<p>Dual-layer optical fiber coating systems rely on magnetohydrodynamic (MHD) flow mechanisms to ensure uniform coating thickness and optimal mechanical performance. However, the coupled nonlinear interactions between electromagnetic fields and fluid dynamics pose significant modeling challenges, particularly for real-time industrial applications. Bayesian Distributed Backpropagation (BDBP), a probabilistic deep learning framework, has emerged as a powerful surrogate modeling approach for these complex Multiphysics systems. This systematic review and meta-analysis aim to consolidate current research on the application of BDBP for modeling MHD flow in optical fiber coating processes. The review evaluates BDBP’s predictive accuracy, computational efficiency, and uncertainty quantification capabilities, while identifying methodological gaps and outlining future research priorities. Following PRISMA 2020 guidelines, a structured search of Scopus, Web of Science, IEEE Xplore, and PubMed databases from 2018 to 2024 yielded 1243 unique records. After duplicate removal and two-stage screening, 58 peer-reviewed studies were included in the final synthesis, with 49 subjected to quantitative meta-analysis. Key variables extracted included neural architecture, prior distribution, RMSE, runtime, and credible interval calibration. Risk of bias was assessed using the ROBINS-I tool. Meta-analytical models were implemented using the DerSimonian–Laird estimator within a random-effects framework. BDBP significantly outperformed conventional CFD solvers in predictive accuracy, achieving a pooled RMSE of 0.12 (95% CI 0.09–0.15) compared to 0.34 (95% CI 0.28–0.40) for traditional methods (p &lt; 0.001). GRU-based architectures and Gaussian priors with variational inference yielded the best performance. However, computational latency remained a challenge, with mean runtimes of 48.2 min on CPUs versus 18.4 min on GPUs. Credible interval calibration was optimal for Gaussian priors (CI width: 0.19; coverage: 89%). High heterogeneity (I² = 78%) was observed, due to inconsistent model configurations and prior parameterizations. BDBP demonstrates superior modeling capabilities for MHD flow dynamics and holds significant potential for integration into real-time control frameworks in fiber manufacturing. However, to facilitate broader industrial adoption, future research should prioritize standardized benchmark datasets, hardware-accelerated deployment, and transparent reporting of Bayesian priors and uncertainty metrics.</p>