Stability Evaluation of Pulsed MIG Welding Based on Current Sample Entropy

Literature Overview

This 2015 paper published in the Welding Journal by Xie Huangsheng, Fu Zhihe, Wang Yuexin, Xu Min, and Xue Jiaxiang introduces a novel approach to evaluating the stability of pulsed metal inert gas (MIG) welding using current sample entropy as the quantitative metric. The work was supported by the Guangdong Provincial Industry-University-Research Collaboration Program (2013B090600098) and the Longyan Advanced Mechanical Design and Manufacturing Technology Public Service Platform (2012LY01), reflecting the strong industrial-academic partnership driving research in China's welding community. The authors address a long-standing challenge in welding process monitoring: the lack of a simple, robust, and real-time-capable index that can characterize the stochastic behavior of welding current signals and translate that behavior into a meaningful stability assessment.

Core Technical Approach

The fundamental idea behind the paper is that the welding current waveform in pulsed MIG welding carries rich information about the droplet transfer process, arc stability, and overall process quality. Rather than relying on conventional time-domain or frequency-domain statistics (mean, standard deviation, peak frequency), the authors adopt sample entropy (SampEn) as a nonlinear complexity measure. Sample entropy quantifies the regularity and predictability of a time series — a highly stable welding process produces a current signal with low complexity (low SampEn), while an unstable process yields high complexity (high SampEn).

The key methodological steps are as follows:

1. Acquire the welding current signal at a sufficiently high sampling rate (typically 10 kHz or above) to capture the full dynamic range of the pulsed waveform.
2. Normalize the current signal to eliminate the influence of absolute current magnitude.
3. Apply the sample entropy algorithm with carefully selected parameters: embedding dimension m (typically 2 or 3), tolerance r (typically 0.1 to 0.25 times the standard deviation of the normalized signal), and time series length N.
4. Compute SampEn for different welding parameter combinations and correlate the results with visual and metallurgical assessments of weld quality.

Why Sample Entropy Over Other Measures

The authors justify the choice of sample entropy over approximate entropy (ApEn) and spectral entropy on several grounds. First, SampEn does not include self-matches in its calculation, reducing the inherent bias of ApEn. Second, SampEn is more robust to noise and less dependent on data length, making it suitable for real-time or near-real-time monitoring where data segments are limited. Third, unlike spectral entropy, SampEn captures nonlinear dynamics in the current waveform that arise from the complex interaction between the arc plasma, droplet dynamics, and melt pool behavior — features that a purely linear frequency analysis would miss.

Process Parameters and Experimental Design

The experimental matrix typically covers the full range of pulsed MIG parameters relevant to mild steel and low-alloy steel welding. The following table summarizes the typical parameter ranges explored:

The study systematically varied each parameter while holding others constant, recording the current waveform for each condition. For each recording, the SampEn value was computed and compared against the visual quality of the deposited bead and the presence of defects such as spatter, undercut, and porosity.

Key Findings and Stability Mapping

The central finding is that there exists a clear monotonic relationship between the computed SampEn value and the qualitative stability of the pulsed MIG process. Under optimal pulse parameter settings — where the pulse current is sufficient to detach droplets in a controlled manner and the background current maintains a stable arc without excessive spatter — the SampEn value drops to a distinct minimum. As parameters drift away from this optimum, SampEn increases, indicating increased irregularity in droplet transfer and arc behavior.

This relationship enables the construction of a "stability map" in the parameter space, where regions of low SampEn correspond to stable, high-quality welding conditions. Such a map is directly useful for process optimization in production environments, as it provides a quantitative criterion for parameter selection that goes beyond the qualitative trial-and-error approach traditionally used in shop-floor settings.

Integration with Engineering Practice

From a practical standpoint, the sample entropy approach has several implications for cladding and overlay operations. In weld overlay applications, where the stability of the process directly affects the dilution rate, microstructure of the overlay layer, and bond strength between the cladding and base metal, a quantitative stability metric is invaluable. For example:

• In nickel-based alloy overlay on carbon steel pressure vessels, excessive process instability can lead to increased dilution, which compromises the corrosion resistance of the overlay layer.
• In stainless steel cladding on hydrogenation reactor internals, arc instability can cause porosity and lack of fusion at the cladding-base metal interface.
• The SampEn-based stability evaluation can be integrated into a process monitoring system that alerts operators when parameters drift into unstable regions, enabling corrective action before defects accumulate.

The method is also compatible with the FMEA (Failure Mode and Effects Analysis) framework. By mapping the relationship between SampEn values and known failure modes (spatter, porosity, undercut), engineers can assign severity and occurrence ratings to each parameter deviation and prioritize controls accordingly.

Key Questions and Reflections

Several questions merit further investigation. First, the sensitivity of SampEn to parameter selection (m, r, N) must be thoroughly characterized for different welding configurations, as inappropriate parameter choices can lead to misleading stability assessments. Second, the method's performance under different shielding gas compositions (pure Ar, Ar/CO₂, Ar/He) and different electrode materials (ER70S-6, ER80S-D2, ER308L, ERNiCrMo-3) needs to be validated. Third, the transition from laboratory-based signal analysis to in-situ real-time monitoring on a production welding cell requires addressing signal acquisition hardware, computational latency, and operator interface design.

The paper also raises a broader philosophical point: welding process stability is not a binary property but a continuous spectrum. The sample entropy approach acknowledges this continuum and provides a graduated measure, which is more informative than a simple "stable/unstable" classification.

Study Insights and Implications

The introduction of sample entropy as a welding stability metric represents a meaningful advancement in welding process monitoring. It bridges the gap between raw electrical signal data and actionable process knowledge, offering a computationally efficient and physically interpretable index. For engineers involved in cladding and bimetal product manufacturing, this method provides a tool for optimizing overlay parameters, reducing defect rates, and improving the consistency of the final product. The approach is particularly valuable in automated and semi-automated welding cells where real-time feedback is essential for maintaining quality.

However, the true value of this method lies not in its novelty but in its practicality. The sample entropy calculation is straightforward, requires no specialized hardware beyond a standard current transducer and data acquisition system, and can be implemented in real time with modest computing resources. This practicality, combined with its demonstrated sensitivity to process instability, makes it a worthwhile addition to the quality assurance toolkit of any welding shop engaged in critical overlay operations.