A Hidden Markov Framework for Physically Interpretable Arc Stability Dynamics in Welding Systems

Hidir Selcuk Nogay · 2026

Read A Hidden Markov Framework for Physically Interpretable Arc Stability Dynamics in Welding Systems by Hidir Selcuk Nogay (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-21839.txt.

Introduction

Nogay models electric arc welding stability with a Hidden Markov framework built from physically interpretable spectral features. The welding current signal passes through a Short-Time Fourier Transform, yielding energy, entropy, and centroid descriptors as observations. The model infers three dominant regimes: transient, stable, and extinction. Stable arcs show rising spectral energy and falling entropy, meaning less variability. Despite partial feature overlap, the inferred state sequence stays temporally coherent, with high persistence and low transition rates.

Catalog record

CREATORHidir Selcuk Nogay
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE43616
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.21839

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A Hidden Markov Framework for Physically Interpretable Arc Stability Dynamics in Welding Systems Hidir Selcuk Nogay Abstract Electric arc welding (EAW) exhibits strongly nonstationary and temporally evolving behavior, making reliable assessment of arc stability difficult using conventional frame-based approaches. In this study, arc dynamics are modeled as a sequence of latent operational regimes within a probabilistic state-space framework. The welding current signal is transformed into a time-frequency domain using Short-Time Fourier Transform (STFT), and a set of physically meaningful spectral descriptors, including energy, entropy, and centroid, is extracted to construct the observation sequence. A Hidden Markov Model (HMM) is employed to capture temporal dependencies and estimate the evolution of arc states. The analysis reveals three dominant regimes, transient, stable, and extinction, with a clear monotonic increase in spectral energy and a corresponding decrease in entropy, indicating reduced variability under stable conditions. Despite partial overlap in the feature space, the inferred state sequence exhibits strong temporal coherence, supported by high state persistence and low transition rates. These findings highlight the limitations of static classification and emphasize the importance of temporal modeling. The proposed framework provides an interpretable and physically consistent representation of arc behavior, enabling more realistic monitoring and analysis of stability dynamics in welding processes. I Introduction Electric arc welding (EAW) is a complex industrial process governed by nonlinear interactions between electromagnetic fields, thermal dynamics, and plasma behavior [ 1 ] . The stability of the electric arc directly determines weld quality, energy transfer efficiency, and structural integrity of the final product [ 2 ] . However, the arc is inherently non-stationary and stochastic, exhibiting rapid temporal fluctuations due to plasma turbulence, impedance variations, and external disturbances. These characteristics make reliable monitoring and control of arc stability a challenging task, particularly in real-time industrial environments [ 3 ] . Traditional approaches to arc monitoring rely on signal analysis techniques that extract descriptive features from the welding current [ 4 ] . Time-frequency representations have shown strong capability in revealing localized spectral structures associated with underlying physical dynamics [ 5 ] . Prior studies demonstrate that spectral characteristics derived from current signals can serve as informative indicators of system behavior across different domains, including power systems and biomechanical signal analysis [ 7 , 6 , 8 ] . In welding applications, these representations provide a physically meaningful description of energy redistribution within the arc, where variations in spectral energy reflect changes in plasma conductivity, arc length, and electromagnetic coupling [ 9 , 10 ] . Building on this perspective, existing data-driven methods utilize time-frequency features with machine learning models to classify arc conditions into regimes such as transient, stable, and extinction states [ 4 , 11 , 12 ] . However, these approaches rely on frame-level evaluation and treat observations independently, neglecting temporal dependencies. As demonstrated in Section IV, feature distributions exhibit partial overlap, indicating that static classifiers are inherently limited under realistic conditions. Consequently, such models provide instantaneous labeling rather than a continuous representation of arc behavior. Another limitation concerns interpretability. Although descriptors such as energy and spectral entropy are widely used, their connection to physical arc dynamics is often not explicitly addressed. Similar challenges have been observed in related welding monitoring studies, where time-frequency and spectral features are employed for automated quality assessment, yet their direct linkage to underlying physical arc behavior remains limited [ 13 ] . In practice, increasing energy combined with decreasing entropy reflects a more structured and constrained arc state, whereas higher entropy indicates increased stochastic fluctuations. Without embedding such physical interpretation, purely data-driven models remain limited in practical applicability [ 4 , 8 ] . Arc stability is inherently a temporally evolving process characterized by gradual transitions between different operating regimes, as also reflected in experimentally derived arc stability features under varying welding conditions [ 14 ] . This observation motivates the use of modeling frameworks that explicitly capture sequential dependencies over time. In this context, HMMs provide a natural probabilistic formulation, where the observed time–frequency features extracted via STFT are interpreted as emissions generated by an underlying sequence of latent states, a perspective consistent with recent process monitoring approaches that incorporate temporal dependencies and hidden state inference in industrial systems [ 15 , 16 ] . In this study, arc dynamics are modeled as a stochastic process evolving over a finite set of hidden states. The welding current signal is transformed into a time-frequency domain using STFT, and a compact set of spectral descriptors is extracted to represent energy distribution and variability. These descriptors are treated as a sequential observation process within an HMM framework, enabling estimation of state transitions and reconstruction of arc evolution over time. Unlike conventional classification approaches, this formulation captures temporal persistence and transition dynamics, providing a more realistic representation of the welding process. The main contributions of this work can be summarized as follows: • A probabilistic state-space formulation of arc stability is developed, modeling the welding process as a sequence of hidden regimes. • A physically-informed observation model is constructed by linking STFT-based spectral features to underlying arc physics. • A temporal inference framework based on Hidden Markov Models is proposed to estimate state transitions and persistence behavior. • The study shifts arc analysis from static classification toward interpretable temporal modeling. Accordingly, the contribution of this work lies in reformulating arc stability analysis as a temporal probabilistic inference problem grounded in both signal characteristics and physical interpretation. The remainder of this paper is organized as follows. Section II presents the data acquisition and preprocessing steps. Section III introduces the Hidden Markov modeling framework. Section IV provides experimental evaluation and analysis. Section V concludes the study. II Data Acquisition and Sequential Signal Representation The primary current signal acquired from an Electric Arc Welding Machine (EAWM) is modeled as a stochastic and non-stationary process in the Hilbert space \mathcal{L}^{2}(\mathbb{R}) . The measured signal is expressed as: i(t)=s(t)+n(t) (1) As defined in ( 1 ), the observed signal consists of the underlying arc dynamics s(t) and an additive noise component n(t) . This formulation provides a physically meaningful interpretation in which the arc behavior is embedded within a stochastic observation model, explicitly accounting for measurement uncertainty and environmental disturbances. To preserve temporal dependencies, the continuous-time signal is segmented into overlapping windows, forming a sequential representation: \{i_{t}\}_{t=1}^{T},\quad i_{t}\in\mathcal{L}^{2}(\mathbb{R}) (2) As expressed in ( 2 ), each segment i_{t} represents a localized observation of the arc process. This transformation enables the conversion of the raw signal into a structured temporal sequence suitable for probabilistic modeling. A nonlinear feature mapping is then defined as: \mathbf{v}_{t}=\Phi(i_{t}),\quad\mathbf{v}_{t}\in\mathbb{R}^{d} (3) As indicated in ( 3 ), the mapping \Phi extracts informative descriptors from each segment, providing a compact representation that links physical signal characteristics with statistical inference. The overall acquisition and transformation process is illustrated in Fig. 1 , which summarizes the complete signal-to-inference pipeline of the proposed framework. The diagram begins with the raw welding current signal i(t) , representing the physical arc behavior, which is inherently non-stationary and influenced by plasma dynamics and external disturbances. Fig. 1: Proposed STFT–HMM framework for temporal arc stability modeling, including signal acquisition, feature extraction, observation sequence construction, and probabilistic inference. In the first stage, the signal is segmented into

Read the complete text