From Generative Bits to Industrial Atoms: A Sensorium–Simulation–Actuation Framework for Physics-Informed AI in Cyber-Physical Systems

The rapid proliferation of generative artificial intelligence (AI) systems has produced substantial advances in natural language processing, creative content generation, and knowledge synthesis. However, the architectural assumptions underlying these systems - notably, tolerance for probabilistic error and hallucinated outputs - render them fundamentally unsuitable for direct deployment within industrial cyber-physical systems (CPS), where operational failures carry consequences ranging from economic loss to physical harm. This paper identifies a critical gap in the literature: the absence of an integrated, deployment-ready framework that reconciles AI's probabilistic nature with the deterministic reliability demands of brownfield industrial infrastructure. To address this gap, the paper proposes the Sensorium–Simulation–Actuation (SSA) framework, a three-layer architectural model for reliable industrial AI deployment. The Sensorium layer addresses data acquisition from heterogeneous industrial sensors and legacy protocols within electromagnetically hostile environments. The Simulation layer centers on Physics-Informed Neural Networks (PINNs) and bidirectional Digital Twins, which constrain AI inference within thermodynamic and physical law boundaries, eliminating physically impossible outputs. The Actuation layer translates validated intelligence into physical interventions including predictive maintenance, generative engineering, and autonomous process control. Drawing on a practitioner-informed synthesis of industry case studies, systems analysis, and comparative framework development, the paper further examines the organizational and cultural barriers - collectively termed 'Pilot Purgatory' - that prevent industrial AI from scaling beyond proof-of-concept deployments. Key findings indicate that IT/OT convergence, physics-constrained modeling, and augmented human-machine collaboration are necessary conditions for industrial AI viability. The paper contributes a structured, actionable framework for scholars, practitioners, and policymakers engaged in the strategic deployment of AI across manufacturing, energy, supply chain, and critical infrastructure domains.