To address the challenges of complex data environments, low accuracy of single-sensor detection, and limited precision in traditional laboratory safety systems, this study presented a multi-sensor fusion early warning model based on an ISBOA and algorithm KELM. First, the KELM framework was employed to integrate heterogeneous sensor data and construct the warning model, where a regularization term was introduced to alleviate overfitting. Then, the improved ISBOA adaptively optimized the regularization coefficient C and kernel parameterσ of the KELM, thereby enhancing parameter robustness and diagnostic accuracy. Finally, simulation and experimental analyses were conducted using both synthetic and real laboratory datasets, and the proposed ISBOA-KELM model was compared with the unimproved Secretary Bird Optimization Algorithm (SBOA), Particle Swarm Optimization (PSO), and Gray Wolf Optimization (GWO) algorithms. The experimental results show that the ISBOA-KELM model improved accuracy by 4%, 3%, and 2%, respectively, compared with the other three models. In four representative laboratory safety scenarios, including fire and gas leakage, the detection accuracy exceeds 96% with the false negative rate below 6%, which significantly improves the reliability and robustness of safety accident early warning.

In order to address the constraints of limited corpus resources, restricted input capacity, and data privacy in applying LLMs to the field of safety engineering, a localized accident question-answering model integrating the DeepSeek with a RAG mechanism was constructed to enable intelligent parsing and knowledge services for complex texts, thereby supporting safety management decision-making. A semantic-feature corpus was built based on accident investigation reports and laws and regulations released by government emergency management systems, and technologies such as PaddleOCR, LayoutLMv3, and YOLOv8 were incorporated to accomplish document structure reconstruction and semantic modeling. The model encompassed four stages—document parsing, semantic alignment, knowledge-base construction, and hybrid retrieval—and was designed with capabilities for causal-chain extraction, regulation matching, and semantic mapping. The results indicated that, compared with the Deepseek-r1:32b model without the RAG mechanism, the enhanced model achieved improvements of 7.7% in automated scoring and 17.6% in human evaluation, and the response-speed and stability metrics presented higher numerical performance than those of the baseline model. The model performance was still influenced by the local parameter scale and the knowledge-updating mechanism, yet the experimental findings demonstrate that it is capable of fulfilling the intended functions in the present study.

This study aims to systematically explore the temporal and spatial patterns, causal mechanisms, and prevention and control strategies of dust explosion accidents in China. Knowledge graph technology combined with statistical analysis was used to analyze the relevant literature and accident cases of dust explosion accidents in China from 1990 to 2024. A knowledge graph was constructed using CiteSpace visualization software to analyze the spatiotemporal distribution characteristics and the accident causality chain of dust explosion accidents. A four-layer prevention and control system framework of "hazard source-facility-management-emergency" was constructed, and its effectiveness was verified by typical cases. It is found that the risk of dust explosion shows the characteristics of "type differentiation" and "geographical migration", with frequent occurrence of dust, high fatality of coal dust and high secondary injury of grain dust explosions. The accident hotspots are showing a trend of shifting from the eastern to the central and western regions. Inadequate dust cleaning and defective dust removal systems are central causes, while operational violations and use of non-explosion-proof equipment are main triggering conditions, and the causal structure varies significantly with dust types. The implementation of differential and precise prevention and control strategies is the key means to improve the prevention and control efficiency of dust explosion accidents.

In-service pipelines are subject to complex stresses, and their performance degradation over time constitutes a dynamic, time-varying stochastic process. To address the challenge that traditional deterministic functions struggle to accurately capture its inherent randomness, a dynamic analysis method for failure probability of in-service pipelines based on a dual stochastic process was proposed. The degradation of pipeline performance was simulated using an inverse Gaussian stochastic process, while the variation of internal pressure loads within the pipeline was described by an equal-interval stationary binomial rectangular wave process probability model. A dual stochastic process probability model for pipeline bearing capacity and internal pressure load was then constructed. Based on statistical parameters and performance degradation data from a specific pipeline's service period, inverse Gaussian distribution was used to fit the performance degradation models at six distinct time points, enabling dynamic failure probability prediction. The results show that the pipeline's service life is predicted to be 16 and 14 years using degradation data from 2 and 4 years, respectively. When utilizing degradation data from 6, 8, and 10 years, the predicted service lives are 12, 11, and 10 years, respectively. Sensitivity analysis indicates that wall thickness, yield strength, pipe diameter, and operating pressure have the most significant impacts on the pipeline's failure probability, followed by the initial depth of defects. In contrast, the initial length of defects, depth corrosion rate, and length corrosion rate have relatively minor effects.

A safety culture reshaping model was studied to enable enterprises to carry out safety culture reshaping in a more effective manner and enhance practical outcomes. Typical domestic and international safety culture models were systematically compared and analyzed to summarize their characteristics in terms of element design and logical structure. Based on actual needs of enterprise safety management scenarios, elements adaptation and frameworks integration were carried out, and an excellent safety culture reshaping model. This model incorporated six core elements—leadership, safety philosophy, risk control, communication, systems, and behavior, and deeply integrated within the "being-knowing-doing unity" logical framework. The results show that "excellence" is reflected in ambitious goals, extreme execution, and measurable mechanisms, while the core of "reshaping" lies in "retaining strengths and correcting weaknesses", which involves transforming unscientific safety concepts, revising imperfect systems, improving inefficient communication, and rectifying misleading leadership behaviors. The application of the excellent safety culture reshaping model enables all employees to first establish a solid foundation in values and safety beliefs, then reach a consensus in safety cognition, and finally standardize their safety behaviors, forming a shared behavioral pattern, thereby enhancing the effectiveness of safety culture reshaping.

To enhance China's work safety governance level, policy texts on work safety issued by the central government and 30 provincial-level governments from 2014 to 2023, as well as safety accident data from the same period, were used as the research sample. By integrating the Levenshtein distance algorithm, the Jaccard similarity algorithm, and other similarity measures, a "theme-content" two-stage policy synergy analysis model was proposed. Central-local policy synergy degree was quantitatively measured through a weighted evaluation of policy theme matching degree and content similarity. A complex synergy network was then constructed. Using social network analysis and modularity algorithms, the structural characteristics of the central-local safety policy synergy network were analyzed in depth, and its spatiotemporal evolution was revealed. Based on panel data, the impact of central-local synergy outcomes of safety policies on accident incidence were further investigated. The results show that China's central-local safety policy synergy network is exhibiting an increasingly integrated development trend over time. The responsiveness of provincial governments to central policies is continuously improving, evolving from early regional differentiation toward comprehensive nationwide coordination, and an initial "central planning-local response" national work safety policy system has been initially established. Moreover, the central-local policy synergy degree is significantly negatively correlated with the frequency of safety accidents, and improving the level of coordination can effectively reduce the risk of work safety accidents.

In order to elucidate the essential differences among safety-I, safety-II, and safety-III, to clarify the operational mechanisms of safety science paradigms, and to promote the sustained development of safety science as a scientific discipline, this study adopted literature review and comparative analysis methods. The theoretical distinctions among the three from epistemological and methodological dimensions were analyzed. Then the construction and transformation of the safety science paradigm from the perspective of philosophy of science were discussed, and their paradigm positions were clarified. The results indicate that: Safety-I represents traditional accident causation theories or models that emphasize causality. Safety-II is a resilience theory that studies safety issues from a positive perspective. Safety-III is a systemic accident model grounded in systems theory and cybernetics. The safety science paradigm comprises one entity with four aspects. These aspects include the paradigm-free stage, paradigm establishment stage, normal science research stage, paradigm crisis stage, and paradigm shift stage. Currently, the safety science paradigm is dominated by accident causation theory and remains in the normal science research stage. Although there are signs of a paradigm crisis, it has not yet entered the paradigm shift period. Safety-I and Safety-III are research contents within the accident causation theory paradigm. However, Safety-II reflects changes in both beliefs and research perspectives. It can be seen as a new research direction in safety science, but it has not yet become a new paradigm. The future development of the safety science paradigm has three possible forms. It requires continuous practice and exploration led by new technologies.

To address the structural safety risks in the workforce arising from the increasing proportion of older miners in the mining industry, this study investigated the causal mechanisms underlying unsafe behaviors among older miners. Drawing on the IMB model, a SEM was developed to examine the causal pathways of miners' unsafe behaviors, and multi-group analysis was employed to compare the path coefficient differences between older and younger miners. The results indicate that knowledge and experience factors such as educational attainment and safety knowledge, as well as psychological characteristic factors including safety attitudes, accident experience, mental health, job burnout, and risk attitudes, have significant effects on miners' unsafe behaviors. Notably, work ability mediates both the pathway from knowledge and experience to unsafe behaviors and the pathway from psychological characteristics to unsafe behaviors. In addition, risk attitudes, mental health, job burnout, educational attainment, and safety knowledge exert stronger influences on unsafe behaviors among older miners compared to their younger counterparts.

In order to effectively predict disasters and evaluate the effectiveness of intervention measures, a multi-disaster coupling evolution intelligent deduction method was proposed. The disaster process, multi-disaster coupling, as well as the advantages and difficulties of evolutionary intelligent deduction, were discussed. The disaster evolution process was described, and an intelligent deduction method was established. Taking the mining process of an open-pit mine as an example, the proposed method was applied for analysis, and its effectiveness was verified. The results show that the multi-disaster coupling process is complex and changeable, with characteristics such as uncertainty, network structure effect and spatio-temporal distribution difference. The system fault evolution theory can provide support for the multi-disaster coupling evolution intelligent deduction from the perspectives of conceptual description, topological structure and mathematical analysis. The evolutionary intelligent deduction method is established, and its steps and mathematical model are provided. This method can qualitatively and quantitatively deduce the multi-disaster coupling evolution process, discover hidden disaster processes and generate emergent knowledge.

To evaluate the application performance of UAVs and rescue robots in earthquake disaster data acquisition and three-dimensional (3D) modeling, an air-ground collaborative emergency response field experiment was conducted at the Beichuan earthquake site. In the experiment, fixed-wing UAVs were deployed to acquire aerial images covering the entire earthquake-affected area, from which high-resolution top-view images were generated, while rotary-wing UAVs were used to supplement data collection in key local regions. Meanwhile, the NuBot rescue robot was deployed to enter damaged buildings and collect indoor disaster images and 3D point cloud data through real-time video transmission and lidar scanning, and 3D reconstructions of interior building structures and artificially constructed collapse scenes were produced. Based on the acquired 3D geographic information, a human-machine interaction platform was developed using virtual reality technology to support visualization and interactive analysis of 3D scenes.The experimental results indicate that multi-scale image data covering both the overall earthquake-affected area and local regions were obtained and corresponding high-precision real-scene models were generated. Indoor images and 3D point cloud data were collected in complex environments, forming 3D models of interior building structures. The 3D scene data were visualized in the virtual reality system and applied to human-machine interaction analysis.