noise interference and affect the accuracy of the welding quality monitoring [5].

Therefore, Adaptive data driven decomposition methods should be applied for denoising the detected electrical signals in welding quality monitoring. Former adaptive data driven decomposition methods can be divided into two classes: Fourier transform (FT)-based methods and non-FT-based methods. FT-based methods include the wavelet transform (WT) and the variational mode decomposition (VMD). Kumari et al. applied WT for extracting the localized defect information to fulfil the welding defect identification process [6]. Bo et al. applied variational mode decomposition (VMD) for extracting the tracking information of the electrical signals for monitoring the arc welding seam quality, the results showed that the VMD can extract accurate seam tracking feature information [7]. Non-FT-based methods contain empirical mode decomposition (EMD), ensemble empirical mode decomposition (EEMD) and local mode decomposition (LMD). Bakker et al. used EMD method to denoise the acquired linear friction-welding machine data, analysis results showed that EMD is suitable for the robust welding data processing [8]. Huang et al. applied EEMD to extract the quality feature from the noisy MIG welding electrical signals, analysis of various welding process parameters showed EEMD combined with the marginal index can effectively uncover the stability MIG welding [9]. He et al. combined LMD with support vector machine to quantitatively estimate the time–frequency energy distribution characteristics of the arc welding current signal [10].

Broadband signals with “sharp corners” such as square signals and sawtooth signals, are often applied as the input pulse electrical signals of DPMIG welding inverter power sources. However, problems will be generated when analyzing broadband DPMIG electronical signals applying former adaptive data driven decomposition methods including both FT-based methods and non-FT-based methods. For FT-based methods, they are based on frequency domain calculations by multiscale adaptive filter to achieve signal decomposition. However, De et al. studied the distortions in time domain caused by the Gibbs phenomenon, which showed that the high-frequency part of a broadband signal will decay or disappear after filtering, which may cause a series of disturbances at the break points of the decomposition results [11]. For non-FT-based methods, the original signals are separated into several IMFs by computing the envelope of the extreme points. Huang et al., proposed EMD method for adaptively dealing with nonlinear and non-stationary time series [12]. Sweeney et al. applied EEMD for canonical correlation analysis [13]. Liu et al. used LMD for the feature extraction of the vibration signals for machinery fault diagnosis [14]. These researches of non-FT-based methods showed that interpolation functions were usually used to construct envelope of the extreme points, which will make the generated IMFs to be “smooth” narrowband signals. Therefore, an unavoidable error will occur when dealing with broadband components with “sharp corners”.

As mentioned above, an unavoidable error may occur while processing broadband signals when applying former adaptive data driven decomposition methods, a method for processing broadband DPMIG electrical signals is needed for accurately evaluating the welding quality. Therefore, broadband mode decomposition (BMD) is proposed in this paper for the feature extraction of complex signals combining typical broadband signals and noise. In BMD, an associative dictionary consisting of typical broadband signals, such as square signals and sawtooth signals, and narrowband signals are established first. Then the sparsest solutions are obtained by searching in the dictionary using a heuristic optimization algorithm. The BMD algorithm uses a regulated smoothness function as the optimized objective function. The amplitude, frequency and initial phase of typical broadband signals are applied as the optimized parameters. Alatas recently proposed the artificial chemical reaction optimization algorithm (ACROA). As the ACROA method is robust and suitable for processing nonstationary signals, it is applied for solving the optimization problems in BMD [15]. The sparsest decomposition results are obtained during the optimization process. Compared to former adaptive time–frequency analysis methods, as BMD obtains the sparsest decomposition results by searching in the time-domain associative dictionary, disturbances of the generated results caused by Gibbs phenomenon in FT-based methods is avoided. Meanwhile, as the interpolation functions are not needed to be applied in BMD to construct envelope of the extreme points, the disadvantages of non-FT-based methods which will make the generated IMFs to be “smooth” narrowband signals can be avoided. Therefore, BMD can accurately extract both broadband effective components and narrowband signals from noise and achieve broadband feature extraction of effective modes when analyzing complex signals. As recent studies including Bo et al. stated in Ref. [7] and Huang et al. stated in Ref. [9] have shown the effectiveness of VMD and EEMD in dealing with the MIG welding quality data, they are chosen as representatives of FT-based methods and non-FT-based methods for the comparison with the proposed method when analyzing simulation and experimental signals in this paper.

Zheng et al. recently introduced the composite multiscale fuzzy entropy (CMFE) approach, which was based on multiscale fuzzy entropy (MFE) and has been proved to be suitable for processing non-stationary signals in engineering [16], [17]. Therefore, it is applied to construct the eigenvectors of the decomposed results generated by BMD. Afterwards, Zeng et al. recently proposed maximum margin classification based on flexible convex hulls (MMC-FCH) based on the basic theory of support vector machine (SVM) [18], [19], [20]. In SVM, the convex hull is inaccurate matching to the classification area of the original data sets when the number of data sets is small. Thus, the flexible convex hull is proposed for solving the problems existing in SVM. As have shown a good performance in the classification of nonlinear data sets, MMC-FCH is chosen to fulfill the quality monitoring process of welding at different current frequencies, duty cycles and welding speeds in this paper.

The rest of the paper is as follows. In section 2, the existing problems in analyzing broadband signal analysis when applying former adaptive data driven methods are illustrated. The detailed procedures of the BMD algorithm are given in section 3. In section 4, the basic theories of CMFE and MMC-FCH are stated. In section 5, the collected experimental data sets of aluminum alloy DPMIG welding inverter power source output electrical signals are analyzed by the proposed method. Conclusions are presented in the end.

Section snippets

Gibbs phenomenon

A Fourier series expands a signal into a sum of multiple sinusoidal signals. Taking the square signal as an example of a broadband signal, the Fourier expansion is as follows:square(t)=i=1+12i-1sin(2i-1)t

As the bandwidth of a filter is limited, the high-frequency part of a broadband signal will decay or disappear after filtering, which may cause a series of disturbances at the break points of the decomposition results. The wider the bandwidth of the applied filter is, the more accurate the

Dictionary

To separate both the broadband signal and the AM-FM narrowband effective components from noise, an associative dictionary consisting of two typical broadband signal dictionary including square signals and sawtooth signals and a narrowband signal dictionary, which was defined in Ref. [21], is established first. The decomposition results are obtained by searching in the three dictionaries. The dictionaries are as follows:Dic1=A1square(ω1n+θ1,D1)Dic2=A2sawtooch(ω2n+θ2,D2)Dic3={A3(n)cos(ω3n+θ3(n))}

A

CMFE algorithm

To alleviate the shortening of the time series in the coarse graining of MFE, CMFE applies the mean MFE of various coarse graining signals with identical scale factor as the extracted feature. If the CMFE of a given signal is larger, the mean MFE with a specific scale of different coarse graining time series of the signal is larger, which means that the given signal contains more mode information and is more complicated. To verify the stability of welding quantitatively and intuitively, the

Data collection

Experimental electrical signals are collected from the constructed aluminum alloy DPMIG welding test bed shown in Fig. 9. The experiment uses a welding robot and a MIG welding machine for bead welding. A 6061 aluminum alloy is applied as the work piece material, and the scantling is 250 mm × 200 mm × 5 mm. The argon flow is 10 L/min, and the radius of the ER5356 welding wire is 0.6 mm. The datasets are collected by Hall sensors and an acquisition box. To test the effectiveness of BMD in

Conclusions

In the quality monitoring process of DPMIG welding, the quality feature should be extracted from noisy broadband signals, as the collected electrical signals are often affected by electrical wiring and electromagnetic interference. However, former adaptive data driven decomposition methods are not suitable for separating noisy broadband signals due to Gibbs phenomenon and the calculation of extreme points. Therefore, BMD method is proposed by searching for the sparsest results in an associative

Author′s contributions

Yanfeng Peng and Kuanfang He conceived and designed the study. Zepei Li, Yanfei Liu, Liangjiang Liu and Ruiqiong Luo performed the experiments. Yanfeng Peng and Kuanfang He wrote the paper. Qinghua Lu and Qingxian Li reviewed and edited the manuscript. All authors read and approved the manuscript.

Declaration of Competing Interest

The authors declared that there is no conflict of interest.

Acknowledgments

This work was supported by the National Key Research and Development Program of China (2018YFF0212902, 2018YFB1308000), the National Natural Science Foundation of China (51805161), the Hunan Provincial Natural Science Foundation of China (2018JJ3187, 2017JJ1015) and the Guangdong Provincial Natural Science Foundation of China (2019A1515011961).

References (22)

  • S. Madhulika et al.

    Residual stress and surface properties of stainless steel welded joints induced by ultrasonic pulsed water jet peening

    Measurement

    (2018)
  • Cited by (11)

    • Monitoring of back bead penetration based on temperature sensing and deep learning

      2022, Measurement: Journal of the International Measurement Confederation
      Citation Excerpt :

      In the field of fusion welding and additive manufacturing, the traditional off-line welding quality inspection cannot fundamentally prevent the occurrence of welding defects, so the realization of on-line welding quality monitoring has become a research hotspot [1–4].

    View all citing articles on Scopus
    View full text