University laboratories, as critical hubs for scientific innovation, face escalating safety management challenges. Traditional approaches, reliant on periodic manual inspections and static compliance checklists, are increasingly inadequate and suffer from inherent deficiencies: static risk perception, fragmented management elements, and reactive response strategies. This misalignment between safety management and the dynamic nature of experimental activities makes it difficult to adequately address the coupled and complex risks in modern research, transforming safety from an enabler of research into a heavy operational and administrative burden. Therefore, this study aims to address these deep-seated limitations by proposing a systematic, integrated framework to drive a fundamental paradigm shift in laboratory safety management from static, passive compliance to proactive, dynamic, and intelligent governance, thereby unifying safety assurance with scientific development.

This study develops a comprehensive solution comprising a novel theoretical model, a supporting technological architecture, and a defined operational mechanism. First, the core innovation is the WSR-T dynamic safety management model, which incorporates a Time (T) dimension divided into three sequential phases: pre-experiment (T1: prevention and preparation), during experiment (T2: monitoring and execution), and post-experiment (T3: restoration and learning). These phases are integrated with the Wuli (physical), Shili (procedural), and Renli (human) dimensions of the WSR systems methodology, creating a dynamic management matrix that reframes safety as continuous control over the entire experimental life cycle. Second, to support this model, a multi-agent collaborative digital twin-enabled cyber-physical system (MA-DT-CPS) architecture is designed. This model is based on a high-fidelity digital twin that integrates five core computational models: a geometric model for spatial semantics, a physical model for real-time monitoring and simulation, a rule model for formalizing regulations and procedures, a behavior model for quantifying human actions and states, and a process model for creating intelligent digital threads for the life cycle of each experiment. Within this digital framework, a collaborative multi-agent system operates, featuring specialized agents for situational awareness, dynamic risk assessment, compliance execution, emergency decision-making, and system learning & optimization. This system is designed to operate in a three-tier hybrid-intelligence mode: full automation for routine tasks, suggestion mode for uncertain scenarios, and co-creation for novel situations. Finally, the study outlines the “perception–mapping–analysis–decision–execution–learning” event-driven closed-loop operational mechanism, detailing how the architecture implements the logic of the WSR-T model for dynamic, intelligent control.

The study produces a holistic and actionable framework. Theoretically, the WSR-T model provides a novel, structured perspective, making complex laboratory safety events analyzable as specific spatiotemporal couplings of W, S, and R elements, thereby moving beyond mere checklist compliance. Technologically, the MA-DT-CPS architecture translates this theoretical model into a concrete, actionable implementation path. The key results include the following: (1) a systematic methodology that recontextualizes safety as a process of dynamic control; (2) a sophisticated technological blueprint enabling high-fidelity digital representation, autonomous agent collaboration, and human-machine synergy; and (3) a well-defined operational mechanism that transforms management from a periodic, plan-driven activity to an event-driven, intelligent closed-loop process. This integrated “theoretical model-technical architecture-operational mechanism” framework systematically addresses the root causes of static, fragmented, and passive management, providing a clear pathway for the intelligent transformation of laboratory safety systems. The framework enables proactive risk assessment, virtual strategy testing through the digital twin sandbox, and a system capable of self-learning that continuously improves based on operational feedback.

This study provides a systematic, end-to-end solution designed to overcome the fundamental challenges in modern university laboratory safety management. By integrating the time dimension into the WSR methodology, the WSR-T model establishes a robust theoretical foundation for dynamic safety governance throughout the life cycle. The MA-DT-CPS architecture provides the necessary technological enablers, fusing digital twin and multiagent system concepts to create a platform for realizing the model logic. Together, they form a coherent framework that paves the way for the development of intelligent management systems that seamlessly embed safety requirements into the entire scientific workflow. This paradigm aims to unify the goals of safety assurance and scientific innovation at a higher level. Future work should focus on developing prototypes and conducting long-term empirical studies in real laboratory settings to validate, optimize, and iteratively improve the proposed system, with particular attention to the quantitative modeling of complex human factors and ensuring cost-effectiveness for widespread adoption.