A long-distance slag conveying numerical self-adaptive control method based on CFD

Abstract

The application provides a long-distance slag conveying numerical self-adaptive control method based on CFD, relates to the technical field of slag conveying pipeline analysis, and comprises the following steps: constructing a pipeline real-time data set; generating a calculation grid; combining a mathematical model and a dynamic boundary condition to construct a CFD dynamic simulation system of a pneumatic conveying pipeline; performing solution calculation on the CFD dynamic simulation system to obtain a flow parameter distribution model in the slag conveying pipeline; performing feature extraction on the pipeline real-time data set and the flow parameter distribution model, mining the correlation between features and fault types through a weight matrix and a cross-term weight matrix, obtaining a multi-modal fault diagnosis judgment result, and generating a slag conveying pipeline self-adaptive control strategy. The technical problems that the simulation accuracy is limited, it is difficult to support high-reliability design, a multi-physical field data fusion mechanism is missing, accurate fault diagnosis cannot be achieved, dynamic self-adaptive analysis capability is lacked, and the timeliness of fault early warning and regulation strategies is insufficient in the prior art are solved.

Description

Technical Field

[0001] This application relates to the field of slag conveying pipeline analysis technology, and in particular to a CFD-based numerical adaptive control method for long-distance slag conveying. Background Technology

[0002] With the acceleration of industrialization, the demand for long-distance slag conveying systems in mining, building materials, and chemical industries is increasing daily. Pneumatic slag conveying technology, with its advantages of high conveying efficiency, wide adaptability, and excellent sealing, has become one of the core means of material transportation. However, the complex gas-solid two-phase flow characteristics during long-distance conveying place extremely high demands on system design and operation. To address this challenge, computational fluid dynamics (CFD) technology is widely used in the numerical simulation of slag conveying systems. By constructing high-precision CFD models, the distribution patterns of velocity, pressure, and particle concentration fields within the pipeline can be comprehensively analyzed, providing a scientific basis for optimizing conveying efficiency, predicting pressure drop losses, and assessing system stability, significantly improving design efficiency and operational reliability.

[0003] While CFD technology has shown promise in slag conveying system research, existing methods still have significant limitations. First, traditional design relies on empirical formulas and simplified physical models, making it difficult to accurately reflect the complex interactions between particles and fluids under real-world conditions, leading to large prediction errors for key parameters. Second, existing CFD models lack the ability to simulate multi-fault coupling scenarios; for example, they cannot simultaneously and accurately characterize the synergistic effects of particle deposition, pipe blockage, and leakage. More importantly, current technology lacks the ability to deeply fuse multi-source data, making it difficult to accurately identify fault types based solely on single flow field characteristics.

[0004] In summary, the design and operation of current long-distance slag conveying systems face the following core challenges: traditional CFD models have limited accuracy in simulating complex gas-solid coupled flows, making it difficult to support high-reliability designs; the lack of multiphysics data fusion mechanisms hinders accurate fault diagnosis; and the absence of dynamic adaptive analysis capabilities results in insufficient timeliness of fault warning and control strategies. These problems severely restrict the operational efficiency and safety of slag conveying systems, necessitating the development of a method to overcome existing technological bottlenecks and promote the intelligent upgrading of long-distance slag conveying systems. Summary of the Invention

[0005] This disclosure provides a CFD-based numerical adaptive control method for long-distance slag conveying to address the technical problems in existing technologies, such as limited simulation accuracy, difficulty in supporting high-reliability designs, lack of multi-physics data fusion mechanisms, inability to achieve accurate fault diagnosis, lack of dynamic adaptive analysis capabilities, and insufficient timeliness of fault warning and control strategies.

[0006] According to a first aspect of this disclosure, a CFD-based numerical adaptive control method for long-distance slag transport is provided, comprising:

[0007] A sensor cluster is deployed at the core monitoring location of the slag conveying pipeline to collect real-time pipeline transportation data and construct a real-time pipeline dataset. A three-dimensional physical model is constructed based on the pipeline's geometry, and the computational domain of this model is determined. The computational domain is discretized using a mesh generation method to generate a computational mesh. A CFD dynamic simulation system for the pneumatic conveying pipeline is constructed by combining a mathematical model and dynamic boundary conditions. The mathematical model includes governing equations and a turbulence model, and the dynamic boundary conditions include inlet boundary conditions, outlet boundary conditions, and wall boundary conditions. The CFD dynamic simulation system is solved to obtain a flow parameter distribution model within the slag conveying pipeline, which includes a velocity field, pressure field, and particle concentration field. Feature extraction is performed on the real-time pipeline dataset and the flow parameter distribution model. The correlation between features and fault types is mined using weight matrices and cross-term weight matrices to obtain multimodal fault diagnosis results. The multimodal fault diagnosis results are analyzed, and each fault type is classified and processed to generate an adaptive control strategy for the slag conveying pipeline.

[0008] One or more technical solutions provided in this disclosure have at least the following technical effects or advantages:

[0009] A sensor cluster is deployed at the core monitoring location of the slag conveying pipeline to collect real-time pipeline transportation data and construct a real-time pipeline dataset. A three-dimensional physical model is constructed based on the pipeline's geometry, and the computational domain of this model is determined. The computational domain is discretized using a mesh generation method to generate a computational mesh. A CFD dynamic simulation system for the pneumatic conveying pipeline is constructed by combining a mathematical model and dynamic boundary conditions. The mathematical model includes governing equations and a turbulence model, and the dynamic boundary conditions include inlet boundary conditions, outlet boundary conditions, and wall boundary conditions. The CFD dynamic simulation system is solved to obtain a flow parameter distribution model within the slag conveying pipeline, which includes a velocity field, pressure field, and particle concentration field. Feature extraction is performed on the real-time pipeline dataset and the flow parameter distribution model. The correlation between features and fault types is mined using weight matrices and cross-term weight matrices to obtain multimodal fault diagnosis results. The multimodal fault diagnosis results are analyzed, and each fault type is classified and processed to generate an adaptive control strategy for the slag conveying pipeline. This invention addresses the technical problems in existing technologies, such as limited simulation accuracy, difficulty in supporting high-reliability designs, lack of multi-physics data fusion mechanisms, inability to achieve accurate fault diagnosis, and lack of dynamic adaptive analysis capabilities, resulting in insufficient timeliness of fault early warning and control strategies. It achieves the technical effects of improving simulation accuracy, enabling high-reliability designs, accurately diagnosing faults, enhancing the timeliness of fault early warning and control, improving operational efficiency and safety, and promoting intelligent upgrades.

[0010] The above description is merely an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in this disclosure or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0012] Figure 1 This application provides a CFD-based numerical adaptive control method for long-distance slag transport. Detailed Implementation

[0013] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0014] Example 1: The personalized teaching strategy generation method based on student data analysis provided in this disclosure is as follows. Figure 1 The methods include:

[0015] S1: Deploy sensor clusters at core monitoring locations of the slag conveying pipeline to collect real-time pipeline transportation data and construct a real-time pipeline dataset.

[0016] Specifically, a data acquisition network for long-distance slag transport pipelines is constructed. Core monitoring locations are systematically calibrated based on pipeline design drawings, covering areas prone to blockage, leak-sensitive sections, and routine detection points. Sensor clusters are deployed at core monitoring locations using a multi-physical quantity collaborative sensing architecture. Using the edge device hardware clock as a reference, sensor time is synchronized to millisecond levels via the NTP protocol, and then a cubic spline interpolation algorithm is used to unify the time axis of signals with different sampling rates. The raw data is preprocessed using a classification and denoising strategy. The processed pressure, vibration, temperature, mass flow rate, and velocity data are encoded and sorted by location number, constructing a spatiotemporal matrix dataset with timestamps as row indices and spatial and sensor composite labels as column indices, thus obtaining the real-time pipeline dataset.

[0017] S2: Construct a three-dimensional physical model based on the geometry of the slag conveying pipeline, determine the computational region of the three-dimensional physical model, and discretize the computational region using a mesh generation method to generate a computational mesh;

[0018] Specifically, based on the slag conveying pipeline design drawings, various geometric features of the pipeline were obtained, and a three-dimensional physical model of the pipeline was constructed using parametric modeling tools. The computational domain was defined, and boundary conditions were preset. The internal cavity of the pipeline was defined as the fluid computational domain, ensuring the model was closed and free of permeable gaps. Virtual extensions of appropriate lengths were added at the inlet and outlet to eliminate the influence of boundary effects on the core flow field. Boundary conditions were preset based on flow characteristics: the inlet was set as a velocity inlet, the outlet as a pressure outlet, and the pipe wall as a no-slip boundary condition. Simultaneously, particle injection and escape surfaces were marked, and corresponding reflection and capture boundary conditions were set. An unstructured mesh generation strategy was adopted, generating a global mesh based on tetrahedral elements. The global mesh size was determined according to the minimum characteristic size of the pipeline, and the mesh growth rate was controlled to ensure a smooth transition between adjacent element sizes. Boundary layer meshing was enabled in the near-wall region. The initial layer height was determined based on the target value, and appropriate total layers and interlayer expansion ratios were set to accurately capture wall shear stress and velocity gradients. Finally, complex flow regions such as the area with the maximum curvature on the inner side of elbows, the separation zone downstream of diameter transition sections, and the heat-affected zone of welds were identified and locally refined. Create a dense control domain, reduce the mesh element size within the dense control domain, and use a size function to achieve a smooth transition from the dense control domain to the normal region. For particle-fluid coupling simulations, increase the element density in regions with dense particle trajectories. After densification, check the element orthogonality and Jacobian determinant, and discard elements that do not meet the requirements.

[0019] S3: Construct a CFD dynamic simulation system for pneumatic conveying pipelines by combining mathematical models and dynamic boundary conditions. The mathematical model includes governing equations and turbulence models, and the dynamic boundary conditions include inlet boundary conditions, outlet boundary conditions, and wall boundary conditions.

[0020] Specifically, a mathematical model is determined based on the type of material transported in the pipeline, encompassing both governing equations and a turbulence model. The governing equations are discrete phase equations, using an Eulerian-Lagrange framework to describe gas-solid phase exchange, while the turbulence model is the RNG k-ε model. Dynamic boundary conditions are set based on real-time pipeline datasets, constructing a CFD dynamic simulation system for pneumatic pipelines to form an intelligent simulation framework. In constructing the mathematical model, the gas phase uses the three-dimensional transient Navier-Stokes equations to describe the flow field evolution; particle motion simulation employs the Lagrange particle tracking method, describing trajectories according to Newton's laws of motion. Phase coupling is achieved through bidirectional exchange using the Schiller-Naumann drag model and the Ranz-Marshall heat transfer model, while particle collisions are modeled using Hertz-Mindlin contact theory. The RNG k-ε model for turbulence simulation introduces a Crowe correction term to characterize the modulation of turbulence by particles. Some model input parameters are derived from the real-time pipeline dataset, while others are manually set and input into the CFD. Dynamic boundary conditions are set using the real-time pipeline dataset, and mass flow rate, pressure, and temperature data are obtained to calculate relevant boundary condition data. The inlet velocity is calculated using the inlet mass flow rate data of the pipeline; the total outlet pressure is obtained by superimposing the static pressure and dynamic pressure terms at the outlet; and the wall heat flux density is calculated based on the centralized temperature data of the pipeline in real time.

[0021] S4: Solve the CFD dynamic simulation system to obtain the flow parameter distribution model in the slag conveying pipeline. The flow parameter distribution model includes the velocity field, pressure field, and particle concentration field.

[0022] Specifically, the transient solution mode is activated in the solver to clarify the time-related mechanisms. When setting the time step, the PISO method is used for pressure-velocity coupling, achieving convergence and overhead in 3-5 sub-iterations. The discretization scheme is configured according to the equation characteristics: the momentum equation uses a second-order upwind scheme for the convection term to suppress numerical dissipation, and a central difference scheme for the diffusion term to maintain accuracy; the turbulence model uses a first-order upwind scheme for the convection term to maintain stability and locally refines the accuracy compensation; the particle phase is updated using explicit time integration within the Lagrangian framework. Finally, the initial field is constructed, and the flow distribution parameter model is solved.

[0023] S5: Extract features from the real-time pipeline dataset and the flow parameter distribution model, and mine the correlation between features and fault types through the weight matrix and cross-term weight matrix to obtain multimodal fault diagnosis and judgment results;

[0024] Specifically, the real-time pipeline dataset is spatiotemporally aligned with the flow parameter distribution model for joint analysis. The CFD simulation time step and sensor sampling frequency are matched, mapping sensor locations to CFD grid nodes. Feature extraction is performed on the spatiotemporally aligned data to achieve fault diagnosis. Features of each field in the CFD full-field distribution data are extracted: the velocity field calculates the axial velocity decay index to reflect velocity changes; the pressure field calculates the pressure gradient along the axial direction, and the proportion of abnormal grids indicates blockage faults; the particle concentration field estimates the wall deposition thickness, thus obtaining a deposition index to determine the deposition status. Temporal dimension analysis is performed on the real-time pipeline dataset. Vibration data is analyzed to obtain kurtosis and impulse factors, thereby determining the operating conditions. Cross-correlation analysis is performed on temperature and mass flow rate data to obtain the thermal hysteresis coefficient, identifying leakage faults. Coherence function analysis is performed on pressure and flow rate data, and blockage or leakage faults are determined based on the phase difference. The correlation between CFD and real-time data features is mined. Seven feature settings and three fault types are obtained, a matrix operation model is established, the input feature vector is used to output the fault probability, and the existence of the fault is determined.

[0025] S6: Analyze the multimodal fault diagnosis results, classify and process each fault type, and generate an adaptive control strategy for the slag conveying pipeline.

[0026] Specifically, the fault type is determined, and the faulty section is located using real-time pipeline datasets and flow parameter distribution models for adaptive adjustment. For particulate deposition faults, a tiered treatment process is initiated. First, the upstream fan speed is increased to bring the airflow velocity in the fault area to 1.3-1.5 times the design value within 30 seconds. Simultaneously, a high-frequency pulse valve group is activated 2 meters upstream of the deposition area to release compressed air pulses. If the deposition volume does not decrease by 30% within 20 minutes, deep treatment is initiated, deploying the guide vane array, injecting fluidizing agent, and using a CFD model for real-time monitoring to ensure a balance between stripping efficiency and energy consumption. When a blockage fault is identified, the intelligent gate valves at both ends of the faulty pipe section close within 0.5 seconds, and the blockage clearing system injects high-pressure gas from downstream, delivering a three-stage pulse impact. If the effect is unsatisfactory, a secondary response is triggered, and a pig moves to the blockage point, switching operating modes according to the blockage density. External vibration sensors monitor the blockage, and for high-temperature molten blockages, the blockage is heated to the softening point before clearing. When a leak occurs, the leak coordinates are first determined, the pressure control module reduces pressure in stages, and a negative pressure generator is activated downstream to suppress leak propagation. Then, the intelligent sealing system performs chemical repair in three stages, with distributed fiber optic sensors monitoring the sealing layer status. After the fault is repaired, an enhanced monitoring period begins, with the monitoring frequency three times the normal rate in the first week to assess the repair effectiveness. Preventative maintenance procedures are also initiated, including regular flushing and coating maintenance. The predictive maintenance module uses machine learning to analyze historical data and build a degradation model. When the predicted remaining lifespan is below a threshold, a purchase order and work order are automatically generated, allowing for advance scheduling of repair units. This significantly reduces unexpected downtime and improves system reliability.

[0027] Furthermore, step S1 of this application also includes:

[0028] The easily clogged areas, leakage-sensitive sections, and routine detection points of the slag conveying pipeline are designated as the core monitoring locations. The easily clogged areas include bends, diameter changes, and vertical sections. The leakage-sensitive sections are welded joints. The routine detection points are evenly marked every 5 meters along the pipeline, starting from the pipeline inlet and ending at the pipeline outlet. If a routine detection point is less than 1 meter away from the easily clogged area, leakage-sensitive section, or pipeline outlet, that point is removed.

[0029] A sensor cluster is installed at the core monitoring location of the slag conveying pipeline. The sensor cluster includes pressure sensors, vibration sensors, temperature sensors, mass flow sensors, and velocity sensors.

[0030] The sensor cluster continuously collects pressure, vibration, temperature, mass flow rate, and velocity data during the operation of the slag conveying pipeline to construct a real-time pipeline dataset. The real-time pipeline dataset uses timestamps as row indexes and spatial and sensor composite labels as column indexes.

[0031] Specifically, a data acquisition network for long-distance slag conveying pipelines is constructed to systematically pinpoint the core monitoring locations of these pipelines. Based on pipeline design drawings, easily clogged areas such as bends, diameter changes, and vertical sections, as well as leakage-sensitive sections such as circumferential welds and longitudinal welds, are accurately identified. The pipeline inlet is designated as the starting point, and the pipeline outlet as the ending point. Laser rangefinders are used to evenly delineate regular monitoring points every 5 meters along the pipeline. To avoid redundant data collection, if a regular monitoring point is less than 1 meter away from a previously identified easily clogged area, leakage-sensitive section, or pipeline outlet, the monitoring point at that location is removed.

[0032] To meet monitoring requirements, sensor clusters are deployed at all core monitoring locations, employing a multi-physical quantity collaborative sensing architecture: Intrusive piezoresistive pressure sensors are installed on the pipe sidewalls to directly measure fluid pressure; triaxial vibration sensors are fixed to the outer surface of the pipe wall using magnetic bases and anti-vibration locking screws to ensure signal stability; armored thermocouples are used as temperature sensors, installed at 60° circumferential intervals, with the sensing ends wrapped in ceramic fiber insulation sleeves to isolate environmental thermal interference; and Coriolis mass flow meters are used as both mass flow and velocity sensors to acquire mass flow and velocity data. In summary, each core monitoring location integrates pressure, vibration, temperature, mass flow, and velocity sensors, all centrally powered via an explosion-proof junction box, and locally networked using an RS-485 bus. Data is encrypted via an industrial gateway and uploaded in real-time to the cloud data center via a 4G fiber optic hybrid link.

[0033] Multi-source signal spatiotemporal alignment is performed, using the hardware clock of the edge devices as a reference. Millisecond-level time synchronization is achieved for each sensor via the NTP protocol. For different sampling rates, a cubic spline interpolation algorithm is used to resample all signals to a unified 100Hz time axis, ensuring strict temporal matching of pressure pulsation, pipe wall vibration, temperature changes, mass flow rate, and fluid velocity. The sensor cluster acquires raw data from the slag conveying pipeline in real time according to a set frequency and performs preprocessing operations on the raw data. In the data preprocessing stage, a classification denoising strategy is employed. Pressure data is first filtered using a 50ms window moving average to eliminate short-term disturbances, and then wavelet packet decomposition is used to identify and remove transient impact noise caused by water hammer. For vibration data, a db4 wavelet basis is used for 5-level wavelet decomposition, and a soft thresholding method is used to filter out high-frequency noise above 500Hz, retaining the dominant frequency components reflecting the dynamic response of the pipeline structure. Temperature data is filtered using median filtering to suppress abnormal jumps. Simultaneously, outlier detection is performed based on physical constraint rules and the 3σ statistical criterion; abnormal data segments are marked as invalid and linear interpolation is used for imputation. Mass flow rate and velocity data are preprocessed using data cleaning and alignment interpolation. The preprocessed pressure, vibration, temperature, mass flow rate, and velocity data are integrated and encoded and sorted according to sensor cluster location numbers. A spatiotemporal matrix dataset is constructed with timestamps as row indices and spatial and sensor composite labels as column indices to obtain the real-time pipeline dataset.

[0034] Furthermore, step S2 of this application also includes:

[0035] A three-dimensional physical model is constructed based on the actual geometry of the pipeline, the computational domain is defined, and boundary conditions are predefined.

[0036] An unstructured mesh is used to divide the interior of the pipe, and global mesh parameters are defined, including global mesh size and mesh growth rate.

[0037] Mark the encryption control domain inside the pipeline, and perform local mesh encryption in the encryption control domain.

[0038] Specifically, based on the slag conveying pipeline design drawings, the precise geometric features of the pipeline are obtained, including pipeline diameter, length, elbow structure, inlet and outlet shape and size, internal components, etc. A parametric modeling tool is used to construct a 3D physical model of the pipeline. A geometry repair engine is used to eliminate minor gaps and overlapping surfaces caused by drawing conversion, and unnecessary details such as bolt holes and nameplates are removed to simplify the model while preserving the continuity and closure of the internal flow channel. For critical areas such as elbows and diameter transition sections, feature locking is implemented to accurately preserve their geometric features; for example, the radius of curvature of elbows must be consistent with reality, and the taper angle and transition length of diameter transition sections must be strictly restored according to the drawings. For connection parts such as welds and flanges, smooth transition surfaces are constructed to avoid sharp edges and reduce the risk of subsequent mesh distortion. Finally, a closed geometry containing the complete internal flow channel is formed, clearly defining the boundaries between the fluid and solid domains.

[0039] Define the computational domain and predefine boundary conditions. Define the internal cavity of the pipe as the fluid computational domain, ensuring the model is completely closed and free of seepage gaps. Extend a virtual section at the inlet and outlet, approximately 3-5 times the pipe diameter in length, to eliminate the influence of boundary effects on the core flow field. Predefine the boundary condition types according to the flow characteristics: set the inlet as a velocity inlet, the outlet as a pressure outlet, and the pipe wall as a no-slip boundary condition. Simultaneously, mark the particle injection surface and escape surface in the model, and set reflection and capture boundary conditions.

[0040] An unstructured meshing strategy is employed, generating a global mesh based on tetrahedral elements. The global mesh size is determined according to the minimum characteristic dimension of the pipe, using 1 / 10 to 1 / 20 of the pipe's inner diameter as the baseline element edge length. The mesh growth rate is set to no more than 1.3 to ensure a smooth transition between adjacent element sizes. Boundary layer meshing is enabled in the near-wall region, with the initial layer height calculated from the target y+=30-50 value. The total number of layers is no less than 5, and the interlayer expansion ratio is controlled within 1.2 to accurately capture wall shear stress and velocity gradients.

[0041] Identify and locally refine complex flow regions. Create cylindrical or spherical refinement control domains at locations such as the maximum curvature on the inner side of elbows, the separation zone downstream of diameter transition sections, and the heat-affected zone of welds. Within the refinement control domain, reduce the mesh element size to 1 / 3 to 1 / 5 of the global mesh size, and use a size function to achieve a smooth transition from the refinement control domain to the normal region. For particle-fluid coupling simulations, additional element density needs to be added in regions with dense particle trajectories. After refinement, check the element orthogonality (greater than 0.1) and Jacobian determinant (greater than 0.5), and remove negative volume elements.

[0042] Furthermore, step S3 of this application also includes:

[0043] The mathematical model is selected based on the type of material transported in the pipeline. The mathematical model includes a governing equation and a turbulence model. The governing equation adopts the discrete phase equation and selects the Eulerian-Lagrange framework to describe the exchange of momentum, mass and energy between the gas and solid phases. The turbulence model adopts the RNG k-ε model.

[0044] Set dynamic boundary conditions for the model based on real-time datasets.

[0045] Specifically, real-time pipeline datasets are acquired, and a CFD dynamic simulation system for pneumatic conveying pipelines is constructed by combining multiphysics mathematical models and dynamic boundary conditions, forming an intelligent simulation framework. A hybrid Eulerian-Lagrange framework is used to describe the gas-solid two-phase flow, and a mathematical model is built. The gas phase is treated as a continuous medium, and the flow field evolution is described by solving the three-dimensional transient Navier-Stokes equations, including continuity, momentum, and energy equations. For particle motion simulation, the Lagrange particle tracking method is used, and Newton's laws of motion are used to describe the particle trajectory. The interphase coupling mechanism achieves bidirectional exchange of momentum and energy through the Schiller-Naumann drag model and the Ranz-Marshall heat transfer model. Particle collisions are modeled using Hertz-Mindlin contact theory, with a friction coefficient set to 0.4-0.6. The RNG k-ε model is used for turbulence simulation, and a Crowe correction term is introduced to characterize the modulation effect of particles on turbulence. The initial turbulence intensity is calculated based on the gas inlet velocity and wall roughness. During the calculation, adjustments are made dynamically based on the real-time pressure fluctuation spectrum. When an abnormal rise in the turbulent energy spectrum is detected in the high wavenumber region, the sub-relaxation factor of the turbulent dissipation rate ε is reduced from 0.7 to 0.5 to suppress numerical oscillations. Some of the input parameters required for the above models are directly obtained from the real-time pipeline dataset or indirectly derived from real-time data. The remaining parameters are manually set and input into the CFD. For example, the initial velocity and pressure fields required by the Navier-Stokes equations are directly obtained from the real-time pipeline dataset, while data such as gas density and viscosity need to be obtained indirectly. The particle diameter involved in the Schiller-Naumann drag model needs to be manually defined as an input.

[0046] Dynamic boundary conditions are set using a real-time pipeline dataset. Mass flow rate, pressure, and temperature data are acquired from the real-time pipeline dataset, and the inlet velocity (inlet boundary condition), outlet total pressure (outlet boundary condition), and heat flux density (wall boundary condition) of the pipeline model are calculated. Specifically, this involves obtaining the mass flow rate data at the pipeline inlet and calculating the inlet velocity data. Because the fluid flow at the inlet is unstable, with vortices and turbulence, it affects the propagation and reception of ultrasonic waves, leading to significant measurement errors. At the outlet, the fluid flow has stabilized, resulting in better measurement conditions and more accurate outlet velocity measurement. Therefore, the inlet velocity is calculated instead of sensor data, using the following formula: Where t represents time, characterizing the specific value of each data point at that time, and uin(t) represents the inlet velocity. The mass flow rate data at the pipe inlet is represented by D, where D is the pipe inner diameter, ρ represents the fluid density, and π is a constant. Pressure and velocity data at the pipe outlet are obtained, and the outlet pressure is calculated. The pressure data collected by the sensor at the outlet is static pressure, which needs to be superimposed with a dynamic pressure term to obtain the total outlet pressure. The specific formula is as follows: Where t represents time, P out (t) represents export pressure, P sensor (t) represents the static pressure data, 0.5uρ 2 The data represents dynamic pressure, u represents outlet velocity, and ρ represents fluid density. Temperature data is obtained from the real-time pipeline dataset. The average temperature of a portion of the pipeline segment is used as the wall temperature, and the outlet temperature is used as the fluid temperature. The heat flux density is calculated using the following formula: Where t represents time, qw(t) represents heat flux density, Tw(t) represents wall temperature, Tf(t) represents fluid temperature, and h(t) represents heat transfer coefficient, typically 5-25 W / (m²). 2 •K).

[0047] Furthermore, step S4 of this application also includes:

[0048] Set the solver parameters, adopt the transient solution mode, set the time step, select the pressure-velocity coupling algorithm, and balance the calculation speed and stability.

[0049] Discrete schemes are defined: the momentum equation adopts a second-order upwind scheme, the turbulence equation adopts a first-order scheme, and the particle phase equation adopts explicit time integration to ensure trajectory tracking accuracy.

[0050] The flow parameter distribution model is iteratively calculated and output, which includes the velocity field, pressure field, and particle concentration field.

[0051] Specifically, the mathematical model constructed in the transient multiphase flow simulation is solved to obtain the final CFD demonstration results. First, the transient solution mode is activated in the solver to clarify the influence mechanism of the time dimension on flow evolution. The time step setting needs to consider both the dynamic characteristics of the physical process and the numerical stability requirements: for the gas phase, the initial step size is estimated based on the inlet velocity and mesh size using the CFL condition (Courant number ≤ 1); for the particulate phase, the particle relaxation time is additionally considered, and the particle tracking time step is typically set to 1 / 10 to 1 / 100 of the fluid step size to avoid spurious dispersion or penetration phenomena caused by time integration errors in trajectory calculation. The pressure-velocity coupling algorithm preferentially adopts the PISO method. This algorithm handles transient pressure fluctuations through predictive correction loops, and is particularly suitable for scenarios with strong unsteady effects, such as eddy shedding and pressure pulsations induced by particle collisions. Its sub-iteration number is usually set to 3-5 times to balance convergence speed and computational cost.

[0052] The definition of discrete schemes is mainly based on the differentiated configuration of different governing equations. The convection term of the momentum equation adopts a second-order upwind scheme to suppress numerical dissipation, which is crucial for accurately capturing complex flow structures such as boundary layer separation and recirculation zones, while the diffusion term retains the central difference scheme to ensure second-order accuracy. Due to the presence of nonlinear source terms, the turbulence model is prone to numerical oscillations, so its convection term needs to be reduced to a first-order upwind scheme. The solution stability is maintained by introducing controllable numerical dissipation, while the accuracy loss is compensated by local mesh refinement, such as in the near-wall region y+≈30. The solution of the particle phase adopts explicit time integration under the Lagrangian framework, such as the Euler method, which directly updates the particle forces and positions based on the flow field information of the current time step. Although this method is subject to strict time step limitations, it can effectively avoid the phase delay problem caused by implicit integration, and is particularly suitable for tracking high-frequency dynamic processes of particle-fluid interaction.

[0053] The initialization phase requires constructing a reasonable initial field distribution. The velocity field is typically assigned values ​​based on the inlet boundary conditions, the pressure field is initialized as a zero-gradient field, and the turbulence parameters are estimated based on the inlet turbulence intensity and pipe diameter, using initial values ​​of k and ε. In the time-step loop, the gas phase solution follows a strict iterative sequence: first, the momentum equation is solved to obtain the predicted velocity field; then, the pressure field is updated and the velocity field is corrected using the pressure correction equation; subsequently, the turbulence equation is solved to update the turbulence parameters; and finally, particle phase tracking is performed. During particle tracking, multiple sub-steps are performed within each fluid time step, updating the particle motion state under drag and lift through explicit integration. If the particle volume fraction is high (>1%), the particle reaction force on the fluid needs to be fed back to the fluid momentum equation via linear interpolation, forming a two-way coupling. The convergence criterion requires simultaneous monitoring of the equation residuals and the flow and pressure fluctuation amplitudes at key monitoring points to ensure the solution reaches statistical steady state or periodic stability.

[0054] The output stage requires an efficient data storage strategy and the generation of a flow parameter distribution model to facilitate subsequent fault diagnosis. Full flow field data can be saved at preset time intervals for transient process playback, while real-time recording of velocity, pressure, and particle concentration data at key sections reduces storage pressure. Post-processing requires the comprehensive application of multi-dimensional analysis methods. Quasi-steady flow characteristics are extracted using the time-averaged field; transient streamline diagrams and vorticity isosurfaces are used to identify vortex structure evolution; spectral analysis of pressure pulsations identifies dominant frequency components; and particle phases are characterized by trajectory animation and spatial concentration distribution statistics to represent their transport characteristics. If residual oscillations or divergences occur during calculation, the time step should be reduced, the sub-relaxation factor strengthened, or the mesh quality checked. For particle penetration problems, the particle time step should be further refined or an elastic collision model introduced. Overall computational efficiency can be optimized through domain decomposition parallel computing and dynamic time step adaptive techniques. Through the above processing methods, a clear and well-defined flow parameter distribution model is finally generated, including velocity field, pressure field, and particle concentration field. This model covers the key parameters describing the flow field characteristics and provides comprehensive basic data for subsequent fault diagnosis, pipeline optimization, and other tasks.

[0055] Furthermore, step S5 of this application also includes:

[0056] The real-time dataset of the pipeline and the flow parameter distribution model are obtained, and the two are spatiotemporally aligned. The CFD transient simulation time step and the sensor cluster sampling frequency are matched, and the physical installation location of the sensor is mapped to the CFD grid node.

[0057] We extract features from the spatiotemporally aligned flow parameter distribution model and the real-time pipeline dataset as the raw data to construct a flow parameter feature set and a real-time data feature set.

[0058] By integrating the flow parameter feature set and the real-time data feature set, the correlation between the two and the fault is explored, a matrix operation model is constructed, and the multimodal fault diagnosis and judgment results are obtained.

[0059] Specifically, the process involves acquiring real-time pipeline datasets and flow parameter distribution models, and then aligning them spatiotemporally to facilitate subsequent joint analysis. Temporal alignment requires matching the time step of the CFD transient simulation with the sensor sampling frequency to ensure synchronization along the time axis. If the time steps are inconsistent, dynamic time warping is used to align the time axis. If a delay exists, cross-correlation analysis is used to determine the time offset. Spatial alignment involves mapping the physical installation locations of the sensors to CFD grid nodes. For mismatched locations, bilinear interpolation or radial basis function methods are used to interpolate the velocity, pressure, and particle concentration data from the flow parameter distribution model to the sensor coordinates, establishing spatial correlation.

[0060] Key features were extracted from the spatiotemporally aligned real-time dataset and the flow parameter distribution model. The features of the flow parameter distribution model mainly included the degree of axial velocity decay in the velocity field, the proportion of anomalous regions in the pressure field, and the wall deposition thickness in the particle concentration field. The features of the real-time dataset included the waveform characteristics of the vibration data, the hysteresis relationship between temperature and mass flow rate, and the phase difference between pressure and flow rate. The extracted feature data were integrated into a unified flow parameter feature set and a real-time data feature set for subsequent analysis.

[0061] A weight matrix is ​​used to establish the correlation between features and fault types, quantifying the contribution of each feature to different faults. To further improve diagnostic accuracy, a synergistic effect between features is introduced. Through an interactive enhancement mechanism, complex fault modes are accurately captured. Finally, the probability of occurrence for each fault is calculated, generating the final multimodal fault diagnosis result.

[0062] Furthermore, step S5 of this application also includes:

[0063] Feature extraction is performed on the flow parameter distribution model to obtain velocity field features, pressure field features, and particle concentration field features, and a flow parameter feature set is constructed. The velocity field feature is the axial velocity decay index, the pressure field feature is the proportion of pressure anomaly region, and the particle concentration field feature is the wall deposition thickness.

[0064] Feature extraction is performed on the real-time pipeline dataset to obtain vibration data features, temperature-mass-flow data coupling features, and pressure-flow data coupling features, and a real-time data feature set is constructed. The vibration data features include kurtosis value and impulse factor, the temperature-mass-flow data coupling feature is thermal hysteresis coefficient, and the pressure-flow data coupling feature is mean phase difference.

[0065] Specifically, spatiotemporally aligned CFD field data and real-time sensor data are used as raw data for feature extraction, transforming them into feature parameters that clearly reflect the fault modes, thus enabling multimodal fault diagnosis of slag conveying pipelines. The CFD output is full-field distributed data, primarily describing the spatial dimension, such as velocity, pressure, and particle concentration at each grid point. It is necessary to extract local or global features characterizing the fault from this data, including velocity field feature extraction, pressure field feature extraction, and particle concentration field feature extraction. The specific extraction methods are as follows:

[0066] Velocity field feature extraction requires calculating the axial velocity decay exponent, the specific formula of which is:

[0067] ;

[0068] Wherein, AVDI(x) represents the velocity decay index at position x, vx is the velocity data at position x, and Δx represents the position difference. This index indicates the velocity change inside the pipe. If the AVDI of a certain section of the pipe is continuously negative and has a large absolute value, it indicates that the fluid velocity in that area is decreasing rapidly, possibly due to particle deposition leading to a reduction in the flow area.

[0069] Pressure field feature extraction requires calculating the pressure gradient along the pipe axis. The specific formula is as follows:

[0070] ;

[0071] Where Pt represents the pressure gradient, x represents the position coordinate, p represents the pressure, and γ is the partial derivative, which is used to represent the rate of change of pressure data in the x-direction of the position coordinate, and is used to describe the change of pressure in different directions in space.

[0072] Set |Pt|>1.5P design The grid is considered an abnormal grid, according to the formula. Obtain the percentage of areas with abnormal pressure, where PAA represents the percentage of abnormal areas, Ab represents the total number of abnormal grid cells, and All represents the total number of grid cells. If this percentage exceeds 15%, it indicates that there is a blockage in the pipeline.

[0073] Particle concentration field feature extraction requires wall deposition thickness estimation. In the grid layer near the pipe wall, the particle volume concentration integral is calculated using the following formula:

[0074] ;

[0075] Where, δ deposit This represents an estimated wall deposition thickness, primarily serving as a key intermediate measure. Further analysis using its relationship to the pipe radius is needed to determine the extent of deposition. `i` is the index variable for the summation, ranging from 1 to N, representing the summation of N grids. N represents the total number of grids in the grid layer closest to the pipe wall. These grids are selected within a specific distance from the wall (e.g., 0.1 mm). `ci` represents the particle volume concentration in the i-th grid, reflecting the volume ratio of particles within each grid; it is a dimensionless quantity. `Vi` represents the volume of the i-th grid.

[0076] Dividing the deposition thickness by the pipe radius R yields the dimensionless deposition index D = δ. deposit / R, if D>0.05 (i.e., the deposition thickness exceeds 5% of the pipe radius), it is determined that deposition exists.

[0077] By integrating the axial velocity decay index, the proportion of pressure anomaly regions, and the wall deposition thickness data after feature extraction, a flow parameter feature set is constructed.

[0078] The real-time pipeline dataset is analyzed in the time dimension using methods including time-domain analysis and time-frequency analysis. Time-domain analysis involves calculating statistical characteristics such as the mean, standard deviation, and peak value of the data, as well as the signal's autocorrelation function and cross-correlation function. These characteristics reflect information such as the signal's amplitude, frequency, and phase. Time-frequency analysis can simultaneously obtain the signal's energy distribution at different times and frequencies, providing a more comprehensive description of the signal's time-varying characteristics. The specific operations are as follows.

[0079] Vibration data is analyzed to obtain kurtosis and impulse factor. Kurtosis is a statistic describing the sharpness of a signal distribution; it is the fourth-order normalized moment of the signal. For a discrete signal s(m), the kurtosis is calculated as follows:

[0080] First, calculate the mean of the signal. ;

[0081] Then calculate the standard deviation of the signal. ;

[0082] Finally, calculate the kurtosis value. ;

[0083] Where M is the signal length, m is the index variable, savg is the signal mean, σ is the signal standard deviation, and the numerator in the kurtosis formula represents the sum of the fourth power of the difference between each signal value and the mean. When kurtosis > 5 and impulse factor > 4, transient impact is identified, such as large particles getting stuck. When kurtosis < 3 and impulse factor < 2, steady vibration is identified, which is the normal operating condition.

[0084] The impulse factor is the ratio of the peak value to the mean value of a signal, used to measure the intensity of the impulse component in a signal. For a discrete signal s(m), the impulse factor is calculated as follows:

[0085] ;

[0086] Where max|s(m)| represents the maximum value in the signal, savg represents the mean value of the signal, and the impulse factor can be used to measure the intensity of the impulse component in the signal. It is very effective for detecting sudden pulses or transient signals in the signal. The larger the impulse factor value, the stronger the impulse component in the signal.

[0087] Cross-correlation analysis was performed on temperature and mass flow rate data to find the time delay corresponding to the maximum correlation. And calculate the thermal hysteresis coefficient TLC, the specific formula is as follows:

[0088] ;

[0089] in, The time delay corresponding to the maximum correlation. The average mass flow rate is TLC, and L is the pipe length. Under normal operating conditions, TLC is approximately 0.8 to 1.2, at which point heat transfer is matched with the flow rate. If there is a leakage fault, TLC will be significantly reduced because heat transfer is delayed due to fluid loss. In this case, the data may be less than 0.5.

[0090] Coherence function analysis was performed on pressure and flow data to calculate the phase difference in the frequency domain. Extract the average phase difference of the main frequency band. ,when When the temperature approaches -90°C, the pressure response lags behind the flow rate, indicating a blockage fault. When the angle approaches 90°, the flow response lags behind the pressure, indicating a leak.

[0091] By integrating the extracted kurtosis values, impulse factors, thermal hysteresis coefficients, and mean phase difference data, a real-time data feature set is constructed.

[0092] Furthermore, step S5 of this application also includes:

[0093] Based on the flow parameter feature set and real-time data feature set, a feature vector E and a fault type D are constructed, and then a weight matrix W is generated. The weight matrix W is specifically defined as follows:

[0094] ;

[0095] Wherein, the matrix row index corresponds to the feature, the column index corresponds to the fault type, and the element w ij This represents the contribution weight of the i-th feature to the j-th fault type, where R represents the set of real numbers, a common representation in matrices.

[0096] Based on the physical mechanisms of the characteristics and faults, the distribution of non-zero terms in the weight matrix and the initial contribution weight ratio are set.

[0097] Based on the feature vector E and the weight matrix W, the initial score for each fault is calculated using matrix multiplication, as follows:

[0098] ;

[0099] Where E represents the feature vector, W represents the weight matrix, and B is the bias term used to adjust the threshold;

[0100] A cross-term weight matrix is ​​introduced to enhance the interaction between features, explicitly modeling the second-order interaction between features, and improving the model's ability to express complex fault modes. The cross-term weight matrix is ​​specifically defined as follows:

[0101] ;

[0102] In this matrix, the first dimension represents the feature, corresponding to the feature vector E; the second dimension represents the fault type, corresponding to the fault type D; and the third dimension represents the interaction feature, also corresponding to the feature vector E. The matrix element W... cross (i,j,k) represents the joint contribution weight of the product of the i-th feature and the k-th feature to the j-th fault;

[0103] The initial assignment of the cross-weight matrix is ​​based on the physical mechanism, and the feature interaction pairs (f) are calculated. i ,f k ), representing the synergistic strength of feature combinations, where f i f represents the i-th feature in the feature vector E. k This represents the k-th feature in the feature vector E;

[0104] For each fault type d j Calculate the weighted contribution of all feature pairs using the following formula:

[0105] ;

[0106] Where i is the feature vector index value, j is the fault type index value, and k is the interaction feature index value;

[0107] Based on the calculated weighted contribution, the fault score is revised, using the following formula:

[0108] S enhanced =S+S cross ;

[0109] Where S is the fault score obtained from the initial calculation, S cross Weighted contribution;

[0110] The Sigmoid function is used to map the corrected fault scores to the [0,1] interval, representing the probability of each fault occurring. Specifically:

[0111] ;

[0112] Where e is a constant, and j represents the fault type S. enhanced (j) represents the score for the fault;

[0113] Set threshold θ j To determine if a fault exists, if fault P dj >θ j If the fault type is determined, then the fault type in the pipeline is identified, and a multimodal fault diagnosis result is generated.

[0114] Specifically, based on the flow parameter feature set and real-time data feature set, CFD features and real-time data features are fused to explore the correlation between features and faults. Seven features are obtained, including axial velocity decay index, proportion of pressure anomaly region, wall deposition thickness, kurtosis, impulse factor, thermal hysteresis coefficient, and mean phase difference, to construct a feature vector E=[f1,f2,f3,f4,f5,f6,f7], where f1 is the axial velocity decay index, and so on. Three fault types are defined: particle deposition, blockage, and leakage, to construct a fault type D=[d1,d2,d3], where d1 is the particle deposition fault, and so on.

[0115] A weight matrix W is constructed based on the feature vector and the fault type. The weight matrix is ​​specifically defined as follows:

[0116] ;

[0117] Wherein, the row index corresponds to feature f i The column index corresponds to the fault type d j element w ij Representing feature f i For fault d j The contribution weight.

[0118] Based on the physical mechanisms of characteristics and faults, the distribution of non-zero terms in the weight matrix is ​​defined. For example, particle deposition fault types are associated with axial velocity decay exponent, wall deposition thickness, and kurtosis characteristics; therefore, w 11 w 31 w 41 Each location will be assigned a weight. The type of blockage fault is related to the proportion of areas with abnormal pressure, the impulse factor, and the thermal hysteresis coefficient; therefore, w 22 w 52 w 62 Each location will be assigned a weight. The type of leakage fault is related to the mean phase difference and the thermal hysteresis coefficient, therefore w 63 w 73 The value will be assigned a weight. The unlisted w... ij If zero weight is set, it means that the feature has no effect on the fault.

[0119] Based on fluid dynamics principles and fault mechanism analysis, an initial contribution weight ratio is set. For particle deposition faults, a weight w is set. 11 :w 31 :w 41 =0.5:0.3:0.2; For congestion faults, set the weight w. 22 :w 52 :w 62 =0.4:0.4:0.2; For leakage faults, set the weight w. 63 :w 73=0.6:0.4.

[0120] Based on the feature vectors and weight matrices, the initial score for each fault is calculated using matrix multiplication. The specific formula is as follows:

[0121] ;

[0122] Where E represents the feature vector, W represents the weight matrix, and B is the bias term used to adjust the threshold. The calculated S=[s1,s2,s3] represents the score of each fault.

[0123] In fault diagnosis of pneumatic slag conveying pipelines, a single feature often fails to accurately reflect the physical nature of complex faults. For example, both velocity decay and wall deposition thickening can lead to particle deposition faults, but their combined effect may nonlinearly amplify the fault probability. The fault score obtained through the weight matrix can only characterize the independent effects of features and cannot capture the synergistic or antagonistic effects between features. Therefore, it is necessary to introduce a cross-term weight matrix to enhance the interaction between features, explicitly model the second-order interactions between features, and improve the model's ability to express complex fault modes.

[0124] A three-dimensional cross-weight matrix W is constructed based on the feature vector and the fault type. cross The cross-weight matrix is ​​specifically defined as follows:

[0125] ;

[0126] Wherein, the first dimension of the matrix represents the feature f i The corresponding feature vector E, with the second dimension representing the fault type d. j The third dimension represents the interaction feature f, corresponding to fault type D. k Similarly, corresponding to the eigenvector E, the matrix element W cross (i,j,k) represents feature f i with f k The product of the fault d j The joint contribution weight.

[0127] Initial assignment of cross-weight matrices is based on physical mechanisms. Initial weights are assigned to key interaction terms according to the physical correlation between features and faults. For example, W... cross (1,1,3) represents the combined contribution weight of the axial velocity decay exponent and wall deposition thickness to particle deposition failure, with its initial value set to 0.5. cross (2,2,5) represents the combined contribution weight of the abnormal pressure region and the impulse factor to the blockage fault, and its initial value is set to 0.4. For characteristics without significant interaction, their combined contribution weight is set to 0, such as W. cross(4,3,2)=0 indicates that the kurtosis value and the proportion of the pressure anomaly area have no direct interaction with the leakage fault.

[0128] Compute feature interaction pairs (f) i ,f k For each pair of features in the feature vector E (including combinations of features themselves), calculate their product. Characterizing the strength of the synergistic effect between the two, for example This represents the interaction between the axial velocity decay index and the wall deposition thickness. Its physical meaning is that the decrease in velocity and the increase in deposition thickness form a positive feedback, which significantly increases the probability of calculating particle deposition failure.

[0129] For each fault type d j Calculate the weighted contribution of all feature pairs using the following formula:

[0130] ;

[0131] Where i is the feature vector index value, j is the fault type index value, and k is the interaction feature index value.

[0132] Based on the calculated weighted contribution, the fault score is revised, using the following formula:

[0133] S enhanced =S+S cross ;

[0134] Where S is the fault score obtained from the initial calculation, S cross Weighted contribution.

[0135] The Sigmoid function is used to map the corrected fault scores to the [0,1] interval, representing the probability of each fault occurring. Specifically:

[0136] ;

[0137] Where e is a constant, j represents the fault type, and S enhanced (j) represents the score for the fault.

[0138] Set threshold θ j Determine if a fault exists; if fault P... dj >θ j If this condition is met, it is determined that a fault of this type has occurred in the pipeline. Multiple faults are allowed to be met simultaneously. For example, if θ1 = 0.5, and P... d1 If the value is 0.8, then a particle deposition fault is determined to exist.

[0139] Furthermore, step S6 of this application also includes:

[0140] The multimodal fault diagnosis and judgment results are obtained, and the fault types are classified to generate corresponding adaptive control strategies for slag conveying pipelines.

[0141] To address particulate deposition issues, the system employs methods such as increasing airflow velocity and using high-frequency pulses to strip away surface deposits, deploying guide vanes to enhance turbulence, and injecting fluidizing agents to decompose stubborn deposits, thereby achieving graded treatment and dynamic regulation.

[0142] To address blockage issues, a three-stage high-pressure air jet is used to break up the blockage, combined with mechanical cutting by a pig and heat treatment to soften the molten material. This multi-mode unblocking strategy achieves rapid unblocking while real-time monitoring prevents damage to the pipe body.

[0143] In response to the leak, a tiered pressure reduction system was implemented based on precise location to control the spread, and chemical materials were applied in stages to seal the cracks, ensuring rapid containment of the leak and structural restoration.

[0144] Specifically, the fault types identified through multimodal fault diagnosis are obtained, and the real-time pipeline dataset and flow parameter distribution model are backtracked to locate the fault section based on spatial information for targeted adaptive adjustments.

[0145] When the system detects a particle deposition fault, it will initiate a tiered treatment procedure. First, the control center automatically increases the speed of the upstream air supply fan, raising the airflow velocity in the fault area to 1.3-1.5 times the design value within 30 seconds. This enhanced fluid shear force helps to peel away loose surface deposits. Simultaneously, a high-frequency pulse valve assembly is activated 2 meters upstream of the deposition area, releasing compressed air pulses of 0.5-0.8 MPa at a frequency of 10-15 Hz. This creates periodic pressure oscillations, disrupting the adhesion between particles and the pipe wall. If the deposit volume does not decrease by more than 30% within 20 minutes, the system will switch to deep treatment mode. At this time, the built-in guide vane array in the pipe automatically deploys to a 45° angle, guiding the main airflow to generate controllable vortices and enhancing local turbulence intensity. In conjunction with this measure, distributed atomizing nozzles inject nano-scale fluidizing agents (such as modified silica) at a rate of 0.2 L / min. The surface-active components of these agents reduce van der Waals forces between particles, promoting deagglomeration of the deposit layer. During this stage, the CFD model monitors the changes in the concentration field in real time and dynamically adjusts the amount of fluidizing agent and the airflow pulsation parameters to ensure an optimal balance between stripping efficiency and energy consumption.

[0146] When a blockage is detected, the intelligent gate valves at both ends of the faulty pipe section close within 0.5 seconds, forming a sealed isolation chamber. The unblocking system then injects 6-8 bar high-pressure gas from the downstream side, implementing a three-stage pulse program for physical impact: an initial 0.5-second transient high-pressure impact aims to break up the blockage's skeletal structure; a 2-second medium-pressure airflow moves the fragments; and a final 5-second low-pressure long-duration purging removes residual particles. The entire process is monitored in real-time by pressure sensors. If the pressure differential does not decrease by 40% after three cycles, a secondary response is triggered. In the secondary response stage, a self-propelled pig (PIG) ​​travels along the inner wall of the pipeline to the blockage point. Its millimeter-wave radar accurately identifies the density distribution of the blockage and automatically switches operating modes: a spiral propulsion head physically pushes loose deposits; for hardened, compacted blocks, a rotary cutting module with diamond blades is activated, reaching speeds of up to 3000 rpm. During the unblocking process, vibration sensors are deployed outside the pipe to monitor vibration characteristics in real time. Operation is immediately suspended upon detecting any abnormal metallic friction noise to prevent damage to the pipe. For high-temperature molten blockage, the system activates a ring-shaped induction heating device to raise the pipe wall temperature to the material's softening point (approximately 300°C for coal ash) at a rate of 5°C / min. Simultaneously, nitrogen is introduced to inertize the environment and inhibit oxidation. Temperature sensors monitor the temperature field distribution to ensure uniform heating. Once the blockage exhibits rheological properties, high-pressure gas purging is immediately implemented, combined with a vacuum adsorption system to recover the molten material.

[0147] When a leak is detected, the system first determines the leak coordinates, and the pressure control module then implements a graded pressure reduction: the upstream gas supply pressure decreases in steps at a rate of 5% per minute, while an ejector-type negative pressure generator is activated 0.5 meters downstream of the leak point to create a local pressure gradient and suppress leak diffusion. During this stage, the pressure propagation path is predicted through real-time CFD simulation, and adjustment parameters are dynamically optimized. Based on pressure control, the intelligent sealing system implements chemical repair in three stages: First, a pneumatic delivery device releases 1-3mm porous epoxy resin particles at a rate of 20g / s. These particles are embedded in cracks under the influence of the leaking airflow, forming a mesh-like framework. Subsequently, a low-viscosity polyurethane prepolymer is injected through an atomizing nozzle, which penetrates deep into the micro-cracks through capillary action. Finally, a humidity-responsive curing agent is activated, completing the cross-linking reaction within 30 seconds to form an elastic seal. Throughout the process, distributed fiber optic sensors monitor the stress and strain state of the sealing layer in real time to ensure structural integrity.

[0148] Following fault recovery, the system enters an enhanced monitoring phase: During the first week, the monitoring frequency is increased to three times the normal level, and the repair effectiveness is evaluated through vibration modal analysis and thermodynamic parameter tracking. Simultaneously, preventative maintenance procedures are initiated, including monthly preventative flushing and quarterly self-healing coating maintenance. The predictive maintenance module analyzes historical operational data using machine learning to build degradation models for each pipe section. When the predicted remaining lifespan falls below a safe threshold, the system automatically generates a spare parts procurement list and maintenance work order, and pre-deploys mobile repair units to designated locations. This proactive health management strategy can reduce unexpected downtime by more than 60%, significantly improving system reliability.

[0149] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0150] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A CFD-based numerical adaptive control method for long-distance slag conveying, characterized in that, The method includes: Deploy sensor clusters at core monitoring locations of slag conveying pipelines to collect real-time pipeline transportation data and construct a real-time pipeline dataset. A three-dimensional physical model is constructed based on the geometry of the slag conveying pipeline. The computational domain of the three-dimensional physical model is determined. The computational domain is discretized using a mesh generation method to generate a computational mesh. A CFD dynamic simulation system for pneumatic conveying pipelines is constructed by combining mathematical models and dynamic boundary conditions. The mathematical model includes governing equations and a turbulence model, and the dynamic boundary conditions include inlet boundary conditions, outlet boundary conditions, and wall boundary conditions. The CFD dynamic simulation system is solved to obtain the flow parameter distribution model in the slag conveying pipeline. The flow parameter distribution model includes the velocity field, pressure field, and particle concentration field. Feature extraction is performed on the real-time pipeline dataset and the flow parameter distribution model. The correlation between features and fault types is mined using weight matrices and cross-term weight matrices to obtain multimodal fault diagnosis results. Obtaining multimodal fault diagnosis results also includes: The real-time dataset of the pipeline and the flow parameter distribution model are obtained, and the two are spatiotemporally aligned. The CFD transient simulation time step and the sensor cluster sampling frequency are matched, and the physical installation location of the sensor is mapped to the CFD grid node. We extract features from the spatiotemporally aligned flow parameter distribution model and the real-time pipeline dataset as the raw data to construct a flow parameter feature set and a real-time data feature set. By integrating the flow parameter feature set and the real-time data feature set, the correlation between the two and the fault is explored, a matrix operation model is constructed, and the multimodal fault diagnosis and judgment results are obtained. The process of constructing a matrix operation model to obtain multimodal fault diagnosis results also includes: Based on the flow parameter feature set and real-time data feature set, a feature vector E and a fault type D are constructed, and then a weight matrix W is generated. The weight matrix W is specifically defined as follows: ； Wherein, the matrix row index corresponds to the feature, the column index corresponds to the fault type, and the element w ij This represents the contribution weight of the i-th feature to the j-th fault type, where R represents the set of real numbers, a common representation in matrices. Based on the physical mechanisms of the characteristics and faults, the distribution of non-zero terms in the weight matrix and the initial contribution weight ratio are set. Based on the feature vector E and the weight matrix W, the initial score for each fault is calculated using matrix multiplication, as follows: ； Where E represents the feature vector, W represents the weight matrix, and B is the bias term used to adjust the threshold; A cross-term weight matrix is ​​introduced to enhance the interaction between features, explicitly modeling the second-order interaction between features, and improving the model's ability to express complex fault modes. The cross-term weight matrix is ​​specifically defined as follows: ； In this matrix, the first dimension represents the feature, corresponding to the feature vector E; the second dimension represents the fault type, corresponding to the fault type D; and the third dimension represents the interaction feature, also corresponding to the feature vector E. The matrix element W... cross (i,j,k) represents the joint contribution weight of the product of the i-th feature and the k-th feature to the j-th fault; The initial assignment of the cross-weight matrix is ​​based on the physical mechanism, and the feature interaction pairs (f) are calculated. i ,f k ), representing the synergistic strength of feature combinations, where f i f represents the i-th feature in the feature vector E. k This represents the k-th feature in the feature vector E; For each fault type d j Calculate the weighted contribution of all feature pairs using the following formula: ； Where i is the feature vector index value, j is the fault type index value, and k is the interaction feature index value; Based on the calculated weighted contribution, the fault score is revised, using the following formula: S enhanced =S+S cross ; Where S is the fault score obtained from the initial calculation, S cross Weighted contribution; The Sigmoid function is used to map the corrected fault scores to the [0,1] interval, representing the probability of each fault occurring. Specifically: ； Where e is a constant, j represents the fault type, and S enhanced (j) represents the score for the fault; Set threshold θ j To determine if a fault exists, if fault P dj >θ j If the fault type is found, it is determined that the fault type has occurred in the pipeline, and a multimodal fault diagnosis result is generated. The multimodal fault diagnosis results are analyzed, each fault type is classified and processed, and an adaptive control strategy for the slag conveying pipeline is generated.

2. The CFD-based numerical adaptive control method for long-distance slag conveying as described in claim 1, characterized in that, Construct a real-time dataset for the pipeline, including: The easily clogged areas, leakage-sensitive sections, and routine detection points of the slag conveying pipeline are designated as the core monitoring locations. The easily clogged areas include bends, diameter changes, and vertical sections. The leakage-sensitive sections are welded joints. The routine detection points are evenly marked every 5 meters along the pipeline, starting from the pipeline inlet and ending at the pipeline outlet. If a routine detection point is less than 1 meter away from the easily clogged area, leakage-sensitive section, or pipeline outlet, that point is removed. A sensor cluster is installed at the core monitoring location of the slag conveying pipeline. The sensor cluster includes pressure sensors, vibration sensors, temperature sensors, mass flow sensors, and velocity sensors. The sensor cluster continuously collects pressure, vibration, temperature, mass flow rate, and velocity data during the operation of the slag conveying pipeline to construct a real-time pipeline dataset. The real-time pipeline dataset uses timestamps as row indexes and spatial and sensor composite labels as column indexes.

3. The CFD-based numerical adaptive control method for long-distance slag conveying as described in claim 1, characterized in that, Generate a computational grid, including: A three-dimensional physical model is constructed based on the actual geometry of the pipeline, the computational domain is defined, and boundary conditions are predefined. An unstructured mesh is used to divide the interior of the pipe, and global mesh parameters are defined, including global mesh size and mesh growth rate. Mark the encryption control domain inside the pipeline, and perform local mesh encryption in the encryption control domain.

4. The CFD-based numerical adaptive control method for long-distance slag conveying as described in claim 1, characterized in that, A CFD dynamic simulation system for pneumatic conveying pipelines is constructed by combining multiphysics mathematical models and dynamic boundary conditions, including: The mathematical model is selected based on the type of material transported in the pipeline. The mathematical model includes a governing equation and a turbulence model. The governing equation adopts the discrete phase equation and selects the Eulerian-Lagrange framework to describe the exchange of momentum, mass and energy between the gas and solid phases. The turbulence model adopts the RNG k-ε model. Set dynamic boundary conditions for the model based on real-time datasets.

5. The CFD-based numerical adaptive control method for long-distance slag conveying as described in claim 1, characterized in that, Obtain the flow parameter distribution model within the slag conveying pipeline, including: Set the solver parameters, adopt the transient solution mode, set the time step, select the pressure-velocity coupling algorithm, and balance the calculation speed and stability. Discrete schemes are defined: the momentum equation adopts a second-order upwind scheme, the turbulence equation adopts a first-order scheme, and the particle phase equation adopts explicit time integration to ensure trajectory tracking accuracy. The flow parameter distribution model is iteratively calculated and output, which includes the velocity field, pressure field, and particle concentration field.

6. The CFD-based numerical adaptive control method for long-distance slag conveying as described in claim 1, characterized in that, Constructing a flow parameter feature set and a real-time data feature set, including: Feature extraction is performed on the flow parameter distribution model to obtain velocity field features, pressure field features, and particle concentration field features, and a flow parameter feature set is constructed. The velocity field feature is the axial velocity decay index, the pressure field feature is the proportion of pressure anomaly region, and the particle concentration field feature is the wall deposition thickness. Feature extraction is performed on the real-time pipeline dataset to obtain vibration data features, temperature-mass-flow data coupling features, and pressure-flow data coupling features, and a real-time data feature set is constructed. The vibration data features include kurtosis value and impulse factor, the temperature-mass-flow data coupling feature is thermal hysteresis coefficient, and the pressure-flow data coupling feature is mean phase difference.

7. The CFD-based numerical adaptive control method for long-distance slag conveying as described in claim 1, characterized in that, Generate an adaptive control strategy for the slag conveying pipeline, including: The multimodal fault diagnosis and judgment results are obtained, and the fault types are classified to generate corresponding adaptive control strategies for slag conveying pipelines. To address particulate deposition issues, the system employs methods such as increasing airflow velocity and using high-frequency pulses to strip away surface deposits, deploying guide vanes to enhance turbulence, and injecting fluidizing agents to decompose stubborn deposits, thereby achieving graded treatment and dynamic regulation. To address blockage issues, a three-stage high-pressure air jet is used to break up the blockage, combined with mechanical cutting by a pig and heat treatment to soften the molten material. This multi-mode unblocking strategy achieves rapid unblocking while real-time monitoring prevents damage to the pipe body. In response to the leak, a tiered pressure reduction system was implemented based on precise location to control the spread, and chemical materials were applied in stages to seal the cracks, ensuring rapid containment of the leak and structural restoration.

Images