Constructing a high-fidelity, multi-scale, and continuously updated geometric model of mine tunnels is fundamental to achieving intelligent mining. To overcome the inefficiency, delayed updates, and poor environmental fidelity of traditional 3D tunnel modeling methods, this study proposes a multi-scale digital twin geometric modeling approach for mine tunnels. First, the overall architecture of the method (MineDT-GM) is designed. At the global scale, parameterized automatic modeling is employed to parse CAD drawings for rapid construction of a 3D topological tunnel network, while traverse point data are integrated to drive the dynamic evolution of the model. At the local scale, adaptive voxel rasterization, composite filtering, dynamic-radius BPA reconstruction, and Trimesh-based hole filling are utilized to achieve high-precision modeling. A parallel computing framework is designed to enhance processing efficiency. For global-local integration, the method achieves multi-scale fusion at both the model and data levels. This is accomplished through coordinate matching and hierarchical storage, ensuring seamless interaction between the global and local models. Experiments were conducted using real data from two large and complex mines. Results show that, for a mine with 1653 tunnels, global modeling is completed in 148 s. For a dataset containing 31,045,538 points, local reconstruction takes only 15.05 s, which is just 0.72% of the time required by Poisson reconstruction, while demonstrating superior modeling quality and robustness. The method enables efficient construction and dynamic updating of global-scale tunnel networks and precise mapping of local-scale models, providing a unified visual representation and digital foundation for mine digitization.

The deformation and stress have vital influences on the geometric accuracy and mechanical property of assembly structure that determine the quality of aircraft product. In order to realize real-time perceptions and controls of deformation and stress in aircraft assembly, a hybrid modelling strategy is proposed to determine the static response of the assembly structure composed of parts with high and low rigidities. The parts are meshed and separately applied by linear and nonlinear analysis models to represent explicit mappings from nodal force to nodal displacement and stress. The linear model of high-rigid part with invariant stiffness is established by the method of influence coefficient, while the nonlinear static response of low-rigid part with variable stiffness is represented by the Kriging model that is trained by the dataset of displacement and stress under multiple groups of loads. Combining individual model of each part by the application of coupled degree of displacement freedom to the nodes on connection surface, a hybrid analysis model of the whole assembly structure is established. Attributed to the decoupling of linear and nonlinear terms in the Kriging model, all the parameters in the hybrid model can be identified offline, and the mappings from force to displacement and stress of the assembly structure are rapid for real-time computation. The indicators evaluating the stress concentration and displacement magnitude are established, and a multi-objective optimization model of holding forces is further established and solved for the regulations of deformation and stress on the assembly structure. A software and a flexible fixture with servo system are developed to monitor and control the assembly stress and deformation. Simulation results show that the hybrid model can represent the static response of the assembly structure precisely and the optimization of holding force can suppress the stress and deformation efficiently. The maximum analysis errors of total displacement and Von-Mises stress on the assembly structure are separately 0.032 mm and 8.2 MPa corresponding to their real data of 0.45 mm and 186 MPa, while the displacement and stress are controlled below 0.009 mm and 1 MPa after the optimization of holding force. Experimental verification demonstrates that the assembly deformation and stress can be instantly perceived and regulated online by the sensing and controlling of holding forces.

The mine ventilation system serves as the cornerstone of safe mine production, and its core task is to achieve timely and on-demand air supply. However, the calculation of the required airflow is intricate and time intensive, and the calculation results usually lag behind, which poses a serious threat to mine safety. In this paper, a hybrid model for feature extraction, component prediction, and result reconstruction is proposed. On the basis of the sensor data, the airflow required for air-consuming areas can be accurately predicted in advance. First, a real-time computational method and architecture to support substantial and reliable data for prediction are described. Then, the variational mode decomposition (VMD) technique is optimized by employing the whale optimization algorithm (WOA) to minimize the envelope entropy, and the enhanced VMD (EVMD) method is proposed. The hybrid prediction model integrates a WOA, a convolutional neural network (CNN) and a bidirectional long short-term memory (BiLSTM) neural network. Finally, two significantly different wind position datasets from a coal mine in Shanxi Province were collected. Different signal decomposition methods, workplace datasets, dynamic time sliding windows, prediction time scales, and ablation experiments were used to verify the performance of the model. 18 experiments were performed. The proposed model achieved greater prediction accuracy and robustness than did the contrast model. The proposed model effectively addresses the issues of delayed calculation of the required airflow and inaccurate prediction of the air volume, thereby ensuring the safety, health, and sustainable production of the mine.

To solve the problem of difficult high-strength bonding between ordinary resin anchoring agents and surrounding rocks under conditions such as soft and fractured rock mass and geological structures, this study proposes a method of adding expansion agents to anchoring agents. The aim is to enhance the bonding properties of the resin anchors with the surrounding rocks by inducing the expansion effect of the resin anchors, and ultimately improve the overall anchoring effect. Three kinds of expansion systems were selected, including polymethylene polyphenyl isocyanate, alum and sodium bicarbonate, aluminum sulfate and sodium bicarbonate. Through the expansion rate test and performance test of resin anchorage agent, the expansion rate of each expansion system and the related performance data of expansive resin anchorage agent with aluminum sulfate and sodium bicarbonate as expansion agent were obtained. The results show that the expansion system of polymethylene polyphenyl isocyanate, alum and sodium bicarbonate only has obvious expansion effect on the medium-speed anchorage agent, while the expansion system of aluminum sulfate and sodium bicarbonate has obvious expansion effect on all kinds of anchorage agents. The addition of aluminum sulfate and sodium bicarbonate has little effect on the gelation time of resin anchorage agent. In the range of 0%-20% expansion rate, the compressive strength is negatively correlated with the mass fraction of aluminum sulfate and sodium bicarbonate, but all can reach the minimum standard of compressive strength, while the anchorage force is positively correlated with the mass fraction of aluminum sulfate and sodium bicarbonate. In mudstone without cracks, the anchoring force can be increased by up to 82.9%, and the anchoring force after exchanging the installation sequence of anchoring agents is higher than that of the original installation sequence; in fractured mudstone, the anchoring force of 20% expansion rate anchoring agent can be increased by 104% compared with that of non-expansion anchoring agent.

Traditional mine ventilation expert systems mainly rely on manually predefined rules and isolated knowledge modules, which limits their adaptability, update efficiency, and support for complex scenario-oriented decision-making. To address this issue, this study proposes an intelligent mine ventilation knowledge base for event capture and scheme reasoning. At the knowledge representation level, a unified framework integrating a ventilation ontology space, a global ventilation knowledge graph, and a task-oriented ventilation event graph is established to organize ventilation entities, semantic relations, abnormal events, and response actions in a structured manner. At the reasoning level, a method combining ontology extraction, rule discovery, and knowledge-based inference is developed to support real-time event identification and scheme generation from multi-source ventilation data. Furthermore, a software prototype is implemented on Neo4j to realize knowledge storage, visualization, and reasoning execution. A simplified real-mine case is used to verify the practical applicability of the proposed method. The results show that the knowledge base can identify coupled abnormal ventilation conditions, trace the corresponding reasoning path, and generate feasible optimization schemes, demonstrating its value for intelligent ventilation decision support and emergency response in coal mines.

To address the challenges encountered in complex mine ventilation networks, such as optimization decisions that heavily rely on human expertise and the inefficiency caused by redundant computations in the existing decision-making algorithms when they are applied to large-scale ventilation network optimization tasks, a reinforcement learning-based decision-making method (RLDM) is proposed in this paper for achieving real-time, proactive optimization and knowledge integration in mine ventilation systems. By integrating a self-learning policy function, the method constructs intelligent agents for main fans and air regulators. Upon perceiving ventilation data, these agents can make optimal decisions based on the current system state and execute them, iteratively performing a “sense–actuate” cycle until the target ventilation system reaches its optimal state. A dynamic reward function with adaptive penalty and incentive mechanisms is designed for the ventilation optimization process, enabling knowledge integration through both value-based and policy-based learning. A centralized decision-making and knowledge integration architecture is established, supporting a collaborative multiagent mechanism with shared global state information, along with a practical engineering implementation workflow. Experiments conducted using real-world mine ventilation data demonstrate that the agents learned rapidly along the gradient direction. Compared with the traditional baseline methods based on evolutionary computations, the proposed RLDM approach effectively reuses historical decision experiences, achieving fast, accurate, and low-resource-cost proactive optimization decisions. Moreover, it supports the continuous accumulation, updating, and reuse of decision-making knowledge. This research has significant implications for reducing the amount of manual labor required in mines, enhancing production safety, and lowering the consumption of ventilation energy.

Ventilation network diagrams play a vital role in mine safety, enabling simulation, airflow control, and emergency planning. Traditional manual drawing methods are time-consuming, error-prone, and difficult to synchronize with evolving mine structures. To address this, we propose an automated framework that generates schematic ventilation diagrams from spatial models while preserving topological fidelity. The framework integrates a hierarchical node layering algorithm based on Hasse diagrams, a trunk-branch decomposition strategy for structural abstraction, and a deformation-based curve layout model inspired by water droplet geometry to enhance visual clarity and structural alignment. A 3D elevation inference module ensures semantic consistency across 2D diagrams, 3D models, and graph representations. The system further supports reactive multi-view synchronized editing, enabling coherent modifications across different representations while preserving user-defined constraints. Although developed and validated in mine ventilation, the framework is generally applicable to directed-graph-based industrial systems. Its generality is further demonstrated on a schematic exhaust pipeline case, where a tree structure rather than a network diagram provides a more suitable abstraction, highlighting that the method supports domain-adaptive abstraction guided by topological characteristics. This approach has been integrated into real-world mining operations through the "Ventilation Brain System," offering a scalable solution for intelligent layout generation, adaptive design iteration, and responsive decision-making. By bridging spatial realism with schematic abstraction, the proposed method streamlines diagram construction and reinforces intelligent mine ventilation management.

To effectively manage and utilize the vast amount of ventilation data in coal mines and meet the national demand for intelligent mining systems, the development of knowledge graphs and intelligent question-answering (Q&A) systems is a critical step in the intelligent transformation of mine ventilation. Using techniques like web scraping and classification analysis, knowledge data from the mine ventilation domain is collected and integrated with expert input to build an ontology model. This model serves as the foundation for constructing the mine ventilation knowledge graph, with 6 935 entity annotations and vertical relationship associations completed. Based on this knowledge graph, a vector-based intelligent Q&A framework was designed. Through the creation of question intent identification rules and answer templates, a vector-based Q&A model was developed. To verify its applicability, 200 professional questions related to mine ventilation were tested. The results show an overall accuracy of 95% for the model, with 97% accuracy for single-turn questions and 93% for multi-turn continuous questions. Compared to rule-based models, the vector-based model demonstrates significant advantages in multi-turn interactions. Future research will further improve the accuracy by integrating large language models for enhanced semantic analysis. This intelligent Q&A system will reduce the workload of ventilation personnel, increase management efficiency, and provide vital support for the full implementation of intelligent ventilation systems in coal mines.

In recent years, the gas control method of “Y-type ventilation+ high drainage roadway” has been widely used with its superior control effect. Affected by the negative pressure of high extraction roadway, the air leakage law of working face is more complicated, which affects the effect of gas extraction. Taking Yuwu Coal Industry as the engineering background, Comsol software is used to simulate the air leakage law and gas extraction effect of different extraction positions and extraction negative pressure conditions of high extraction roadway, and the optimal layout position and extraction negative pressure of high extraction roadway are determined and verified on site. The results show that the negative pressure of extraction has a significant effect on the air leakage of the working face. When the negative pressure of extraction is 0 kPa, the airflow field in the goaf is fan-shaped, and deflects from the working face to the direction of gob-side entry retaining. With the increase of extraction negative pressure, the goaf in the range of 260-328 m of the working face tends to form a negative pressure interception effect, and a triangular reflux area is formed inside the goaf on the side of the gob-side entry retaining. The location of the high drainage roadway and the change of the negative pressure of the drainage mainly affect the air leakage wind speed in the 170 - 328 m area of the working face and the 0 - 140 m area of the gob-side entry retaining. The air leakage and gas extraction parameters of high extraction roadway under different extraction conditions are comprehensively compared and analyzed. When the extraction negative pressure is 11 kPa, the extraction position is 30 m horizontally and 35 m vertically, the gas extraction effect is the best. The field measured wind speed and gas concentration are consistent with the numerical simulation results.

This study investigates the maximum temperature in tunnel fires under forced ventilation conditions by constructing a 1:10 scale experimental model and using numerical simulations. A dimensionless derivation of the maximum temperature is provided for the case where the fire source is at the end of the excavation tunnel. A correction factor for the maximum temperature prediction coefficient is suggested in situations when the fire source is situated in the center of the excavation tunnel. The results of the highest temperature in the experimental and FDS simulation results under different fire source conditions have a good fitting performance, and the correlation is 0.899 and 0.913. The effect of the forced ventilation outlet distance on maximum temperature was analyzed through wind flow and temperature field distributions. The study concludes that an optimal layout for the ventilation system is important, particularly by considering both the maximum temperature and the ventilation volume. This research addresses the gap in understanding maximum temperatures in excavation tunnel fires and offers valuable insights for fire suppression strategies, rescue operations, and the prevention of secondary disasters in such environments.

Mine ventilation corpora contain fragmented entities, attributes, and rule-based knowledge, marked by heterogeneous expressions, implicit structures, and ambiguous terminology. These characteristics hinder systematic modeling and intelligent utilization. To address these challenges, we propose an automated extraction and semantic alignment method based on large language models (LLMs), aiming to construct a high-quality knowledge graph (KG) tailored for mine ventilation. We design an ontology-driven extraction framework for three textual sources - regulations, books, and websites - using prompt engineering and few-shot strategies to extract information, domain knowledge, entity attributes, and rule-based relations in a unified way. For entity alignment, we develop a dual-filtering mechanism that integrates semantic similarity and structural adjacency, and leverage large language models (LLMs) for verification, enabling high-confidence alignment of entities and predicates. We propose a rule-path modeling strategy using the structure "subject entity (with condition)-predicate-object entity (with condition)," integrating multi-source triples into conditional rule chains. These are mapped into a graph database to support structured knowledge representation. Under few-shot conditions, the extraction accuracy of entity attributes and rule-based relations reached 94% and 99%, respectively. The final alignment of entities and predicates, manually verified, achieved 100% precision. The resulting knowledge graph (KG) comprises 58,358 entities and 63,630 edges, demonstrating strong semantic consistency and structural integrity. It provides structured knowledge support for risk warning, question answering (QA), and intelligent decision-making in mine ventilation systems as part of intelligent mining applications.

Pressure swirl nozzles are widely applied in various mineral industries, and their excellent spray performances have been found to be directly related to dust removal results. In this study, in order to examine the influencing effects of the swirl intensity levels of pressure swirl nozzles on atomization characteristics and dust-reduction efficiency, five spiral guide grooves were added for the purpose of improving the quality of the atomization and dust control. The swirl intensities of pressure swirl nozzles were first theoretically investigated using the established models. The effects of the swirl intensities on the atomization angles and distributions of the atomization sizes were comparatively studied using theoretical calculations, numerical simulations, and experimental methods. A dynamic wind tunnel was built in order to conduct dust-reduction experiments with different swirl intensities. The proximate and ultimate analysis results and the sizes of the coal samples were characterized, and the mechanism of the dust reduction was summarized. The results demonstrated that the applications of spiral guide grooves improved the atomization angles and coverage areas by up to 98.8 degrees and 73%, respectively. At the same time, the enhancements of the swirl intensities were observed to reduce the uniformity index of the distribution sizes,` which significantly decreased the Sauter mean diameter (SMD) from 101.3 mu m to 22.1 mu m. The atomization results revealed that the increasing of the stagger angles benefited the larger atomization angles and smaller SMD. In addition, the proposed theoretical model had achieved better prediction accuracy. Also, the high atomization quality had obviously improved the dust-reduction ability of the pressure swirl nozzles, with the efficiency of total dust and respirable dust removeable increased by approximately 11.23% and 9.33%, respectively.

To examine the permeability alterations in coal seams induced by liquid CO2 injection, we employed COMSOL simulations to recreate the flow, force, and temperature fields, along with the corresponding physical parameters of the geological formation. We established a mathematical model to track the evolution of the coal body’s permeability, enabling us to capture the phase distribution of CO2, the dynamics of fluid transportation, and the coal body’s mechanical responses during the liquid CO2 injection process. Our findings revealed that the coal seam undergoes significant changes due to low-temperature freezing and shrinkage, gas-liquid phase transitions, and expansion impacts from solid-liquid phase changes. These factors collectively modify the pore structure, subsequently influencing the coal seam’s permeability. Notably, the coal body’s permeability, influenced by the pore structure and the Klinkenberg effect, increases logarithmically with rising injection pressure and falling temperatures. The structural changes in the coal body are markedly impacted by seepage and phase change pressures, with the permeability enhancement from secondary phase changes being 2.28 times greater than that from primary phase changes, and five times greater than in scenarios without phase change.

According to the demand of regional gas control in Wangpo Coal Mine crushed soft coal seam area,the pneumatic direc-tional drilling technology was introduced to optimize the existing pneumatic directional drilling process and equipment.Through up-grading the pneumatic screw drilling tool,the rated torque was increased to more than 450 N·m.The 89 mm spiral cable drilling pipe with low wind resistance was developed to realize the construction hole depth of the connected cable measuring system of pneumat-ic directional drilling to more than 300 m.Based on the tecnique of the"wave type"accurate detection of coal seam top and bottom trajectory,from the principle of directional drilling and branching,the technology of rapid branching is realized by improving the strength of drilling tool.We systematically analyzed pneumatic directional drilling in different strata trajectory variation characterist-ics,summed up the precise trajectory control technology,and the pneumatic directional drilling process construction efficiency is greatly improved.During the field test,a total of 34 main holes and 49 branch holes were constructed,with a total footage of 17 287 m,an average hole depth of 305 m,an average daily footage of 103 m/d and a maximum of 154 m/d.The average extraction concen-tration of the borehole was above 60%,and the pure extraction volume of the 100-meter borehole was 2.7 times that of the ordinary borehole in the adjacent area.

In order to study the influence of bedding on the mechanical properties and permeability of coal samples, a triaxial gas seepage device and acoustic emission monitoring system developed by self-developed were used to conduct triaxial compression tests on coal samples at different bedding angles and to collect the corresponding acoustic emission data, and the failure forms of coal samples were analyzed combined with the Mohr-Coulomb strength criterion. The results show that the mechanical deformation characteristics of coal samples with different bedding angles have obvious bedding effects under triaxial loading, and the peak strength, elastic modulus and Poisson’s ratio all decrease first and then increase with the increase of bedding angles. The permeability of coal samples at each bedding angle shows the same trend, showing a U-shaped change of first decreasing and then increasing. At the same time, the initial permeability and minimum permeability of coal samples increase with the increase of bedding angle, and the C-90° bedding coal sample is 2.5-5 times that of other bedding coal samples. The acoustic emission signals are mainly concentrated in the yield stage and the post-peak failure stage, and in the post-peak failure stage, the ringing meter value of the coal sample reaches the peak. The acoustic emission signal generated by the coal sample at 0° is also the strongest, while the acoustic emission signal generated by the coal sample at 45° is the weakest. The Mohr-Coulomb strength criterion can well reflect the failure mode of coal sample with single stratified group, and the experimental results are basically in agreement with the theoretical analysis.

Resistance-alteration-based multiple-fault diagnosis of mine ventilation systems is essential to ensuring the safety of mine production. The basic assumptions, definitions, framework, theory, algorithms and experiments regarding the resistance-alteration-based multiple-fault diagnosis of mine ventilation systems are systematically studied here. First, the problems of the single-fault assumption in the conventional ventilation system fault diagnosis framework are analyzed, and a real-time online multiple-fault diagnosis framework is proposed. Then, based on the theory of resistance observability, the sensor layout scheme is optimized, the properties of the ventilation subnetwork are studied, and the interpretable resistance-alteration-based multiple-fault diagnosis (RMFD) algorithm is designed. Finally, a diagnosis experiment with multiple faults was carried out for a real coal mine. The experimental results show that the RMFD algorithm can achieve 100% accuracy of identification and positioning at the ventilation subnetwork level and can perform quantitative resistance-alteration analysis for k - 1 roadways within a k-order star subnetwork, which verifies the effectiveness of the real-time online multiplefault diagnosis framework and the RMFD algorithm. This study achieves real-time online multiple-fault diagnosis of mine ventilation systems and provides a theoretical reference and technical support for the intelligentization of mine ventilation systems and similar fluid networks.

In order to meet the strategic needs of intelligent upgrading of mine ventilation system, the design principle, logical structure and specific implementation of intelligent ventilation precision control system are studied. The current situation of intelligent ventilation control system is reviewed from three aspects of state perception, control algorithm and control strategy. In view of the shortcomings of the current intelligent ventilation control system that each research field is separated from each other and the data are difficult to support each other, starting from the top-level design, the whole life cycle design principle of the system with the fusion of perception algorithm and control algorithm as the core is proposed. Based on this, the logical framework with the edge’s integration of perception and control as the core is proposed. The hardware and software platform of mine intelligent air conditioning is designed and built, and various measures are taken to ensure the safety of people and vehicles and the reliable operation of the system. Aiming at the specific problem that the state perception and precise control algorithm are relatively separated in the traditional framework, which leads to the sensitivity of air volume control to noise, a proportional integral differential closed-loop control algorithm combined with the Kalman filter is proposed to accurately track the air volume of roadway, and the decision and control concepts of expert modeling, algorithm optimization and equipment linkage are integrated. Finally, in the hardware and software platform of mine intelligent air conditioning, the accurate tracking experiment of wind window air volume in normal period and the smoke flow elimination experiment in disaster period are carried out. The experimental results show that compared with the traditional proportional-integral-derivative closed-loop control algorithm, the proportional-integral-derivative closed-loop control algorithm combined with the Kalman filter improves the accuracy, anti-interference and overall efficiency of air volume tracking by 20.3%, 17.4% and 13.9%, respectively, and realizes the accurate and efficient control of air volume in the normal period. In the smoke flow exclusion experiment, the real-time control of constant pressure difference on both the sides of the wind window can take into account the suffocation fire extinguishing and the prevention of smoke flow invasion, so as to achieve the precise control of smoke flow in the disaster period. The two experiments verify the effectiveness of design principles and logical framework of intelligent ventilation precise control system.

Propose This work has been done in order to solve the problems of long time consuming, complex operation, and poor environmental applicability of the existing methods for detection of calcium hydroxide absorption rate in emergency rescue work. Methods A method to invert the availability of calcium hydroxide by using acoustic velocity was proposed. By measuring the change of acoustic velocity in calcium hydroxide sample, the residual CO2 absorption rate was determined, and the availability was evaluated according to coal industry standards. On the basis of the experiment of CO2 absorption by calcium hydroxide with time, the law of CO2 absorption by calcium hydroxide with time was discussed. This variation rule is caused by the combination of the product CaCO3 coating barrier to the inward CO2 diffusion and the continuous reduction of the effective volume of calcium hydroxide sample along with the continuous reduction of the contact surface between Ca(OH)2 and CO2. According to the test experiments of calcium hydroxide absorption rate and acoustic velocity, a mathematical model for judging the available of calcium hydroxide by acoustic velocity is established. Findings The test results show that the relative error of the test results is less than 1% with a single test time of 1 to 2 min, which achieves high efficiency and high precision test of calcium hydroxide absorption rate.

To study the effect of evaporation and humidification characteristics of spray water curtain on the relative humidity of the tunnel, the calculation model for relative humidity was established in the spray space. The effect of different control parameters on the humidity in the tunnel was explored based on numerical simulation. The results show that the mathematical model of spray droplet evaporation and humidification can effectively predict the relative humidity of roadway; the spray flow rate and atomization pressure, the air supply volume of the air duct, the ambient temperature, and humidity all contributed significantly to the relative humidity of the tunnel. By optimizing the parameters, it is possible to not only reduce the effective range of spray humidification on tunnel humidity but to also improve the moisture removal efficiency after spraying.

Semiclosed tunnels are very common in engineering construction. They are not connected, so they easily accumulate heat. Once a fire breaks out in a semiclosed tunnel, the route for rescue workers to enter is limited, so it is tough to get close to the fire source. In this paper, taking a mine excavation roadway with local pressure ventilation as an example, the temperature field distribution and water spray fire prevention characteristics of the excavation roadway face were studied using numerical simulation and theoretical analysis. This paper provides an explanation of a dynamics-based smoke management method for water spraying in a semiclosed tunnel as well as the equilibrium relationship between droplet drag force and smoke buoyancy. A method was first developed to calculate the quantity of smoke blockage based on the thickness of the smoke congestion. The local ventilation and smoke movement created a circulating flow in the excavation face, which was discovered by investigating the velocity and temperature fields of the excavation face. The size of the high-temperature area and the pattern of temperature stratification varied due to this circulating flow. When local ventilation and sprinkler systems were operating simultaneously, when the volume of smoke was small, the smoke avoided the majority of the water spray effect with the circulation flow; however, when the volume of smoke was large, the effect of the circulation flow decreased and the smoke gathered close to the sprinkler head. At this time, the blocking effect of the water spray was significant. The mean square error analysis revealed that activating the sprinkler had the most significant cooling impact on the wall on one side of the air duct.