Optical Fiber Coating Using Bayesian Distributed Back Propagation: Historical Development, Current State and Future Perspective

Abstract

Modeling magnetohydrodynamic (MHD) flows in double-layer optical fibre coatings poses significant computational challenges due to their nonlinear and anisotropic nature. Traditional computational fluid dynamics (CFD) techniques often struggle with scalability and precision in such high-dimensional systems. This paper presents a systematic review of Bayesian distributed backpropagation, highlighting its integration with neural networks to address uncertainty quantification and improve model generalization. The study reformulates key physical laws—Navier-Stokes with Lorentz force and Maxwell’s equations—within machine learning frameworks optimized via distributed Bayesian learning. Comparative analysis demonstrates that Bayesian methods outperform conventional backpropagation and optimization algorithms in accuracy and robustness, particularly under complex electromagnetic-fluid interactions. Nevertheless, high computational costs and convergence time remain major limitations, especially in real-time applications. The review identifies key breakthroughs in uncertainty modeling and intelligent neuro-structure optimization, offering practical relevance for optical fibre manufacturing. Future directions include hybrid Bayesian methods and scalable distributed learning strategies to address nonlinear, anisotropic systems more effectively and support broader industrial deployment of MHD flow simulation technologies.