In order to improve the efficiency of facepiece-wearing detection for tunnel operation workers, a feature fusion-based facepiece detection model was proposed. First, high quality query images were selected, and an image gallery was established. An image retrieval method was adopted to obtain samples and measure the similarity between query and gallery images, thereby iteratively expanding the dataset scale. Then, Histogram of Oriented Gradients (HOG) and Fisher features were extracted from the images. The Ant Lion Optimizer (ALO) was introduced to compute the optimal weight combination for the two types of features, which were subsequently fused. Finally, based on the fused features, a SVM was utilized to train a facepiece detection model, and experimental evaluations were conducted on the self-constructed dataset. The results indicate that the proposed model effectively accomplishes the task of facepiece-wearing detection in tunnel operation scenarios. Feature fusion enhances the image description and improves the detection accuracy of the model. Compared to using only HOG features or Fisher features, the accuracy is increased by 6% and 14%, respectively. The model meets the accuracy requirements for facepiece-wearing detection of workers in tunnel construction environments.

In order to clarify the potential research value and application prospects of LLMs and KG technologies in the field of construction safety, the existing problems in the knowledge-driven digital transformation of this domain were comprehensively analyzed, and the current state of technological development regarding knowledge graphs and large language models within it was reviewed. First, relevant literature was retrieved from the China National Knowledge Infrastructure (CNKI) and Web of Science(WoS) databases to define the scope of analysis. Then, the existing problems and challenges in the digital transformation of construction safety were analyzed, and the necessity of introducing KG and LLM was elaborated. Subsequently, the technological development of KG and LLM, along with the current status of their application research in construction safety, was briefly described. Preliminary achievements in the integrated application of KG and LLM within the construction safety domain were explored. Finally, the shortcomings of existing research were summarized, and future research directions are outlined. The results indicate that LLMs and KG technologies have demonstrated significant application potential in various scenarios within construction safety, including knowledge management, risk identification, and intelligent decision-making, highlighting considerable prospects for future implementation. However, current research still faces challenges such as a lack of real-time capabilities and insufficient integration. Future efforts should focus on establishing comprehensive and precise construction safety KG and exploring novel approaches for the integrated application of LLM and KG.

To achieve safe and efficient work in job shops, this paper proposed a BP-DRFJSP model that considers workers' boredom perception. First, the impact of boredom arising from repetitive tasks on work efficiency and work safety accident risks was analyzed. A mathematical model was developed to quantify this impact, and a scheduling model was established with the optimization objectives of minimizing makespan and production safety risks. Then, a MO-FTTA was designed to solve the model. Finally, comparative experiments and a case study involving MO-FTTA and three other common algorithms were conducted to verify the performance and feasibility of the proposed approach. The results show that when solving BP-DRFJSP model, MO-FTTA achieves better hypervolume (HV) and inverted generational distance (IGD) in 95% of the test instances, with C-metric values close to 1. Moreover, it yields superior objective function values in 90% of the instances compared to other algorithms. In the case study, MO-FTTA reduces the makespan to 79.635 minutes and lowers work safety risk by 1.25%, demonstrating its effectiveness in enhancing production efficiency while mitigating safety risks, thereby supporting safe and efficient workshop operations.

To control the simulated smoke flow field (used as a substitute for real fire smoke in the airworthiness verification experiments for aircraft cargo compartment smoke detection), this paper established a closed-loop flow control experimental platform for simulated smoke based on a full-scale aircraft cargo compartment mock-up. Control law design was conducted using MFAC theory. The closed-loop flow control effects of simulated smoke were verified with real fire smoke mass concentration as the control target. Additionally, the influence of control law parameters on control effect was explored. The experimental results demonstrate that MFAC method can effectively overcome the adverse effects of unmodeled dynamics in turbulent flow fields on closed-loop control, and achieve the approximation of the non-stationary flow fields between simulated smoke and real fire smoke. Compared with the PID control method, the control performance can be improved by more than 20%.

Non-caking coal was often used as gasification and power generation coal, and the heat generation was low due to the coal formation stage, and the heating rate had a greater influence on the heat release of coal-oxygen composite at different stages. The heat release process of Non-caking coal at different heating rates (10, 20, 30, 40 ℃/min) up to 1 000 ℃ was tested by synchronous thermal analyzer, and the changes of heat release and ignition heat in the coal-oxygen composite process were analyzed. The apparent activation energy was solved by the Starink method, and the effect of apparent activation energy on the heat release was clarified. The coal-oxygen composite stage with the greatest influence of heating rate was analyzed by the gray correlation method. The gray correlation method was used to analyze the coal-oxygen composite stage that was most affected by the heating rate. The results show that the heat flow changes of Non-caking coal can be divided into five stages: evaporation heat absorption, oxidation exothermic, decomposition heat storage, combustion exothermic and combustion exhaustion. With the increase of heating rate, the heat flow density gradually shifts to the high temperature area and the energy redistribution phenomenon occurs, and the heat release per unit mass of Non-caking coal increases by 19.2%. The percentage of exothermic heat in the exothermic stage of combustion increases from 70.6% to 75.1%, and the heat of ignition is 360~380 J/g. The apparent activation energy decreases gradually from 0 to 0.90, and increased abruptly at 0.95, with a peak value of 143.44 kJ/mol. The exothermic stage of combustion and exhaustion stage were most affected by the warming rate, and the evaporation and heat absorption stage was least affected by the warming rate.

To address the safety risks and geological hazards encountered during urban subway shield tunneling in soft-hard uneven strata, the applicability and identification characteristics of the surface transient electromagnetic method for advanced detection in such formations were investigated. Taking a metro section in Qingdao as an example, and considering the generally shallow burial depth of urban subways (typically 30-50 m), a numerical model of upper-soft and lower-hard strata was established using COMSOL finite element software. Through simulation of the transient electromagnetic response process, electric field response results and anomalous characteristics at different measurement points were obtained. Combined with field-measured data, the numerical simulation results were compared and validated, and the criteria for identifying soft-hard uneven anomalies in the strata were summarized. The results indicate that soft-hard uneven strata induce significant variations in the electric field response. The response curve at the measurement point directly above the anomalous body exhibits the most drastic changes and the fastest decay rate, demonstrating that the surface transient electromagnetic method can be utilized to identify anomalies in soft-hard uneven strata. The rate of electric field attenuation is jointly influenced by the properties of the medium and the distance from the anomalous zone, manifesting as a "trough phenomenon", with faster attenuation near the anomaly and slower attenuation farther away. In the apparent resistivity contour maps, the tunnel-crossing areas exhibit intense small-scale contour fluctuations, with soft rock showing low-resistivity anomalies and hard rock showing high-resistivity anomalies. Resistivity often transitions from low to high or from high to low, reflecting the heterogeneous characteristics of the strata.

To enhance the safety and on-time performance of flight rerouting under thunderstorm conditions, the airspace was first discretized into grids. Considering meteorological avoidance zones, aircraft performance, and fuel consumption, an ARRPTC model incorporating multiple safety factors was established based on the aircraft's origin, destination, and departure time. The model determined both the aircraft's airborne holding time at the departure point and its optimal rerouting path, with the objective of minimizing the total flight time. Subsequently, a multi-strategy improved DBO (MSDBO) was designed according to the problem characteristics to improve convergence speed and accuracy while avoiding premature convergence of the algorithm. The performance of the proposed MSDBO was then compared with four other swarm intelligence algorithms by means of six benchmark test functions to verify the effectiveness of the improvement strategies. Finally, a case study based on an actual aircraft rerouting scenario under thunderstorm conditions was conducted. The results indicate that, compared with traditional models, the proposed ARRPTC model reduces the detour time and distance by 52.3% and 53.0%, respectively, while the total flight time and distance decrease by 16.5% and 14.3%, and fuel consumption is reduced by 13.6%. Compared with the other four swarm intelligence algorithms, MSDBO demonstrates faster convergence, stronger global exploration, and superior local exploitation capabilities. In the ARRPTC model, MSDBO can obtain more optimal flight trajectories, significantly improving the efficiency of flight rerouting. Furthermore, as the thunderstorm safety threshold and turbulence coefficient threshold decrease, the total rerouting distance tends to increase.

In order to ensure the safe operation of autonomous ships, it is necessary to deeply analyze the risk factors of contact accidents and formulate targeted safety prevention and control strategies. Based on the contact accident investigation reports and international literature on autonomous ships, this paper adopts systems theoretic process analysis, the decision-making trial and evaluation laboratory method and the fuzzy cognitive map model to analyze the risk causes of contact accidents for autonomous ships from multiple perspectives, and assess the risk level of two typical accident scenarios, namely, berthing/unberthing in port waters and navigating through bridge areas, and base on this, propose corresponding prevention and control measures. The results show that during the port waters, autonomous ships face ten types of key risks such as inaccurate perception and interrupted information transmission and during the bridge areas, there are nine types of key risks such as perceptual blindness and decision-making errors. And insufficient communication and collaboration, significant wind/current interference and other factors are the key triggers of contact accidents for autonomous ships. These factors involve the key control behavior of perception, decision-making, control and communication of autonomous ships, and there is also a complex coupling and amplification effect among different factors. Finally, prevention and control measures are proposed, including improving the ship-shore data exchange protocol and joint exercise mechanism, integrating multi-source sensors and machine learning algorithms, developing a plan for human-machine permission switching, and using VR/AR technology to conduct immersive emergency training.

In order to prevent dust cloud combustion and explosion accidents in the high sulfur content concentrate warehouse after beneficiation, lead concentrate and zinc concentrate of Panlong Lead Zinc Mine were taken as examples. Tests on explosion intensity and the lower explosion limit at different mass concentrations were conducted using a 20 L balls. Main parameters, including the maximum explosion pressure, explosion lower limit, and explosion index of the lead zinc concentrate powder dust cloud, were obtained. The results show that maximum explosion pressures of lead concentrate and zinc concentrate dust clouds were 0.335 and 0.251 MPa, respectively, both occurring when the mass concentration of the dust cloud was 1 000 g/m³. The lower explosive limits of mass concentration are 160-170 and 210-220 g/m³, respectively, indicating low explosion risk. The explosion risk and hazard of dust clouds in lead concentrate are higher than those in zinc concentrate. The explosion mechanism of concentrate dust cloud particles being ignited, exploded, and subjected to a chain reaction is obtained. These findings provide an experimental and theoretical basis for preventing dust cloud explosion of concentrate in the concentrate warehouse.

To address the challenges that static prediction models face in accurately forecasting dynamic landslide trends and the high computational costs associated with dynamic models—which hinder real-time prediction—this study proposes a novel real-time model for landslide displacement prediction. The model integrated Particle Swarm Optimization (PSO), a Gated Recurrent Unit (GRU) network, Online Learning (OL), and dynamic Model Retraining (MRT). First, a static landslide prediction model was established by integrating PSO and GRU. Second, an OL strategy was incorporated into the static model, enabling dynamic updates and real-time predictions as new monitoring data were acquired. Then, small-batch MRT was performed based on prediction accuracy evaluation to predict landslide trends dynamically and in real time. Finally, a comparative analysis of several related models was conducted using the Wangeryan landslide in Sichuan Province as a case study. The results indicate that the OL and MRT methods significantly improve prediction accuracy. Specifically, the PSO-GRU-OL-MRT model achieved Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), and R² values of 0.795, 3.53, 1.40, and 0.954, respectively, with an average prediction time of 25.0 seconds per instance, demonstrating the highest prediction accuracy. In comparison, the GRU-OL-MRT model yielded values of 1.73, 7.82, 2.54, and 0.917 for the same four metrics, with an average prediction time of 0.496 seconds per instance, significantly reducing computational costs while maintaining relatively high prediction accuracy.