Curve chute material flow accurate control method based on DEM modeling

Abstract

The invention discloses a curve chute material flow accurate control method based on DEM modeling, and the method comprises the following steps: S1, carrying out high-precision DEM modeling, building a non-spherical particle model based on actual material characteristics, introducing a power model, and accurately simulating a gas-solid coupling effect; s2, flow state prediction and optimization: analyzing speed distribution, segregation phenomenon and accumulation risk of materials in the curve chute under different working conditions through DEM simulation, establishing flow state evaluation indexes, and optimizing chute structure parameters based on simulation results; and S3, the intelligent control module controls the execution mechanism to output actions, sensors are arranged at key positions of the chute, real-time data are input into the DEM model for rapid simulation, short-term flowing behaviors are predicted, and the material flow is dynamically adjusted through the execution mechanism. According to the method, through non-spherical particle modeling, electric power model introduction and contact parameter calibration, the reduction degree of DEM simulation on actual material flow behaviors is improved, and the method is suitable for complex materials such as high-humidity materials and superfine powder materials.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of bulk material conveying, and particularly relates to a curve chute material flow precision control method based on DEM modeling. BACKGROUND

[0002] As a key component in the bulk material conveying system, the curve chute is widely used in the mining, metallurgy, chemical industry, grain storage and transportation industries, and its rationality of design directly affects the material conveying efficiency, equipment life and operation safety. In the actual operation process, the material in the curve section is prone to uneven flow rate, segregation, accumulation and even blockage, especially when dealing with complex materials with high humidity, high viscosity and multi-particle size distribution, the traditional design method based on experience is difficult to accurately predict the flow behavior, resulting in conservative design of the chute structure and lagging operation control.

[0003] Currently, the optimization of chute material flow mainly depends on physical tests and empirical formulas, which has problems such as long cycle, high cost and difficulty in fully covering complex working conditions. Although some individual studies have attempted to use the discrete element method (DEM) for simulation analysis, most of them are still based on the assumption of spherical particles, and the non-spherical characteristics of actual materials and the gas-solid coupling effect are not fully considered, resulting in limited simulation accuracy. In addition, the existing control methods are mostly based on threshold alarm and passive response, lacking real-time prediction and active control ability of the flow state, making it difficult to achieve precise and stable control of material flow.

[0004] Therefore, there is an urgent need for a systematic method that can integrate high-precision modeling, multi-condition simulation optimization and intelligent real-time control to improve the performance and reliability of the curve chute in the design and operation stages. SUMMARY

[0005] The application provides a curve chute material flow precision control method based on DEM modeling, which aims to solve the problems of limited simulation accuracy and difficulty in achieving precise and stable control of material flow in the prior art.

[0006] A curve chute material flow precision control method based on DEM modeling, comprising the following steps:

[0007] S1: High-precision DEM modeling, a non-spherical particle model is established based on the actual material characteristics, an electric power model is introduced, and the gas-solid coupling effect is accurately simulated;

[0008] S2: Flow state prediction and optimization, the velocity distribution, segregation phenomenon and accumulation risk of the material in the curve chute under different working conditions are analyzed through DEM simulation, the flow state evaluation index is established, and the chute structure parameters are optimized based on the simulation results;

[0009] S3: The intelligent control module controls the actuator to output actions, sensors are arranged at key positions of the chute, real-time data is input into the DEM model for rapid simulation, short-term flow behavior is predicted, and the material flow is dynamically adjusted through the actuator.

[0010] Optionally, the S1 specifically comprises:

[0011] S1.1: Material sampling and particle size analysis, particle size analysis is performed on the sample, complete particle size distribution data is obtained, and a normal distribution function is fitted;

[0012] S1.2: Particle shape characterization, a large number of particle images are taken, the images are subjected to grayscale and binaryzation processing, and the following parameters of a single particle are automatically calculated using image processing algorithms: aspect ratio, circularity, sphericity, and the average value and standard deviation of the above parameters are calculated as quantitative indicators of particle shape;

[0013] S1.3: Humidity determination and viscosity force calibration, the oven drying method is used to measure the moisture content and viscosity of the material;

[0014] S1.4: Non-spherical particle model construction, the multi-sphere cluster method is selected, a plurality of spheres are used to simulate the shape of a real particle, and a cluster particle template is created;

[0015] S1.5: Intrinsic parameter setting, the parameters of the particle material include: density, Poisson's ratio, and shear modulus;

[0016] The parameters of the geometric body material include: setting the density, Poisson's ratio, and Young's modulus according to the actual material;

[0017] S1.6: Contact parameter calibration;

[0018] S1.7: Special force model introduction, if the simulation involves gas-solid two-phase flow, the Ganser-DiFelice electric force model is selected, and if ultrafine powder is processed, the van der Waals force model is selected

[0019] S1.8: Establishing an accurate three-dimensional model and importing the model into DEM;

[0020] S1.9: Verification of grid independence in calculation.

[0021] Optionally, the step of non-spherical particle model construction comprises:

[0022] S1.4.1: Particle contour extraction and simplification, using the binaryzation image processed in the material characteristic representation, the clear contour of a single typical particle is extracted, and the complex particle contour is simplified into a polygon composed of multiple key points using an image algorithm;

[0023] S1.4.2: Skeleton line guided sphere filling method, a non-spherical particle is determined by a plurality of overlapping spheres;

[0024] S1.4.3: Output the cluster model parameters, the final output of each sphere parameters: sphere center coordinates (x, y, z) and radius r, parameter input simulation software API, automatic generation of particle template.

[0025] Optionally, the skeleton line guided sphere filling method comprises the following steps:

[0026] S1.4.2.1: Input the binary image of a single particle obtained in the material characteristic representation;

[0027] S1.4.2.2: Simplify the contour polygon using the Douglas-Peucker algorithm;

[0028] S1.4.2.3: Extract the skeleton line from the simplified contour using the morphological thinning algorithm;

[0029] S1.4.2.4: Identify key points, analyze the skeleton line, and automatically identify three types of key points: end points, intersection points, and high curvature points. Between these key points, additional points are inserted at a fixed step length;

[0030] S1.4.2.5: Take the coordinates of all identified and inserted points as the candidate sphere center set C = {C1, C2,..., C n};

[0031] S1.4.2.6: Calculate the maximum inscribed sphere radius, for each candidate sphere center C i , calculate its shortest distance d i to the contour boundary, the shortest distance d i is the maximum inscribed sphere radius R i with C i as the sphere center;

[0032] S1.4.2.7: Initialize an empty set Clusters to store the final determined spheres, remove all candidate sphere centers contained in the newly added spheres more than the threshold proportion from the candidate set C, repeat the above process until the candidate set C is empty or all contour regions are fully covered;

[0033] S1.4.2.8: Remove redundant spheres;

[0034] S1.4.2.9: Verification and accuracy evaluation, calculate the coincidence degree of the contour reconstructed by the multi-sphere cluster with the original contour;

[0035] If the accuracy does not meet the requirements, return to step two, generate more candidate sphere centers, and re-fill until the accuracy requirements are met.

[0036] Optionally, S2 specifically comprises:

[0037] S2.1: Define the simulation condition matrix, define a simulation parameter matrix covering all possible cases based on the fluctuation range in actual production;

[0038] S2.2: Perform batch simulation and data acquisition, use batch processing function to automatically run DEM simulation of all preset conditions, set virtual monitoring points, monitoring lines or monitoring surfaces in key areas of the chute;

[0039] S2.3: Extract and quantify flow state indicators, post-process simulation results to extract quantitative indicators:

[0040] S2.4: Build a response surface model to correlate simulation results with input parameters;

[0041] S2.5: Multi-objective optimization and Pareto frontier search, define optimization objectives and constraints;

[0042] S2.6: Propose improvement schemes based on optimization results, select the optimal configuration from the Pareto optimal solution set, and convert it into a specific chute structure improvement scheme;

[0043] S2.7: Final verification simulation, re-import the optimized chute 3D model into the DEM software and perform final verification simulation under extreme conditions.

[0044] Optionally, the intelligent control module includes a sensor network and a digital twin platform.

[0045] Optionally, S3 specifically comprises the following steps:

[0046] S3.1: Sensor network deployment, deploy multiple sensors in key areas of the curved chute to build a real-time data acquisition network;

[0047] S3.2: Data fusion and state recognition, filter and fuse all sensor data, use image recognition algorithm to analyze camera images in real time, determine whether there are "holes" or "accumulations", calculate the average value and gradient of the pressure film sensor, and if the pressure continues to rise, trigger a jam warning.

[0048] Optionally, the specific steps of the image recognition algorithm analysis include:

[0049] S3.2.1: Data preparation and model training, collect a large number of chute image and video data covering different conditions, label the abnormal areas in the image, and the label categories are: accumulation, hole, normal;

[0050] Then pre-process and enhance the data, select YOLOv8s target detection model, frame the abnormal area and classify;

[0051] S3.2.2: Model deployment and real-time inference, deploying the trained model file, using a fixed frequency to capture image frames from the video stream, inputting the captured image frames into the deployed model, subsequently obtaining the output results of the model, applying non-maximum suppression to eliminate overlapping redundant detection boxes, and obtaining the final prediction results;

[0052] S3.2.3: Result analysis and control linkage, according to the prediction results of the model and the set confidence threshold, logical judgment is performed, and an alarm signal is generated, and the generated alarm signal is published to the digital twin platform;

[0053] The intelligent control module receives the early warning signal, and triggers the preset control logic to the actuator immediately;

[0054] S3.2.4: Negative sample feedback and model iteration, establishing a false positive sample library.

[0055] Optionally, the digital twin platform comprises a real-time visualization and remote monitoring module, and a fault diagnosis and predictive maintenance module, an embedded expert knowledge base, and an adaptive control optimization module.

[0056] Compared with the prior art, the present application has at least the following beneficial effects:

[0057] The present application improves the restoration degree of DEM simulation to the actual material flow behavior through non-spherical particle modeling, power model introduction and contact parameter calibration, is suitable for complex materials such as high humidity and ultra-fine powder, evaluates the flow performance of the chute under different operation and structure parameters through a multi-working-condition simulation matrix and a response surface modeling, and in combination with a multi-objective optimization algorithm, can quickly identify an optimal chute structure configuration, and improves design efficiency and scientificity.

[0058] The present application also relies on a sensing network and a digital twin platform to realize real-time identification and prediction of material flow states, early warning of abnormal states such as blockage and surging through multi-source data fusion such as image recognition and pressure monitoring, and linkage of an actuator for dynamic adjustment, and realizes preventive control. BRIEF DESCRIPTION OF DRAWINGS

[0059] Figure 1 A flowchart of a precise control method for material flow in a curved chute based on DEM modeling is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application is further described in detail below with reference to the drawings and embodiments.

[0061] The application provides a curve chute material flow precision control method based on DEM modeling, which comprises the following steps:

[0062] S1: high-precision DEM modeling:

[0063] Based on the actual material characteristics (particle size distribution, shape, humidity), a non-spherical particle model (using the multi-sphere cluster method) is established, the Ganser-DiFelice electric power model is introduced, the gas-solid coupling effect is accurately simulated, the chute wall friction, collision restitution coefficient and other parameters are considered, and the grid independence verification is performed;

[0064] Among them, S1 specifically includes:

[0065] S1.1: material sampling and particle size analysis, representative samples are extracted from the curve chute conveying material in the actual industrial field, standard sieving or laser particle size analyzer (such as Malvern Mastersizer) is used for particle size analysis of the samples, complete particle size distribution (PSD) data is obtained, and normal distribution function is fitted for defining the particle factory in DEM;

[0066] S1.2: particle shape characterization, image acquisition: a high-resolution industrial camera (such as EV-E1600C mentioned in the text) is used to shoot a large number of particle images in a static environment, the images are subjected to grayscale and binary processing, the influence of adhered particles is eliminated, and the following parameters of a single particle are automatically calculated by using image processing algorithm (such as based on OpenCV): aspect ratio, circularity, sphericity

[0067] The average value and standard deviation of these parameters are calculated as the quantitative indicators of particle shape;

[0068] S1.3: humidity determination and viscosity force calibration, the oven drying method is used to measure the moisture content of the material, and the viscosity force of high-moisture material (such as high-moisture lignite) is the key to cause blockage, the liquid bridge force model is used to simulate the inter-particle adhesion force caused by water, the contact cohesion force and internal friction angle between particles under different moisture contents are back calculated by using the powder shear tester or direct shear experiment, and the contact model parameters in DEM are calibrated;

[0069] S1.4: non-spherical particle model construction, the multi-sphere cluster method is selected, a plurality of overlapping or non-overlapping spheres are used to approximate the real particle shape, and a cluster particle template is directly created in EDEM, RockyDEM and other software;

[0070] The steps of non-spherical particle model construction include:

[0071] S1.4.1: Particle contour extraction and simplification, using the binarized image after processing in material property characterization, extract the clear contour of a single typical particle, use image algorithms (such as edge detection, polygon fitting) to simplify the complex particle contour into a polygon composed of multiple key points, to reduce the complexity of subsequent modeling;

[0072] S1.4.2: Skeleton-guided sphere packing method, a non-spherical particle is uniquely determined by a plurality of overlapping spheres;

[0073] The skeleton-guided sphere packing method includes the following steps:

[0074] S1.4.2.1: The input is the binarized image of a single particle obtained in material property characterization;

[0075] S1.4.2.2: Simplify the contour polygon using the Douglas-Peucker algorithm, reduce the number of vertices while retaining the main shape features, and reduce the computational complexity;

[0076] S1.4.2.3: Use a morphological thinning algorithm to extract the skeleton line from the simplified contour, the skeleton line is one or more lines inside the contour, representing the topological structure and geometric center of the figure, for long strip-shaped particles, the skeleton line may be a main line; for irregular particles, it may be a branch structure;

[0077] S1.4.2.4: Identify key points, analyze the skeleton line and automatically identify three types of key points: end points, end points of the skeleton line; intersection points, intersection points of the skeleton line branches; high curvature points, points along the skeleton line whose curvature changes exceed a certain threshold;

[0078] Between these key points, additional points are inserted at a fixed step length to ensure that shape details are retained;

[0079] S1.4.2.5: Take the coordinates of all identified and inserted points as the set of candidate sphere centers C = {C1, C2,..., C n};

[0080] S1.4.2.6: Calculate the maximum inscribed sphere radius, for each candidate sphere center C i , calculate its shortest distance d i to the contour boundary, the shortest distance d i is the radius R i of the largest inscribed sphere that can be placed with C i as the sphere center;

[0081] In 2D, this is a point-to-polygon shortest distance calculation; in 3D, it is a point-to-surface shortest distance calculation;

[0082] S1.4.2.7: Initialize an empty set Clusters to store the final determined spheres, with radius R i Sort the candidate sphere centers in descending order of their current radius, select the candidate sphere C with the largest radius max , and add it to the empty set Clusters. max , R max .

[0083] Remove all candidate sphere centers that are contained in the newly added sphere more than a certain percentage (e.g. 95%) from the candidate set C, effectively avoiding overlap and reducing the number of spheres, repeat the above process until the candidate set C is empty or all contour regions have been sufficiently covered;

[0084] S1.4.2.8: Remove redundant spheres, iterate through the Clusters set, if a sphere S i is more than 99% covered by other larger spheres, remove S i from the set, further optimize the model, eliminate internal spheres that contribute little to the shape;

[0085] S1.4.2.9: Verification and accuracy evaluation, calculate the overlap between the contour reconstructed by the multi-sphere cluster and the original contour. A commonly used indicator is the Hausdorff distance, which measures the maximum mismatch between two shapes;

[0086] If the accuracy does not meet the requirements (e.g. Hausdorff distance greater than 5% of the particle size), return to step two, reduce the sampling step or increase the identification sensitivity of high curvature points, generate more candidate sphere centers, re-fill, until the accuracy requirements are met;

[0087] S1.4.3: Output cluster model parameters, finally output the parameters of each sphere: sphere center coordinates (x, y, z) and radius r, these parameters are directly input into the API of EDEM, RockyDEM and other software, automatically generating particle templates;

[0088] S1.5: Intrinsic parameter setting, the parameters of the particle material include: density (determined by specific gravity bottle method), Poisson's ratio, shear modulus (obtained by nanoindentation experiment);

[0089] The parameters of the geometric body material (chute wall) include: setting its density, Poisson's ratio, Young's modulus according to the actual material;

[0090] S1.6: Contact parameter calibration, particle-particle contact: Set restitution coefficient, static friction coefficient, rolling friction coefficient, initial values of these parameters are calibrated by Angle of Repose Test, simulate the Angle of Repose Test in DEM, constantly adjust the contact parameters until the simulated Angle of Repose and the Angle of Repose measured in the real experiment have an error of less than 5%.

[0091] Particle-wall contact: The same method as particle-particle contact is selected, and the static / dynamic friction coefficient between the particle and the wall is calibrated by the inclined plane sliding experiment or the chute experiment;

[0092] S1.7: Introduction of special force model:

[0093] Ganser-DiFelice electric force model: If the simulation involves gas-solid two-phase flow (such as the presence of dust removal wind or air cushion), in the CFD-DEM coupling simulation, abandon the default spherical particle electric force model, select or customize the DiFelice-Ganser model. This model modifies the electric force of non-spherical particles by introducing the sphericity φ and Ganser coefficients (K1, K2), which significantly improves the accuracy of gas-solid coupling simulation;

[0094] Van der Waals force model: If dealing with micron or nanometer ultra-fine powder, the Van der Waals force option needs to be activated in the contact model, and the Hamaker constant needs to be correctly set to simulate the strong agglomeration effect between particles;

[0095] S1.8: High-precision chute three-dimensional model, use a three-dimensional scanner to scan the actual curved chute, or establish a 1:1 accurate three-dimensional model in CAD software according to engineering drawings, pay special attention to details such as interface parts, bend curvature, baffle shape, etc., import the model into the DEM software in STEP or IGES format;

[0096] S1.9: Verification of calculation grid independence, used to ensure that the simulation results are not dependent on the grid size, in CFD software (such as Fluent) or DEM software, divide a series of different size grids (usually 3-4 sets from coarse to fine) for the chute calculation domain, simulate under the same initial and boundary conditions, monitor a key output variable (such as outlet mass flow, average velocity at a specific point, bin bottom pressure);

[0097] When the grid is continuously encrypted, the change amplitude of the key variable is less than 2-5%, and it is considered that the grid independence has been reached. At this time, the grid size and quantity used are the grid for the final simulation.

[0098] S2: Flow state prediction and optimization, through DEM simulation to analyze the velocity distribution, segregation phenomenon, and accumulation risk of material in the curved chute under different working conditions, establish flow state evaluation indexes (such as mixing index, velocity uniformity, and accumulation angle), and optimize the chute structure parameters (such as curvature radius, outlet shape, and baffle arrangement) based on the simulation results;

[0099] S2 specifically includes:

[0100] S2.1: Define the simulation condition matrix, based on the fluctuation range in actual production, define a simulation parameter matrix covering all possible cases;

[0101] The key variables of the simulation condition matrix include:

[0102] Operating parameters: feed rate (low, medium, high), moisture content of material (dry, normal humidity, high humidity).

[0103] Equipment parameters: as optimization variables, pre-set an initial range (such as chute inclination: 45°-55°; curvature radius: R1.5m-R2.5m).

[0104] Use the experimental design (DOE) method (such as orthogonal experiment method) to scientifically reduce the number of simulations while ensuring that the main influencing factors are covered;

[0105] S2.2: Perform batch simulation and data collection, use batch processing function to automatically run DEM simulation for all pre-set working conditions, set virtual monitoring points, monitoring lines or monitoring surfaces in key areas of the chute during simulation for data collection;

[0106] Key areas include:

[0107] Inlet area: monitor impact force and initial velocity;

[0108] Outside of the bend: monitor wear rate and particle velocity;

[0109] Inside the bend: monitor particle retention and accumulation risk;

[0110] Discharge port: monitor discharge velocity distribution and flow rate;

[0111] S2.3: Extract and quantify flow state indexes, post-process from simulation results to extract quantitative indexes for objective evaluation of flow performance:

[0112] Quantitative indexes include:

[0113] Velocity uniformity index (SVI): coefficient of variation (standard deviation / average value) of particle velocity on the discharge port monitoring surface, the lower the SVI, the more uniform the discharge, and the less likely to segregate;

[0114] Mixing Index (MI): The Lacey Mixing Index is used to quantitatively assess the degree of separation of different components at the discharge port if the conveyed material is a mixture. MI = 0 indicates complete separation, and MI = 1 indicates complete mixing.

[0115] Accumulation Risk Index (BRI): This is the percentage of "stationary" particles with a velocity close to 0 in monitoring zone C at the end of the simulation, out of the total number of particles. A higher BRI indicates a greater risk of clogging.

[0116] Wall wear potential: The average collision energy between particles and the wall within the statistical monitoring area B is used to predict wear hotspot areas;

[0117] Dynamic angle of repose: In simulation, the angle between the surface formed by the material as it naturally accumulates in the chute and the horizontal plane. This angle can indirectly reflect the flowability of the material;

[0118] S2.4: Construct a response surface model (RSM) and correlate simulation results (such as SVI, BRI, and wear potential as response variables) with input parameters (feed rate, moisture content, inclination angle, etc. as design variables).

[0119] Using the Kriging method, a high-precision surrogate model (Metamodel) is fitted and constructed. This model can predict the flow state under any set of parameters at an extremely fast speed (seconds) without the need for time-consuming DEM simulation.

[0120] S2.5: Multi-objective optimization and Pareto front search, defining optimization objectives (usually conflicting) and constraints;

[0121] For example:

[0122] Objective 1: Minimize the Bulk Accumulation Risk Index (BRI)

[0123] Objective 2: Maximize the reciprocal of the velocity uniformity index (SVI) (i.e., maximize uniformity).

[0124] Objective 3: Minimize wall wear potential;

[0125] Constraints: Equipment size limitations, processing capacity requirements.

[0126] Multi-objective optimization algorithms (such as NSGA-II) are used to quickly find optimal solutions on the surrogate model, thus finding a set of Pareto optimal solutions. These solutions represent the best trade-offs between different optimization objectives.

[0127] S2.6: Based on the optimization results, propose improvement schemes. From the Pareto optimal solution set, select one or more optimal configurations according to the actual production focus (such as more attention to anti-clogging or more attention to uniform discharge). Convert these optimal configurations into specific chute structure improvement schemes and select the curvature that can minimize centrifugal segregation and outer wear.

[0128] Design anti-segregation baffles: Design and simulate the installation of guide baffles on the outside of the bend to guide the material flow toward the center;

[0129] Alternatively, anti-accumulation baffles can be installed on the inside of the curve to disrupt the formation of the angle of repose. The effects of different baffle heights, angles, and positions can be verified through simulation.

[0130] Optimize the outlet shape: Change the flat outlet to a diffuser outlet or add guide fins to improve discharge uniformity (SVI).

[0131] Surface treatment recommendations: For high-wear areas, it is recommended to use special wear-resistant materials or coatings;

[0132] S2.7: Final verification simulation. The optimized chute 3D model is re-imported into the DEM software, and the final verification simulation is carried out under several extreme working conditions (such as the highest feed rate and the highest moisture content).

[0133] By comparing the improvement in key performance indicators (BRI, SVI) before and after optimization, the benefits of this optimization can be quantified.

[0134] S3: The intelligent control module controls the output action of the actuator. Sensors (pressure, speed, image) are placed at key positions in the chute to input real-time data into the DEM model for rapid simulation, predict short-term flow behavior, and dynamically adjust the material flow through actuators (such as pneumatic baffles, vibrators, water spray systems) to prevent material blockage or flow deviation.

[0135] S3 specifically includes the following steps:

[0136] S3.1: Sensor network deployment: Deploy multiple sensors in the key areas of the curved chute (sensitive points determined based on DEM analysis) to build a real-time data acquisition network;

[0137] In one specific embodiment, an impact flow meter or a weighing sensor is selected for the feed inlet to monitor the feed flow rate (Q_in) in real time, which serves as the main input and feedforward control signal of the system.

[0138] The chute body, especially at bends, is equipped with microwave / radar level gauges to non-contactly measure the material layer thickness (H_material) at specific points. It is also equipped with array-type pressure film sensors attached to the chute wall (especially the inner side) to directly measure the pressure distribution of particles on the wall, which is the most direct indicator for blockage early warning. In addition, it is equipped with high-definition industrial cameras for visual monitoring.

[0139] At the discharge port, a high-speed linear array camera is used to capture images of the material flow to analyze the discharge shape and uniformity; a lidar scanner is used to scan the dynamic angle of repose and profile of the discharge pile to assess flowability and segregation.

[0140] S3.2: Data fusion and status recognition. The industrial computer receives all sensor data, performs filtering and fusion processing, uses image recognition algorithms to analyze camera images in real time, and uses a trained model to determine whether there is a "hole" (indicating a surge) or "accumulation" (indicating a blockage). It calculates the average value and gradient of the pressure film sensor. If the pressure continues to rise (dP / dt>threshold), a blockage warning is triggered.

[0141] The specific steps of image recognition algorithm analysis include:

[0142] S3.2.1: Data preparation and model training: Collect a large amount of chute images and video data covering different working conditions on the installed industrial cameras;

[0143] It must specifically include: normal flow (meaning the material falls smoothly and continuously);

[0144] Accumulation: The accumulation of material in a chute (especially inside a bend or at the outlet) that is stationary or moving slowly.

[0145] Void / Surge: Voids are formed in the material, causing short circuits in airflow or material flow, or large surges.

[0146] Use annotation tools to add bounding boxes or pixel-level annotations to abnormal regions in the image. Annotation categories are: stacking, holes, and normal.

[0147] Subsequently, data preprocessing and enhancement were performed, image sizes were adjusted to meet model input requirements, grayscale or normalization operations were performed, and methods such as rotation, cropping, brightness / contrast adjustment, and addition of simulated dust noise were applied to expand the dataset size and improve the robustness of the model in complex industrial environments.

[0148] For model selection and training, the lightweight YOLOv8s object detection model was chosen. It can directly outline and classify abnormal regions, and is fast enough to meet real-time requirements. The labeled dataset was divided into training, validation and test sets in a ratio (e.g., 8:1:1). The model was trained on a GPU server, and the performance was optimized by adjusting hyperparameters. Finally, the optimal model weight file was obtained.

[0149] S3.2.2: Model Deployment and Real-time Inference. The trained model file is deployed on an edge computing device. An inference optimization framework is used to quantize and accelerate the model, further improving inference speed and reducing latency. The edge computing device reads the real-time video stream from an industrial camera and uses libraries such as OpenCV to capture image frames from the video stream at a fixed frequency. The captured image frames are input into the deployed accelerated model, and then the model's output results (classification label, confidence score, bounding box coordinates) are obtained. Non-maximum suppression (NMS) is applied to eliminate overlapping redundant detection boxes to obtain the final prediction result.

[0150] S3.2.3: Result analysis and control linkage: Based on the model's prediction results and the set confidence threshold (e.g., >0.8), perform logical judgments:

[0151] If accumulation is detected and confidence level > 0.8 THEN, a blockage warning signal is generated, and the approximate blockage area can be located using bounding box coordinates.

[0152] If a void is detected and the confidence level is >0.8, a surge warning signal is generated.

[0153] At the same time, a status counter is deployed. To avoid occasional false detections, it is set to trigger an alarm only if an anomaly is detected for 3 consecutive frames, thereby improving the system's anti-interference capability.

[0154] The generated alarm signals are published to the digital twin platform via MQTT or OPCUA protocols;

[0155] When the intelligent control module receives an early warning signal, it immediately triggers the preset control logic to the actuator (for example, upon receiving a blockage warning, it immediately starts the high-frequency pneumatic vibrator in the corresponding area) to achieve vision-based preventive control.

[0156] S3.2.4: Negative sample feedback and model iteration. Establish a false alarm sample library. When the system issues an alarm but the operator confirms that it is a false alarm, the image frames within that time period can be saved and marked as "normal". The model can be incrementally trained or fine-tuned periodically using the newly collected samples so that the model can continuously adapt to changes in the field environment.

[0157] The digital twin platform includes: a real-time visualization and remote monitoring module, which displays a 3D model of the equipment and overlays real-time data (such as material flow pattern, pressure distribution cloud map, velocity vector field) onto the 3D model in the form of color mapping, animation, etc., to achieve transparent monitoring of the working conditions;

[0158] The fault diagnosis and predictive maintenance module has a built-in expert knowledge base. For example, when "the pressure on the inside of the curve is consistently higher than the threshold X and continues to increase," a "blockage warning" is automatically triggered, and the risk area is highlighted on the 3D model.

[0159] By using historical operating data (especially fault data) to train machine learning models, more complex and potential fault modes can be identified. For example, by analyzing vibration spectrum characteristics, bearing wear conditions can be judged in advance, enabling predictive maintenance and avoiding unplanned downtime.

[0160] The adaptive control optimization module continuously compares the prediction results with the actual effects and has a self-learning mechanism: if it finds that the prediction of the surrogate model has a continuous deviation, it can automatically schedule the high-fidelity DEM model in the cloud, re-simulate based on the new working condition data, and update and calibrate the parameters of the surrogate model to make the next prediction more accurate.

[0161] 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.

Claims

1. A method for precise control of material flow in a curved chute based on DEM modeling, characterized in that, Includes the following steps: S1: High-precision DEM modeling, establishing a non-spherical particle model based on actual material properties, introducing an electric power model, and accurately simulating the gas-solid coupling effect; S2: Flow state prediction and optimization. Through DEM simulation analysis, the velocity distribution, segregation phenomenon and accumulation risk of materials in the curved chute under different working conditions are analyzed. Flow state evaluation index is established, and the chute structure parameters are optimized based on the simulation results. S3: The intelligent control module controls the output actions of the actuator, places sensors at key locations in the chute, inputs real-time data into the DEM model for rapid simulation, predicts short-term flow behavior, and dynamically adjusts the material flow through the actuator.

2. The method for precise control of material flow in a curved chute based on DEM modeling according to claim 1, characterized in that, S1 specifically includes: S1.1: Material sampling and particle size analysis. Perform particle size analysis on the sample to obtain complete particle size distribution data and fit it to a normal distribution function; S1.2: Particle shape characterization. A large number of particle images are captured, and the images are processed by grayscale and binarization. The following parameters of a single particle are automatically calculated using image processing algorithms: aspect ratio, roundness, and sphericity. The average value and standard deviation of the above parameters are calculated as quantitative indicators of particle shape. S1.3: Moisture content and viscosity calibration: The moisture content and viscosity of the material are measured using the oven drying method. S1.4: Construction of non-spherical particle model. Select the multi-sphere cluster method, use multiple spheres to simulate the shape of real particles, and create cluster particle templates; S1.5: Intrinsic parameter settings, the parameters of particulate materials include: density, Poisson's ratio, and shear modulus; The parameters of the geometric material include: density, Poisson's ratio, and Young's modulus, which are set according to the actual material. S1.6: Contact parameter calibration; S1.7: Special force models are introduced. If the simulation involves gas-solid two-phase flow, the Ganser-DiFelice electric power model is selected; if ultrafine powders are being processed, the van der Waals force model is selected. S1.8: Create an accurate 3D model and import the model into the DEM; S1.9: Verification of computational grid independence.

3. The method for precise control of material flow in a curved chute based on DEM modeling according to claim 2, characterized in that, The steps for constructing the non-spherical particle model include: S1.4.1: Particle contour extraction and simplification. Using the binarized image processed in the material property characterization, the clear contour of a single typical particle is extracted. The complex particle contour is simplified into a polygon composed of multiple key points using image algorithms. S1.4.2: Using the spherical filling method guided by the skeleton line, a non-spherical particle is determined by multiple overlapping spheres; S1.4.3: Output cluster model parameters. The final output parameters for each sphere are: center coordinates (x, y, z) and radius r. The parameters are input into the simulation software's API to automatically generate particle templates.

4. The method for precise control of material flow in a curved chute based on DEM modeling according to claim 3, characterized in that, The spherical filling method guided by the skeleton line includes the following steps: S1.4.2.1: The input is a binarized image of a single particle obtained from the material property characterization; S1.4.2.2: Use the Douglas-Peucker algorithm to simplify the outline polygon; S1.4.2.3: Employ a morphological thinning algorithm to extract the skeleton lines from the simplified contour; S1.4.2.4: Identify key points, analyze the skeleton lines, and automatically identify three types of key points: endpoints, intersections, and high curvature points. Insert additional points between these key points at fixed step sizes. S1.4.2.5: The coordinates of all identified and inserted points are used as the set C = {C1, C2, ..., C...} of candidate sphere centers. n }; S1.4.2.6: Calculate the maximum inscribed sphere radius for each candidate sphere center C. i Calculate the shortest distance d from the contour boundary. i The shortest distance d i That is, using C i The radius R of the largest inscribed sphere that can be placed at the center of the sphere. i ; S1.4.2.7: Initialize an empty set Clusters to store the finally determined spheres. Remove all candidate spheres whose newly added spheres contain a proportion exceeding the threshold from the candidate set C. Repeat the above process until the candidate set C is empty or all contour regions have been fully covered. S1.4.2.8: Remove redundant spheres; S1.4.2.9: Verification and accuracy evaluation, calculate the overlap between the contour reconstructed from the multi-sphere cluster and the original contour; If the accuracy requirement is not met, return to step two to generate more candidate sphere centers and refill until the accuracy requirement is met.

5. The method for precise control of material flow in a curved chute based on DEM modeling according to claim 1, characterized in that, S2 specifically includes: S2.1: Define the simulation condition matrix. Based on the fluctuation range in actual production, define a simulation parameter matrix. S2.2: Perform batch simulation and data acquisition. Utilize the batch processing function to automatically run DEM simulations for all preset working conditions and set virtual monitoring points, monitoring lines, or monitoring surfaces in key areas of the chute. S2.3: Extract and quantify flow state indices. Post-process the simulation results to extract and quantify the indices: S2.4: Construct a response surface model and correlate the simulation results with the input parameters; S2.5: Multi-objective optimization and Pareto front search, defining optimization objectives and constraints; S2.6: Based on the optimization results, propose an improvement scheme, select the optimal configuration from the Pareto optimal solution set, and convert it into a specific chute structure improvement scheme; S2.7: Final verification simulation. The optimized chute 3D model is re-imported into the DEM software, and a final verification simulation is performed under extreme working conditions.

6. The method for precise control of material flow in a curved chute based on DEM modeling according to claim 1, characterized in that, The intelligent control module includes a sensor network and a digital twin platform.

7. The method for precise control of material flow in a curved chute based on DEM modeling according to claim 1, characterized in that, S3 specifically includes the following steps: S3.1: Sensor network deployment: Deploy multiple sensors in key areas of the curved chute to build a real-time data acquisition network; S3.2: Data fusion and status recognition. All sensor data is filtered and fused. Image recognition algorithms are used to analyze camera images in real time to determine whether there are "holes" or "accumulations". The average value and gradient of the pressure film sensor are calculated. If the pressure continues to rise, a blockage warning is triggered.

8. The method for precise control of material flow in a curved chute based on DEM modeling according to claim 7, characterized in that, The specific steps of the image recognition algorithm analysis include: S3.2.1: Data preparation and model training: Collect a large amount of chute images and video data covering different working conditions, and label abnormal areas in the images. The labeling categories are: accumulation, holes, and normal. Subsequently, data preprocessing and enhancement were performed, and the YOLOv8s target detection model was selected to outline and classify abnormal regions. S3.2.2: Model Deployment and Real-time Inference. Deploy the trained model file, capture image frames from the video stream at a fixed frequency, input the captured image frames into the deployed model, obtain the model's output, apply non-maximum suppression to eliminate overlapping redundant detection boxes, and obtain the final prediction result. S3.2.3: Results analysis and control linkage: Based on the model's prediction results and the set confidence threshold, perform logical judgments and generate alarm signals, and publish the generated alarm signals to the digital twin platform; Upon receiving the warning signal, the intelligent control module immediately triggers the preset control logic to the actuator; S3.2.4: Negative sample feedback and model iteration to establish a false alarm sample library.

9. The method for precise control of material flow in a curved chute based on DEM modeling according to claim 8, characterized in that, The digital twin platform includes: a real-time visualization and remote monitoring module, a fault diagnosis and predictive maintenance module, a built-in expert knowledge base, and an adaptive control optimization module.

Images