Issue 08, 2026
Determination of the self-sustainability for smoldering propagation by dynamic energy balance method
Song Zeyang;Wang Huajuan;Deng Jun;Determining the self-sustainability of smoldering propagation is crucial for effective smoldering disposal in solid waste management and for preventing and controlling smoldering fires. Traditional methods, such as local energy balance and global energy balance approaches, do not accurately assess the self-sustainability of smoldering propagation. This article presents a dynamic energy balance evaluation method based on a dynamic control volume, which takes into account the relationship between the control volume and the energy balance method. The effectiveness and accuracy of this method were verified through multi-condition smoldering propagation experiments and a calibrated two-dimensional numerical model. The advantages of the dynamic energy balance method are highlighted through comparative analysis with traditional global and local energy balance approaches. Research findings indicate that the local energy balance method fails to account for the impact of the preheating zone on the smoldering system, leading to an overly optimistic assessment of self-sustainability. Conversely, the global energy balance method overestimates the influence of heat loss in the cooling region, resulting in inaccurate assessments of the propagation of weakly self-sustaining smoldering. The dynamic energy balance method resolves these issues and allows for a systematic determination of both self-sustainability and the strength of self-sustenance. When the rate of the dynamic control volume is positive, smoldering propagation can be considered self-sustaining; otherwise, it cannot sustain itself without an additional heat source. If Eret,ceramic, dynamic is stable or continuously increasing, the smoldering propagation is classified as strong self-sustaining. Conversely, if Eret,ceramic, dynamic continues to decrease and approaches zero, the propagation is deemed weakly self-sustaining. When Eret,ceramic, dynamic equals zero, the propagation ceases, and smoldering is extinguished. The proposed dynamic energy balance method not only serves as an effective analytical tool for gaining deeper insights into the behavior of smoldering propagation and optimizing the disposal of high-moisture organic solid waste but also provides theoretical and technical support for the prevention and control of smoldering fires.
Numerical simulation study on shock wave propagation laws in mine- scale gas and coal dust explosions
Fang Yingxiang;Gao Jianliang;Zhang Xuebo;To investigate the propagation laws of shock waves induced by gas and coal dust explosions along return air networks at a mine scale, this paper conducts numerical simulations using a segmented successive simulation method, which incorporates geological conditions, ventilation parameters, and the mining layout of working face Wu8-20120 in Pingdingshan No.10 coal mine. This approach divides the entire simulation model into two distinct sections to balance computational accuracy and efficiency, which is crucial for simulating large-scale mine return air networks with complex structures. A three-dimensional model is established for the explosion zone to accurately capture the explosion process, pressure release, shock wave initiation, and initial propagation characteristics. In contrast, a two-dimensional model is adopted for the propagation zone to reduce computational costs and improve efficiency without sacrificing key propagation information. Shock wave parameters at the section interface, including pressure, velocity, and temperature, are applied as transient inlet boundary conditions to ensure the continuity and reliability of the simulation results. The reliability of the selected mathematical model is verified by comparing numerical results with experimental data from full-scale mine roadway tests reported in the literature, with error analysis indicating good consistency. Results show that overpressure and impulse in roadway regions beyond 10 meters from the outlet boundary are minimally affected by boundary conditions, providing important guidance for setting reasonable simulation boundaries. The attenuation laws obtained from both the full-section two-dimensional and three-dimensional models are consistent; however, the full-section two-dimensional model generally underestimates peak overpressure with significant maximum relative error, proving it unreliable for simulating real large-scale mine roadway gas explosions. The segmented simulation yields similar attenuation laws to the full three-dimensional simulation, with peak overpressures along the tunnel closely aligned and a maximum relative error of only 2.77%, demonstrating the necessity of using a three-dimensional model for the explosion zone. Propagation and attenuation characteristics within the return air network are analyzed, revealing that overpressure at each position varies with explosion equivalent in a power function form(p=a+bxc), and peak overpressure attenuates significantly at bends and bifurcations. It is noted that greater gas and coal dust filling lengths lead to stronger attenuation.
Temperature prediction model for coal spontaneous combustion based on IBWO-CNN-RF
Zhang Yilong;Liu Zhenyang;Yang Zhibin;Ding Xinpu;Xu Yang;Liu Jingyi;Bai Gang;To enhance the accuracy of predicting coal spontaneous combustion temperatures, we propose an Improved Beluga Whale Optimization-Convolutional Neural Network-Random Forest(IBWO-CNN-RF) prediction model. This model integrates deep feature extraction, ensemble regression, and intelligent optimization techniques. The standard Beluga Whale Optimization(BWO) algorithm is enhanced by introducing a Circle mapping-based population initialization mechanism, which increases population diversity, along with an adaptive spiral predation strategy that dynamically balances global exploration and local exploitation throughout the iterative process. To evaluate the optimization capability of the proposed IBWO algorithm, six classical single-modal and multi-modal test functions are employed, and its performance is compared with four other swarm intelligence algorithms. The results indicate that the IBWO algorithm achieves faster convergence, enhanced global search capability, and greater stability across various function types. Following this, the IBWO algorithm is utilized to optimize key hyperparameters of the CNN-RF model, including the initial learning rate, maximum training epochs, the number of convolutional kernels in both the first and second CNN layers, and the number of trees in the RF. This optimization improves the model's generalization ability and robustness. In this hybrid architecture, the CNN automatically extracts deep spatial representations from six characteristic gas indicators, while the RF model conducts nonlinear regression for temperature prediction. Comparative experiments with other models—including CNN, RF, Grey Wolf Optimizer-Support Vector Regression(GWO-SVR), Particle Swarm Optimization-Back Propagation Neural Network(PSO-BPNN), and BWO-CNN-RF—demonstrate that the proposed IBWO-CNN-RF model achieves the lowest values for σRMSE, σMAPEand σMAE, as well as the highest R2 on both training and testing sets. Finally, the model is applied to the programmed temperature-rise experimental dataset from the No.2 coal seam in the Chiyu coal mine, Shanxi Province, China. The results confirm that the IBWO-CNN-RF model exhibits superior prediction accuracy, stability, and generalization capability, providing an effective technical foundation for the intelligent prediction and early warning of coal spontaneous combustion disasters.
Research on predicting blasting vibration velocity based on SSA-XGBoost model
Xu Zhichao;Chen Yunshan;Deng Chao;Zhao Xu;Zhou Shuangshuang;Ji Jing;Accurate prediction of blasting vibration velocity is crucial for ensuring engineering safety, minimizing environmental disturbance, and optimizing blasting operations. To address the challenges of low prediction accuracy and poor adaptability associated with traditional empirical formulas, this paper proposes a blasting vibration velocity prediction model based on the Sparrow Search Algorithm(SSA)-optimized eXtreme Gradient Boosting(XGBoost). To effectively construct a prediction index system, we selected several evaluation indicators for blasting vibration velocity, including the distance from the blast center, elevation difference, rock mass integrity coefficient, total explosive charge, maximum single-shot charge, and minimum burden. A dataset for predicting blasting vibration velocity was established based on these indicators. To overcome the limitations of the XGBoost model arising from manual parameter tuning, which can lead to insufficient generalization ability, SSA was employed to adaptively optimize the hyperparameters of the XGBoost model, thereby enhancing both its prediction performance and generalization capability. The model was trained and evaluated using 5-fold cross-validation and compared against other models, including Support Vector Regression(SVR), Multi-Layer Perceptron(MLP), Random Forest(RF), unoptimized XGBoost, and SSA-RF. The prediction accuracy of each model was comprehensively assessed using metrics such as the coefficient of determination(R2), Root Mean Squared Error(ERMSE), Mean Absolute Error(EMAE), and Variance Accountability Fraction(VAF). Additionally, feature importance analysis within XGBoost was performed to quantify the relative contribution of each influencing factor in predicting blasting vibration velocity. The results indicated that the SSA-XGBoost model achieved the best prediction performance during cross-validation, with the highest R2 value of 0.961 4 and a VAF of 96.09%, along with the lowest ERMSE and EMAE values of 0.273 5 and 0.071 8, respectively, surpassing all comparative models. Compared to the unoptimized XGBoost model, the ERMSE and EMAE of SSA-XGBoost decreased by 34% and 65%, respectively, validating the effectiveness of the SSA algorithm in improving model accuracy. Moreover, the feature importance analysis revealed that the distance from the blast center, maximum single-shot charge, and rock mass integrity coefficient are the three core factors influencing blasting vibration velocity. This study provides a new and efficient method for predicting blasting vibration velocity and offers a scientific basis for optimizing blasting parameter design and controlling vibration effects.
Microscopic influence investigation of defects on the fracture behavior of X80 hydrogen-blended transmission pipelines
Chen Liqiong;Du Shiqi;He Ting;Lai Xin;During the transportation of hydrogen-blended natural gas through X80 pipelines, inherent porosity and crack defects serve as significant hydrogen traps, promoting hydrogen adsorption, localized accumulation, and subsequent initiation of hydrogen-induced damage. Understanding the influence of these microdefects on fracture behavior is essential for evaluating the structural integrity of hydrogen-containing pipeline steels. In this study, molecular dynamics simulations were conducted using the LAMMPS package to investigate the uniaxial tensile response of polycrystalline Fe-C alloy models containing voids of varying sizes and cracks of different lengths. The corresponding mechanical properties, fracture modes, and underlying deformation mechanisms were systematically analyzed. The results indicate that increasing pore size significantly intensifies local stress concentration, causing atoms near the void surface to enter the nonlinear deformation regime at relatively low stress levels. This premature dislocation nucleation reduces the tensile strength from 10.2 GPa to 9.2 GPa, reflecting a decline of approximately 9.8%. The peak strain of the models varies with crack length; specifically, as the crack length increases from 1 nm to 3 nm, the tensile strength decreases from 9.9 GPa to 8.8 GPa, representing a reduction of approximately 11.1%. Mechanistically, voids promote necking, dislocation pile-up, and vacancy-assisted propagation, leading to the formation of intergranular fracture pathways. In contrast, crack-driven fractures are dominated by constrained plasticity at the crack tip, shear-band linkage, and localized brittle extension. Structural analysis further reveals pronounced hydrogen enrichment at void edges and crack tips, which weakens atomic coordination and increases the fraction of Hexagonal Close-Packed(HCP)/Face-Centered Cubic(FCC) phases. Consequently, grain boundaries become preferred fracture routes, and all models consistently exhibit intergranular fracture characteristics. Overall, this study provides atomistic-level insight into the coupled effects of voids and cracks on the fracture behavior of hydrogen-charged X80 steel. The findings elucidate how defect geometry governs load-bearing capacity, ductility, and microfracture evolution, offering a valuable theoretical foundation for assessing hydrogen embrittlement susceptibility, optimizing pipeline steel microstructures, and guiding the safe design of hydrogen transport infrastructure.
Safety monitoring method for tower crane climbing and descending operations based on displacement calculation and IoT technology
Li Hang;Li Zhe;Wang Tingdong;Han Yuhao;Yu Bo;A safety monitoring method that integrates displacement calculation with IoT technology has been developed to address limitations in data fusion, real-time performance, and warning delays during tower crane climbing and descending operations. The system employs five key sensors: a draw-wire displacement sensor with 1 mm resolution, a wind speed sensor installed at the top of the tower, an inclinometer on the sleeve frame, micro-switches on climbing claws and pin shafts, and AI cameras for personnel behavior recognition. All sensor data are transmitted via wired connections to an edge-cloud platform for unified analysis. A dynamic calculation model for jacking cylinder displacement was established based on standard section statistics and operational step analysis. For two-stroke cranes utilizing 2.8 m standard sections, key step-change criteria were defined. For three-stroke cranes with 3.0 m sections, statistical thresholds such as Δa=5 mm and Δb=8 mm were derived from 280 operational datasets. The integrated monitoring system features a “perception-analysis-warning-transmission” framework, which includes real-time sensing devices, edge intelligent hosts, cloud servers, embedded software, cloud platforms, and mobile applications. Thirteen monitoring indicators encompass mechanical status, dynamic parameters, environmental factors, and personnel behavior. The execution mechanism combines “step-jumping logic” with “single-factor triggered three-level alarm logic”. Level I alarms for critical risks trigger continuous 110 dB sound-light alarms, along with mobile notifications to on-site managers and SMS/WeChat alerts to safety supervisors. Level II alarms initiate intermittent 90 dB alarms and technician notifications, while Level III alarms activate low-frequency 70 dB prompts. Empirical validation across seven engineering projects involving 14 tower cranes demonstrated that the displacement sensors achieved a measurement accuracy of ±1 mm. The system recorded a 0 false alarm rate, with average response delays of 0.75 seconds for Level I alarms and ≤2 seconds for mobile notifications. AI cameras achieved recognition accuracies of 96.8% for hard hats, 94.5% for safety belts, and 98.2% for personnel intrusion detection. Micro-switches at critical positions attained a 100% abnormal detection rate.
Multi-spectral adaptive gated hypergraph detection algorithm towards safe autonomous driving
Liao Huan;Gu Chengjie;Zhu Dongjun;Meng Yi;Tian Shu;To address the challenges of intra-modal feature degradation and insufficient cross-modal interaction in multispectral object detection for autonomous driving, we propose the Hypergraph Dynamic Gated Fusion Network(HyperDGFNet). This network integrates hybrid adaptive gating optimization with hypergraph semantic guidance to enhance cross-modal interaction while mitigating the impact of intra-modal feature loss. Specifically, an interactive gated hybrid attention module—designed with a CNN-based architecture—employs channel-spatial hybrid self-attention equipped with an adaptive gating mechanism to suppress intra-modal redundancy and improve cross-modal complementarity. Furthermore, a hypergraph fusion module captures cross-modal correlations through cross-layer semantic modeling, explicitly encoding non-local spatial dependencies across spectral domains. The hypergraph structure represents regional features as nodes and connects semantically related regions across modalities and layers via hyperedges. This approach overcomes the locality constraints of standard convolutions, enabling long-range interactions among heterogeneous features and facilitating coherent multimodal integration, particularly benefiting complex driving scenarios. For precise localization of small distant objects, a high-resolution detection head preserves fine-grained spatial details through multi-scale feature aggregation, enhancing sensitivity to objects with limited pixel coverage. This ensures accurate detection of challenging targets, such as pedestrians and cyclists, at long ranges. Experimental results on the Multi-modality Full-time dataset for Autonomous Driving(MFAD) dataset demonstrate that HyperDGFNet achieved 80.5% mAP@0.5 and 57.4% mAP@0.5:0.95, surpassing EI~2Det by 1.5 percentage points and 4.1 percentage points, respectively. Ablation studies validated the effectiveness of each module: the gated attention significantly improves cross-modal alignment, the hypergraph fusion enhances robustness, and the high-resolution head markedly boosts small-object detection. This method establishes a new benchmark through dynamic gating, hypergraph modeling, and high-resolution detection, providing reliable support for autonomous driving environmental perception.
Pedestrian safety distance warning method that integrates feature histograms and binocular vision
Liu Xunxing;Zhang Haimin;This article proposes the integration of HOG features with an improved YOLOv3 network for pedestrian detection and employs binocular vision for safety warnings. Methodologically, HOG features were utilized to extract multi-scale distribution characteristics from pedestrian images. Feature vectors were generated through several steps, including image preprocessing, gradient calculation, cell histogram construction, and block normalization, and were subsequently fed into the modified YOLOv3 network. A feature preprocessing module was added to the input layer to adjust the dimensionality of the HOG feature vectors, while a feature fusion layer was introduced to merge features using weighted concatenation. The YOLOv3 network was adapted into a pedestrian-specific detector by removing unnecessary loss functions related to pedestrian category probability prediction. By leveraging binocular vision information based on feature histograms, detection accuracy was enhanced through the extraction and fusion of feature histograms from left and right camera images. The distance between pedestrians and vehicles was calculated using the binocular vision method based on the identified pedestrian locations. Multiple factors were considered to compute a safe distance, which was then compared with the measured distance to provide warnings. Additionally, high-precision calibration plates were employed to calibrate the binocular cameras, and high-performance hardware alongside optimized algorithms was selected to address calibration and time-consumption challenges. A reaction time model was developed to obtain accurate reaction times, with brake actuation time determined through braking tests. Experiments were conducted using the Caltech, CityPersons, and KITTI datasets. Results demonstrated that the proposed method accurately labeled pedestrian locations and counts under various lighting conditions. The estimated minimum braking distance closely matched the actual value, and data processing time significantly influenced the minimum braking distance. The real-time warning performance remained above 0.8 throughout, reaching over 0.95 after 40 iterations. The pedestrian detection accuracy, missed detection rate, false alarm rate, recall rate, and F1 score were 91%, 9%, 2%, 89%, and 90%, respectively. Overall, the performance surpassed that of comparative methods.
Research and application of micro fog dust isolation technology under induced airflow influence
Jing Deji;Liu Shaohang;Liu Hongwei;Ren Shuaishuai;Kan Qisheng;Li Donghui;This article presents a micro fog dust isolation system based on supersonic atomization technology to tackle the intensified dust diffusion caused by airflow generated during the operation of coal mining machines in fully mechanized coal mining faces. By integrating Computational Fluid Dynamics, the k-ε turbulence model, and the finite element method, we systematically analyzed the transport behavior of dust under the combined influence of induced airflow and environmental airflow. Numerical simulations indicate that large-particle dust(>40 μm), primarily affected by centrifugal force, is ejected tangentially and settles down, while fine dust particles(<10 μm) form a suspended dust cloud due to the balance between drag force and centrifugal force. The upper drum area of the shearer emerges as a critical zone for dust accumulation because of airflow turbulence. The developed dust control system includes a full-section fog curtain installed between adjacent hydraulic supports and a supersonic atomization device mounted externally on the shearer. Experimental validation shows that the full-section fog curtain exhibits excellent wind resistance under typical site air velocities ranging from 1 m/s to 2 m/s, maintaining a deflection angle of less than 8 degrees. The ultrafine droplets produced by the system effectively capture dust, a process well explained by Stokes collision theory. After the field implementation of this system at the 45205 working face in the Sandaogou Coal Mine, sampling and measurement data indicated that the average dust suppression efficiency exceeded 88% for both total dust and respirable dust concentrations. Specifically, the total dust concentration at the high-concentration point(measuring point 4) was reduced from 465 mg/m3 to 44.3 mg/m3, while respirable dust decreased from 220 mg/m3 to 19.5 mg/m3. This system significantly lowered overall dust concentration levels within the working face, thereby substantially improving the working environment conditions on-site. This study elucidates the multi-field coupled dust control mechanism involving airflow, dust, and mist droplets in the 45205 fully mechanized mining face of Sandaogou Coal Mine, providing a theoretical foundation and technical support for dust control in fully mechanized mining operations.
Research on the pre-control of dust hazards in the regeneration construction of old industrial buildings based on NSGA-Ⅱ
Tian Wei;Hu Jiale;Li Xinyang;Guan Xiaojie;Wang Nan;The challenging working conditions during the regeneration of old industrial buildings can lead to significant dust accumulation, presenting serious health risks to construction workers. Unlike traditional dust control strategies that primarily address mid-and post-exposure phases, this study proposes an innovative proactive approach to preventing dust hazards through the optimization of construction schedules. Initially, the operational characteristics of regeneration activities were analyzed. Similarity-based experiments on typical old industrial buildings were conducted to identify variations in dust hazards associated with different construction tasks, specifically focusing on four high-dust operations: concrete grinding, rebar drilling, concrete crushing, and waste handling. Field investigations were performed to determine exposure parameters for workers engaged in these tasks, allowing for accurate quantification of dust hazard levels across various construction activities. To validate the dust dispersion behavior, Computational Fluid Dynamics(CFD) numerical simulations were employed, using concrete grinding as a primary example. Building on these findings, a “worker health” objective function was integrated into the construction scheduling model. In addition to objectives related to time and cost, a multi-objective optimization model for the regeneration of old industrial buildings was developed using the NSGA-Ⅱ algorithm. Chain-coding was utilized to represent the construction schedule, with the algorithm configured for a crossover probability of 0.78 and a mutation probability of 0.12. Non-dominated sorting was employed as the method to solve the model. Following this, multi-attribute decision analysis was applied to optimize the allocation of resources and scheduling for dust-generating activities. By adjusting task durations and resequencing operations, the number of simultaneous dust-producing activities was reduced, leading to a decrease in onsite dust concentration and enabling proactive control measures. An empirical analysis was conducted on an old industrial plant in Beijing's Shougang Park. The results indicated that the optimized schedule achieved a 3.09% reduction in project duration and a 0.88% cost saving, alongside a significant 20.9% decrease in dust exposure. This approach effectively mitigates the generation and diffusion of dust at the source, demonstrating successful preemptive control of dust hazards in the regeneration of old industrial buildings.