A digital twin method and system for a mineral slurry dewatering process

Abstract

The present application belongs to the field of digital twinning, and discloses a kind of digital twinning method of mineral pulp dehydration process, comprising the following steps: real-time acquisition and control of mineral pulp dehydration process equipment data;Industrial test database is constructed;Real-time transmission of virtual mapping model data is established and visualization is realized;Establishment of equipment geometric macroscopic model hosting platform;Digital visualization and regulation and control.The present application also discloses a kind of digital twinning system of mineral pulp dehydration process, comprising industrial test database construction unit, virtual mapping model establishment and data real-time transmission and visualization unit, establishment of equipment geometric macroscopic model hosting platform unit, visualization and regulation and control unit.The present application solves the problem of insufficient filter cake dehydration, poor effect or dehydration failure in the filter press dehydration process in the existing mineral processing plant.The present application monitors the state of mineral pulp dehydration process such as mineral pulp water by various sensors monitoring equipment in coal preparation plant at all times and adjusts each equipment in real time, so that the related pressure dehydration operation of coal preparation plant is optimized.

Description

Technical Field

[0001] This invention belongs to the field of digital twins, specifically relating to a digital twin method for a slurry dewatering process. This invention also provides a digital twin method and system for a slurry dewatering process. Background Technology

[0002] As coal mining progresses, production from high-quality mines decreases, thinner coal seams increase, and changes in raw coal quality lead to increasingly prominent problems in slurry dewatering at coal preparation plants. Major coal preparation plants and other slurry treatment plants are focusing on the impact of slurry dewatering capacity on the plant's economic benefits. The formula for maximizing the economic benefit of slurry dewatering is:

[0003]

[0004] In the formula, E: maximum economic benefit (unit: yuan); Q: the amount of slurry processed after dewatering (unit: tons), that is, the total amount of slurry before dewatering minus the residual amount after dewatering; P: the market price per ton of slurry (unit: yuan / ton), or the potential utilization value of the slurry, such as energy value, sales or other derivative product revenue; Savings in slurry transportation costs (unit: yuan / ton): Dewatering reduces the volume and weight of the transported slurry. : Savings in slurry storage costs (unit: yuan / ton), as the storage space occupied by dewatered slurry is reduced; Initial investment cost of dehydration equipment (unit: yuan), including equipment purchase and installation costs; Operating and maintenance costs of dehydration equipment (unit: yuan), including electricity consumption, labor costs, equipment maintenance, etc.

[0005] Mineral slurry dewatering equipment is relatively complex, and its maintenance and replacement costs are high. Furthermore, in some cases, the moisture content of the output material may still be high, requiring secondary dewatering treatment. If this is not addressed promptly, it can lead to a significant waste of slurry water and added chemicals, reducing economic efficiency. This can cause problems in subsequent processing stages (such as coal washing and drying), impacting overall production efficiency.

[0006] Currently, most coal preparation plants still rely on manual experience to add feed and chemicals, and carry out dewatering work in a standardized process. The monitoring equipment is limited to simple online moisture meters, laser moisture analyzers, and pressure and flow meters, which are not systematic.

[0007] In order to comprehensively and systematically analyze the current application status of existing slurry dewatering equipment, based on the actual situation of pressure filters and plate and frame filter presses in coal washing plants and laboratory dewatering experiments, a digital twin method and system for slurry dewatering is proposed. This system integrates and visualizes various sensor monitoring devices in coal preparation plants to monitor the status of slurry water and other slurry dewatering processes in real time. Summary of the Invention

[0008] To overcome the problems of insufficient filter cake dewatering, poor effect, or dewatering failure in the pressure filtration and dewatering process of existing coal preparation plants and other mineral processing plants, this invention provides a digital twin method for optimizing the slurry dewatering process by integrating and visualizing various sensor monitoring devices in coal preparation plants, constantly monitoring the status of slurry dewatering process and adjusting various process equipment in real time.

[0009] The technical solution adopted by the present invention to achieve the above objectives is as follows:

[0010] A digital twin method for a slurry dewatering process includes the following steps:

[0011] S1: Real-time data acquisition and control of slurry dewatering equipment: Sensors are installed at the feed inlet and discharge outlet of the slurry dewatering equipment, and multi-module data of the sensors are constructed in real time; Feed or chemical and temperature PID controllers are installed at the feed inlet or chemical inlet of the slurry dewatering equipment.

[0012] S2: Construct an industrial test database: Construct an industrial test database using multi-module sensor data and relevant fluid particle state data obtained from a numerical simulation platform;

[0013] S3: Establish a virtual mapping model for real-time data transmission and visualization: Based on the relevant fluid particle state data obtained from the numerical simulation platform, a virtual mapping model is established through a reduced-order model and a neural network training model to achieve real-time data transmission and visualization.

[0014] S4: Establish a macroscopic geometric model mounting platform for the equipment: Import the multi-module sensor data and slurry dewatering process equipment from the industrial test database into the equipment theoretical geometric model mounting platform through Unity 3D to establish a macroscopic geometric model mounting platform for the equipment; conduct preliminary numerical simulation in industrial simulation software, import the training model process, and dynamically map the output results of the trained model to the virtual mapping model to achieve preliminary digital twinning and obtain digital twin results;

[0015] S5: Digital visualization and control: Visualize the digital twin results in ECharts, and use PID control to control the main body of the equipment using twin data adaptive control model and digital panel control.

[0016] Furthermore, the sensors installed at the feed inlet include: a raw material tank temperature sensor, a feed flow sensor, and a liquid pH sensor; the sensors installed on the main body of the slurry dewatering equipment at the feed inlet include: a filter press pressure sensor, a filter press temperature sensor, a filter press liquid level sensor, a filter press displacement sensor, a filter press vibration sensor, and a filter press torque sensor; the sensors installed at the discharge outlet include feed and discharge flow sensors, slurry and post-filter coal cake humidity sensors, coal cake density sensors, and a filter press effect image sensor; the PID controller includes a PID controller for the feed inlet and chemical inlet, a PID controller for the feed inlet and chemical inlet temperature, a PID controller for the main body pressure of the dewatering equipment, a PID controller for the main body timing of the dewatering equipment, and a PID controller for the main body temperature of the dewatering equipment.

[0017] Furthermore, the industrial test database is established using a MySQL relational database management system to create a digital twin database for slurry solid-liquid separation. This database includes basic data of the physical entity of the slurry dewatering process, various model data from the virtual entity, and simulation data from process simulation, behavioral simulation, evaluation, and analysis based on the aforementioned models. Before data import and export, a relevant database framework is established using SQL. The slurry solid-liquid separation digital twin database comprises two types of data: sensor multi-module data and relevant fluid particle state data obtained from the numerical simulation platform. The database table structures for FluidSimulationResults and SensorData are designed to store fluid numerical simulation results and sensor multi-module data, respectively.

[0018] Furthermore, the specific steps for establishing the virtual mapping model data for real-time transmission and visualization are as follows: Controlling and simulating the physical mechanical components of the device and visualizing sensor data utilizes the binlog binary format log file (ROW format) of MySQL software. The incremental data capture tool Maxwell is used to monitor the binlog file in real time, and the Maxwell output endpoint, the Kafka message queue system, is configured. The API provided by the message queue client library is used to receive data changes. A Kafka consumer is written in Java to receive sensor data changes sent from Maxwell to Kafka, and the parsed JSON data changes are sent to Kafka. Real-time data transmission is then transmitted to Unity. For phase entity displacement, adaptation to digital space is achieved using C# for spatial and temporal model matching. Data related to stress and phase interactions is visualized in digital space using C# scripts for synchronous real-time color rendering, which intuitively reflects the stress and flow velocity state of the phase. A digital twin model of fluid dynamics is constructed by coupling the reduced-order model (ROM) with a neural network (NN).

[0019] Furthermore, the specific steps for constructing the digital twin model of fluid dynamics are as follows:

[0020] Finite volume volume (FV) model for order reduction and intrinsic orthogonal decomposition (POD) were used for simulation. Computational fluid dynamics (CFD) was employed, utilizing a numerical simulation module for dispersed multiphase flow and a particle tracking module. The experimental method employed Euler-Euler. Unsteady flow field simulations were performed on the slurry dewatering process, acquiring flow field data at different time steps. Time series data were extracted, including particle flow and flow field position, velocity, acceleration, phase volume fractions of the particle and liquid phases, drag and lift distributions between particles and liquid, and interphase momentum transfer. Data preprocessing included outlier removal and normalization. Key characteristic modes were extracted, and POD decomposed the flow field data into a modal matrix and a time coefficient matrix using singular value decomposition (SVD). After data preprocessing, data standardization was performed, followed by POV order reduction. SVD decomposition was applied to the matrices.

[0021] (1)

[0022] In equation (1), X is the data matrix, U is the left singular vector matrix, V is the right singular vector matrix, and ∑ is the singular value matrix;

[0023] Based on actual slurry dewatering processes, in simulations using fluid dynamics and fine particle data, the energy accumulation ratio is calculated. With the principal component number k set to 5-10, the energy accumulation ratio reaches 99%, typically covering most data characteristics. The calculation formula is as follows:

[0024] Energy accumulation ratio (2)

[0025] In equation (2), n is the total number of singular values. It is the first One singular value;

[0026] Two reduced-order models are coupled, and the reduced-order governing equations are solved at each time step, including data on velocity and pressure field variables. The reduced-order model data is compared and verified with relevant original data, and the reduced-order model is adjusted to optimize the model's representation of the state and interaction forces between the flow field and particles.

[0027] A training neural network was constructed, employing a combined architecture of Long Short-Term Memory Recurrent Neural Network (LSTM-RNN), Attention Mechanism (AM), and Physical Information Neural Network (PINN). During training, the attention mechanism was used, taking the boundary and initial conditions of the flow field as input. The input data included particle flow and the flow field's position, velocity, acceleration, phase volume fractions of the particle and liquid phases, drag and lift distributions between particles and the liquid phase, and momentum transfer between phases. Modal coefficients were used as output to construct training and testing datasets. An error fitting algorithm was used for the LSTM-RNN, attention mechanism, and PINN neural networks, employing Mean Squared Error (MSE) and residual loss functions to calculate the difference between predicted and true values. The Navier-Stokes equation and particle settling velocity formula were incorporated as constraints into the loss function. An optimizer algorithm was selected and the Adam optimizer was used to update the network parameters. Iterative training was performed until the loss function converged to a suitable value. The combined use of PINN and LSTM improved prediction accuracy.

[0028] Specifically, the input layer contains boundary and initial conditions of the flow field, including particle position, velocity, acceleration, volume fraction of the particle and liquid phases, drag and lift distributions, and data on interphase momentum transfer. The LSTM-RNN layer uses a multi-layer LSTM-RNN structure to capture more complex spatiotemporal features. The AM layer applies an attention mechanism to the LSTM-RNN output, dynamically weighting features at different time steps to make the model focus more on important time steps. The PINN layer embeds physical equations into the network, constraining the network output to conform to physical laws through a loss function. In particular, the Navier-Stokes equations, the physical equations of the flow field, are embedded into the network, and the k feature vectors determined above form the reduced-dimensional basis space.

[0029] The original Navier-Stokes equations

[0030] (3)

[0031] Using the Galerkin projection method, the governing equations are reduced in order to obtain the governing equations.

[0032] (4)

[0033] Projection

[0034] (5)

[0035] In equation (5), p is pressure, v is dynamic viscosity, u is velocity, f is external force, and q is a time-dependent coefficient vector.

[0036] The phase volume fractions of the particle flow group and the liquid phase obtained from numerical simulation.

[0037] Furthermore, the specific steps for establishing the equipment geometric macroscopic model mounting platform are as follows:

[0038] The parameters of the slurry dewatering process equipment and the initial sensor multi-module data are imported to establish a theoretical geometric model of the equipment mounting platform, and preliminary digital twinning is carried out in industrial simulation software.

[0039] The theoretical geometric models of the equipment were created and adjusted in digital space using 3D modeling software. This included the establishment of steady-state and dynamic models. Each dynamic model was adjusted and controlled using corresponding control programs based on the collected physical data or processed data. Within the digital space, 3D modeling tools were used to construct physical models of the equipment, and these models were simplified to improve efficiency. C# control scripts were written for these equipment models in the Unity 3D environment to dynamically update and manage relevant data information such as feed flow rate and pressure. The macroscopic geometric model of the equipment includes the geometric creation of various components, the pressure shaft of the main body of the filter press, hardware devices with minimal impact on the dewatering effect monitoring such as the sample container, and a small amount of monitoring data. The monitoring data includes sensors installed at the raw material inlet: temperature sensor, flow sensor, and pH sensor; and sensors installed on the main body of the filter press: temperature sensor, level sensor, vibration sensor, and torque sensor.

[0040] Firstly, monitor the data visualization transmission process by creating a C# script in Unity to connect to the Kafka message queue system and receive data changes from Maxwell; secondly, control the Unity model visualization by parsing the received data changes into a format that Unity can understand and updating the model state in Unity as needed; and thirdly, visualize the solid and liquid phases of the slurry on the visualization platform.

[0041] Secondly, visualize the data obtained from the numerical simulation, save the model in ONNX format after training, and load and run it using Unity's Barracuda plugin; set up a TCP server, create a TCP server script in Unity (Python) to receive data sent by the client; create a TCP client script in Unity to send data to the server; load the ONNX model using Barracuda, and send the model's output to the client via TCP.

[0042] Based on the output of the trained model, the flow state of coal slurry water is simulated in real time using the particle system and fluid simulation module in Unity. This dynamically displays the fluid trajectory, velocity distribution, and particle motion and distribution. A particle system is created, including particle attributes, quantity, and lifecycle, and each particle is assigned a direction vector and velocity. Using Unity's physics engine simulation algorithm, the flow field is meshed, storing the flow velocity and pressure field of each mesh. Stokes equations are solved, and fluid parameters and boundary conditions are adjusted based on the training model's prediction results and real-time dynamic data collected from the database, resulting in a more accurate and real-time simulation of the dehydration flow process. Unity's computational shaders are used to enhance the visualization and dynamic simulation of particles and the flow field. A custom shader is created, and its parameters are adjusted to control the rich fluid flow and particle gradients. Unity's UI system is used to create a visualization interface to display the model's output. A UI Text component is created to display the model's output, and a RawImage component is created to display the image of the input data. Vertex colors are changed in real time to represent the pressure field, and material texture maps are used to represent the temperature field, achieving visualization of numerical simulation data and dynamically mapping the model's output to the virtual model.

[0043] Set up a parameter control panel in Unity to input and adjust the data flow rate, pressure, and particle distribution data of the above training model in real time, so as to achieve real-time parameter control of the model;

[0044] To improve visualization efficiency, data compression and transmission optimization were implemented. This included data cleaning using the Huffman lossless compression algorithm to convert data into binary data and then into a compressed format, with data differential technology used to improve transmission efficiency. Secondly, to optimize visualization rendering in Unity, specifically for industrial applications, Unity's Level of Detail and Draw Call technologies were used. Draw Call technology was employed to render equipment and related static models, while Level of Detail technology was used to dynamically adjust fluid and particle model details based on camera distance, setting multiple LOD levels for appropriate rendering.

[0045] Furthermore, the digital twin results are visualized in ECharts, and the main body of the equipment is controlled using PID control. The steps of the twin data adaptive control model and digital panel control are as follows:

[0046] The results obtained from the digital twin are visualized in ECharts, and PID control is used to regulate the main body of the equipment. A user interface is built using JavaScript via the web, embedding ECharts charts to display the data, and a timer is used to periodically retrieve data from the backend and update the charts. Control commands are input into the user interface, and these commands need to be transmitted to the backend server via communication. The backend server Node.js application receives the frontend requests, processes the corresponding logic, and converts them into a binary format that the PID controller can accept. Then, these control commands are sent to the PID controller for control via serial communication.

[0047] Furthermore, the steps of using PID control to regulate the main body of the equipment, the twin data adaptive regulation model, and the digital panel regulation are as follows: relevant data is sent to the adaptive model of the adaptive regulation process using the Kafka message queue system; relevant data is calculated from the twin data according to the mathematical model of slurry dewatering efficiency and sent to the Echart chart; automatic control or manual regulation is set; control commands are sent to the WebSocket server; and the data is processed by the WebSocket server to achieve adaptive regulation.

[0048] Based on the monitored dehydration data, the Compression Dehydration Index (CDI) is calculated. By adjusting the pressure and filter cake thickness, the dehydration efficiency can be optimized and the CDI increased.

[0049] (6)

[0050] In equation (6), It is the change in filter cake thickness. It is the change in pressure.

[0051] A digital twin system for a slurry dewatering process includes:

[0052] Real-time data acquisition and control unit for slurry dewatering equipment: Sensors are installed at the inlet and outlet of the slurry dewatering equipment and multi-module data of the sensors are constructed in real time; PID controllers for feeding or chemical and temperature are installed at the inlet or chemical inlet of the slurry dewatering equipment.

[0053] Industrial test database construction unit: The industrial test database is constructed using multi-module sensor data and relevant fluid particle state data obtained from the numerical simulation platform;

[0054] Virtual mapping model establishment and real-time data transmission and visualization unit: Based on the relevant fluid particle state data obtained from the numerical simulation platform, a virtual mapping model is established through a reduced-order model and a neural network training model to transmit data in real time and achieve visualization.

[0055] Establish a macroscopic geometric model mounting platform for the equipment: Import multi-module sensor data and slurry dewatering process equipment from the industrial test database into the equipment theoretical geometric model mounting platform through Unity 3D to establish the macroscopic geometric model mounting platform for the equipment; conduct preliminary numerical simulation in industrial simulation software, import the training model process, and dynamically map the output results of the trained model to the virtual mapping model to achieve preliminary digital twinning and obtain digital twin results;

[0056] Visualization and Control Unit: Visualizes the digital twin results in ECharts, and uses PID control to regulate the main body of the equipment, using a twin data adaptive control model and digital panel control.

[0057] This invention provides real-time monitoring of the working status and dewatering effect of pressure dewatering equipment, reducing labor costs and increasing the efficiency of slurry dewatering, making mineral processing engineering more stable and efficient in pressure filtration dewatering. A process system for real-time data import and export in a twin database is created, making the linkage between monitoring data and visualization more convenient and enabling timely adjustment. A database is established for the real-time monitored data to achieve PID control, which includes two control processes: one is a preliminary adaptive control process for slurry dewatering based on a corresponding dewatering efficiency mathematical model; the other is human intervention and adjustment via a web control panel according to required instructions. By combining numerical simulation with digital twin visualization of the slurry dewatering process, the monitoring results can be scientifically analyzed, and multiple dewatering controls can achieve more accurate and precise regulation. Through appropriate selection, simplification, and compression of the range, model, and data of the numerical simulation, compared with existing digital twin technologies in the field of slurry dewatering, it offers more efficient visualization and control, improving its visualization speed and accuracy. Attached Figure Description

[0058] The present invention will now be further described with reference to the accompanying drawings, wherein:

[0059] Figure 1 Overall route hierarchy diagram.

[0060] Figure 2 Database creation flowchart.

[0061] Figure 3 Real-time data monitoring and transmission process.

[0062] Figure 4 Virtual mapping model and visualization process.

[0063] Figure 5 Improve the efficiency of model transfer and visualization processes.

[0064] Figure 6 Neural network flowchart.

[0065] Figure 7 LSTM logic structure diagram.

[0066] Figure 8 PID control flow and adaptive flow diagram. Detailed Implementation

[0067] Example 1

[0068] like Figure 1 As shown, the digital twin method for slurry dewatering process in this embodiment includes the following steps:

[0069] S1: Real-time data acquisition and control of slurry dewatering equipment: Sensors are installed at the feed inlet and discharge outlet of the slurry dewatering equipment, and multi-module data of the sensors are constructed in real time; Feed or chemical and temperature PID controllers are installed at the feed inlet or chemical inlet of the slurry dewatering equipment.

[0070] Adding monitoring sensors

[0071] A. Sensors installed at the raw material inlet include:

[0072] 1) Temperature sensor: Monitors the temperature change of the raw material tank to prevent the physical properties of the raw material from being affected by excessively hot or cold temperatures, thus ensuring the dehydration effect.

[0073] 2) Flow sensor: monitors the feed flow rate to control the feed speed and processing capacity of the filter press.

[0074] 3) pH sensor: monitors the acidity or alkalinity of the liquid to control the pH value of the raw materials.

[0075] B. Install sensors on the main body of the filter press dewatering equipment:

[0076] 1) Pressure sensor: Used to monitor the pressure of the filter press to ensure that the filtration process is carried out under the set or real-time adjusted pressure.

[0077] 2) Temperature sensor: Monitors temperature changes during the filtration process to prevent overheating or underheating from affecting the dehydration effect.

[0078] 3) Liquid level sensor: Monitors the liquid level inside the filter press dewatering equipment to prevent overload or idling and ensure dewatering efficiency.

[0079] 4) Displacement sensor: Combined with a liquid level sensor, it monitors the degree of sample deformation in the filter press dewatering equipment.

[0080] 5) Vibration sensor: Monitors the vibration of the filter press dewatering equipment to detect equipment failure and maintain the equipment.

[0081] 6) Torque sensor: Monitors the torque of the pressurizing components of the filter press dewatering equipment to ensure that the equipment operates in optimal condition.

[0082] C. Install various sensors and other sensor monitoring components at the discharge port.

[0083] 1) Flow sensor: monitors the flow rate of feed and discharge to control the feed rate and processing capacity of the filter press.

[0084] 2) Humidity sensor: detects the humidity of the slurry and the filter cake after pressing to evaluate the dewatering effect.

[0085] 3) Density sensor: measures the density of the coal cake after filter pressing to assess its dewatering degree.

[0086] 4) Image sensor: Visually monitors the filtration effect, such as the uniformity and condition of the coal cake.

[0087] PID controller installation

[0088] A. Install PID controllers at the feed inlet and reagent inlet to adjust the addition and quantity of slurry and reagents, control the slurry feed rate and the added reagents to control the relevant physicochemical properties of the slurry (adhesion and pH value, etc.); install a temperature PID controller to initially adjust the slurry temperature and maximize the slurry dewatering efficiency.

[0089] B. A pressure PID controller and a timer PID controller are installed in the main body of the dewatering equipment to adjust the dewatering pressure and the dewatering time under the relevant pressure, so as to achieve better dewatering effect; a second temperature PID controller is installed in the main body to adjust the effect of temperature on dewatering efficiency.

[0090] S2: Construct an industrial test database: Construct an industrial test database using multi-module sensor data and relevant fluid particle state data obtained from a numerical simulation platform;

[0091] The industrial test database includes the establishment of two types of data: one is sensor multi-module data, and the other is relevant fluid particle state data obtained from the numerical simulation platform.

[0092] like Figure 2 As shown, the database framework utilizes a relational database management system such as MySQL to establish a digital twin database for slurry solid-liquid separation. The database includes basic data of the aforementioned physical entities, various model data from the virtual entities, and simulation data for process simulation, behavioral simulation, evaluation, and analysis based on these models. The relevant database framework is established using SQL before data import and export.

[0093] Design the database table structures for FluidSimulationResults and SensorData to store fluid numerical simulation results and sensor data, respectively.

[0094] like Figure 3 As shown, the two types of data collection are monitoring source data and theoretical source data.

[0095] Monitoring source data includes real-time data collected by sensors in the main body of the dewatering equipment and related devices, including data collected from the raw material tank, feeding device and dewatering body, including real-time data such as temperature, pH, flow rate, liquid level, displacement, pressure and so on (data monitored by the relevant sensors introduced in S1); for the raw material base raw data, the exported CSV file is used to write a Python script to read the data from the local file and insert it into the MySQL database.

[0096] Theoretical source data: Numerical simulations were performed using industrial simulation software, such as Comsol and Fluent. The relevant simulation results include parameters such as fluid velocity, fluid resistance, migration position changes of fixed-position slurry particles, lift force, and resistance force on the slurry particles. Due to the enormous computational complexity of considering the overall slurry dewatering process, and the fact that local fluid strain patterns can be verified for overall slurry dewatering, a simplified simulation object was chosen. The data for slurry dewatering was used to simulate the central slurry within a cylinder with an XY axis diameter of 10 mm and a Z-axis length of 20 mm.

[0097] For monitoring source data, Python is used to collect data from sensors in real time using a data acquisition card. A Python database connection library is used to create a connection to a MySQL database. The collected data is then sent to the MySQL system and interacted with using the Python (mysql-connector-python library).

[0098] For theoretical source data, Python scripts are used to acquire data in real time. For example, Fluent's User-Defined Functions (UDFs) can be used. The UDF code uses JDBC to connect to a MySQL database. A UDF written in C and compiled and loaded in Fluent is created to write simulation results to the MySQL database at each time step or specific time interval. Within the UDF, trigger conditions (such as time step size or iteration count) are set to write simulation results to the MySQL database in real time via the JDBC interface.

[0099] S3: Establish virtual mapping model data for real-time transmission and visualization: Based on the relevant fluid particle state data obtained from the numerical simulation platform, a virtual mapping model is established through a reduced-order model and a neural network training model to achieve real-time data transmission and visualization, including specific visualization of the two types of relevant data.

[0100] like Figure 4 As shown, this demonstrates the control and simulation of the physical mechanical components of the device and the visualization of related sensor data. The binlog binary format log file (ROW format) of MySQL software is used, and the binlog file is monitored in real time using an incremental data capture tool (Maxwell). The Maxwell output endpoint (Kafka) message queue system is configured (using the API provided by the message queue client library to receive data changes). A Kafka consumer is written in Java to receive sensor data changes sent to Kafka by Maxwell, and sends the parsed JSON data changes back to Kafka. Real-time data is then transmitted to Unity.

[0101] like Figure 5 As shown, the model is adapted to digital space in terms of phase displacement and related aspects, and C# is used for spatial and temporal model matching. For example, data related to stress and phase interactions are visualized in digital space using C# scripts, with synchronous real-time color rendering to intuitively represent the stress, flow rate, and other states of the phases. A digital twin model of fluid dynamics is constructed by coupling a reduced-order model (ROM) with a neural network (NN).

[0102] The specific implementation process is as follows: Finite Volume (FV) model and intrinsic orthogonal decomposition (POD) are used to significantly improve computational efficiency while maintaining high accuracy. Computational fluid dynamics (CFD) is employed for simulation, using a numerical simulation module for dispersed multiphase flow and a particle tracking module. The experimental method is Euler-Euler. Unsteady flow field simulation is performed on the slurry dewatering process to obtain flow field data at different time steps. Time series data related to particle flow and flow field position, velocity, acceleration, phase volume fraction of particle and liquid phases, drag and lift distribution between particles and liquid phases, and interphase momentum transfer are extracted. Data preprocessing is performed, including outlier removal and normalization. Key characteristic modes are extracted, and POD decomposes the flow field data into a mode matrix and a time coefficient matrix using singular value decomposition (SVD). Specifically, after data preprocessing, data standardization is performed, followed by POV model reduction. SVD decomposition is then applied to the matrices.

[0103] (1)

[0104] In equation (1), X is the data matrix, U is the left singular vector matrix, V is the right singular vector matrix, and ∑ is the singular value matrix.

[0105] Based on actual slurry dewatering processes, in simulations using fluid dynamics and fine particle data, the energy accumulation ratio is calculated. With the principal component number k set to 5-10, the energy accumulation ratio reaches 99%, typically covering most data characteristics. The calculation formula is as follows:

[0106] Energy accumulation ratio (2)

[0107] In equation (2), n is the total number of singular values. It is the first A singular value.

[0108] Two reduced-order models are coupled, and the reduced-order governing equations are solved at each time step to obtain data on variables including velocity and pressure fields. The reduced-order model data is compared and validated with relevant original data, and the reduced-order model is adjusted to optimize the model's representation of the state and interaction forces between the flow field and particles.

[0109] like Figure 6As shown, a training neural network is constructed, employing a combined architecture of Long Short-Term Memory Recurrent Neural Network (LSTM-RNN), Attention Mechanism (AM), and Physical Information Neural Network (PINN). During training, the Attention Mechanism is used, taking the boundary and initial conditions of the flow field as inputs (data including particle flow and flow field position, velocity, acceleration, phase volume fraction of the particle and liquid phases, drag and lift distribution between particles and liquid, and momentum transfer between phases). Modal coefficients are used as outputs to construct training and testing datasets. An error fitting algorithm is used in the LSTM-RNN, Attention Mechanism, and PINN neural networks, employing mean squared error (MSE) and residual loss functions to calculate the difference between predicted and true values. The Navier-Stokes equation and particle settling velocity formula are added as constraints to the loss function. The optimizer algorithm selects and uses the Adam optimizer to update network parameters, performing iterative training until the loss function converges to a suitable value. The combined use of PINN and LSTM improves prediction accuracy.

[0110] Specifically,

[0111] (1) Input layer (boundary and initial conditions of the flow field, including particle position, velocity, acceleration, volume fraction of particle phase and liquid phase, drag and lift distribution, and data on interphase momentum transfer, etc.)

[0112] (2) LSTM-RNN layers (using multi-layer LSTM-RNN structures to capture more complex spatiotemporal features.) (e.g.) Figure 7 LSTM logic flowchart

[0113] (3) AM layer (an attention mechanism is applied to the output of LSTM-RNN to dynamically weight the features at different time steps, making the model pay more attention to important time steps)

[0114] (4) PINN layer (embedding physical equations into the network and constraining the network output to conform to physical laws through a loss function. Specifically, the physical equations of the flow field (such as the Navier-Stokes equations) are embedded into the network, and the k feature vectors determined above are used to construct the reduced dimensionality space).

[0115] The original Navier-Stokes equations

[0116] (3)

[0117] Using the Galerkin projection method, the governing equations are reduced in order to obtain the governing equations.

[0118] (4)

[0119] Projection

[0120] (5)

[0121] In equation (5), p is pressure, v is dynamic viscosity, u is velocity, f is external force, and q is a time-dependent coefficient vector.

[0122] The phase volume fractions of the particle flow group and the liquid phase obtained from numerical simulation.

[0123] S4: Establish a macroscopic geometric model mounting platform for the equipment: Import the multi-module sensor data and slurry dewatering process equipment from the industrial test database into the equipment theoretical geometric model mounting platform through Unity 3D to establish a macroscopic geometric model mounting platform for the equipment; conduct preliminary numerical simulation in industrial simulation software, import the training model process, and dynamically map the output results of the trained model to the virtual mapping model to achieve preliminary digital twinning and obtain digital twin results.

[0124] like Figure 4 As shown, the theoretical geometric models of the equipment involved were created and adjusted in digital space using 3D modeling software. This included the establishment of steady-state and dynamic models. Each dynamic model was adjusted and controlled using corresponding control programs based on the collected physical data or processed data. Within the digital space, 3D modeling tools such as UGNX and Maya were used to construct the physical models of the equipment, and the models were simplified to improve efficiency. Then, C# control scripts were written for these equipment models in the Unity 3D environment to dynamically update and manage relevant data information (feed flow rate, pressure, etc.). The macroscopic geometric model of the equipment includes the geometric creation of various components, hardware devices with minimal impact on the monitoring of dewatering effects such as the pressure shaft of the main body of the filter press and sample containers, and a small amount of monitoring data. The monitoring data consists of sensors installed at the raw material inlet: temperature sensor, flow sensor, and pH sensor; and sensors installed on the main body of the filter press: temperature sensor, level sensor, vibration sensor, and torque sensor.

[0125] Firstly, the data visualization transmission process is monitored. A C# script is created in Unity to connect to the Kafka message queue system and receive data changes from Maxwell. Unity model visualization control: After receiving data changes, they are parsed into a format that Unity can understand, and the model state in Unity is updated as needed. The visualization subject is the solid and liquid phases of the slurry, represented on a visualization platform such as Unity.

[0126] Secondly, the data obtained from the numerical simulation is visualized. After training, the model is saved in ONNX format and loaded and run using Unity's Barracuda plugin. A TCP server is set up. A TCP server script (Python) is created in Unity to receive data sent by the client; a TCP client script is created in Unity to send data to the server; the ONNX model is loaded using Barracuda, and the model's output is sent to the client via TCP.

[0127] Based on the output of the trained model, the flow state of coal slurry water is simulated in real time using the particle system and fluid simulation module in Unity, dynamically displaying the fluid trajectory, velocity distribution, and particle motion and distribution. A particle system is created, including particle attributes, quantity, and lifecycle, and a direction vector and velocity are assigned to each particle. Using Unity's physics engine simulation algorithm, the flow field is meshed, storing the flow velocity and pressure field of each mesh, solving Stokes' equations, and adjusting relevant fluid parameters and boundary conditions based on the prediction results of the trained model and real-time dynamic data collected from the database, resulting in a more accurate and real-time simulation of the dehydration flow process. Unity's computational shaders are used to enhance the visualization and dynamic simulation of particles and the flow field. A custom shader is created, and its parameters are adjusted to create rich fluid flow and particle gradations (e.g., color gradients can be used to represent different velocity levels, with red for high speed and blue for low speed). Unity's UI system is used to create a visualization interface to display the model's output results. A UIText component is created to display the model's output, and a RawImage component is created to display the image of the input data. The vertex color is changed in real time to represent the pressure field, and the temperature field is represented by the texture map of the material, realizing the visualization of numerical simulation data and dynamically mapping the output results of the model to the virtual model.

[0128] Set up a parameter control panel in Unity to input and adjust the data (flow rate, pressure, particle distribution, etc.) of the above-mentioned training model in real time, so as to achieve real-time parameter control of the model.

[0129] To improve visualization efficiency, data compression and transmission optimization were implemented. This included data cleaning, using the Huffman lossless compression algorithm to convert the data into binary data and then into a compressed format, and employing data differential technology to improve transmission efficiency. Secondly, to optimize visualization rendering in Unity, specifically for industrial applications, Unity's Level of Detail and Draw Call technologies were used. Draw Call technology was used to render equipment and related static models, while Level of Detail technology was used to dynamically adjust fluid and particle model details based on camera distance, setting multiple LOD levels for appropriate rendering.

[0130] S5: Digital visualization and control: Visualize the digital twin results in ECharts, and use PID control to control the main body of the equipment using twin data adaptive control model and digital panel control.

[0131] A user interface is built via the web (using JavaScript) that embeds ECharts charts to display data, and a timer periodically retrieves data from the backend and updates the charts. The user interface inputs control commands, which are transmitted to the backend server via communication. The backend server (a Node.js application) receives the frontend requests and processes the corresponding logic (converting it to a binary format acceptable to the PID controller). Then, these control commands are sent to the PID controller via serial communication for control.

[0132] like Figure 8 As shown, the twin data adaptive control model and digital panel control utilize the Kafka message queue system to send relevant data to the adaptive model of the adaptive control process. Based on the mathematical model of slurry dewatering efficiency, the twin data is used to calculate relevant data and send it to the Echart chart. Automatic control or manual control is set, and control commands are sent to the WebSocket server. The WebSocket server performs logical processing on the data for the PID controller to achieve adaptive control.

[0133] Based on the monitored dehydration data, the Compression Dehydration Index (CDI) is calculated. By adjusting the pressure and filter cake thickness, the dehydration efficiency can be optimized and the CDI increased.

[0134] (6)

[0135] In equation (6), It is the change in filter cake thickness. It is the change in pressure.

[0136] like Figure 1As shown, the digital twin system for the slurry dewatering process in this embodiment includes:

[0137] Real-time data acquisition and control unit for slurry dewatering equipment: Sensors are installed at the inlet and outlet of the slurry dewatering equipment and multi-module data of the sensors are constructed in real time; PID controllers for feeding or chemical and temperature are installed at the inlet or chemical inlet of the slurry dewatering equipment.

[0138] Industrial test database construction unit: The industrial test database is constructed using multi-module sensor data and relevant fluid particle state data obtained from the numerical simulation platform;

[0139] Virtual mapping model establishment and real-time data transmission and visualization unit: Based on the relevant fluid particle state data obtained from the numerical simulation platform, a virtual mapping model is established through a reduced-order model and a neural network training model to transmit data in real time and achieve visualization.

[0140] Establish a macroscopic mounting platform unit for the equipment geometric model: Import the sensor multi-module data and slurry dewatering process equipment from the industrial test database into the geometric model mounting platform through Unity 3D to establish a macroscopic mounting platform for the geometric model; conduct preliminary numerical simulation in industrial simulation software, import the training model process, and dynamically map the output results of the trained model to the virtual mapping model to achieve preliminary digital twinning and obtain digital twin results;

[0141] Visualization and Control Unit: Visualizes the digital twin results in ECharts, and uses PID control to regulate the main body of the equipment, using a twin data adaptive control model and digital panel control.

[0142] This system is divided into four layers: equipment control layer, twin data layer, visual transmission layer, and digital service layer. Its specific design and implementation steps are as follows: The equipment control layer is the real-time data acquisition and control unit for the slurry dewatering process equipment; the twin data layer includes an industrial test database construction unit and a virtual mapping model establishment and real-time data transmission and visualization unit; the visual transmission layer is the unit for establishing the equipment geometric model and macroscopic mounting platform; and the digital service layer is the visualization and control unit.

[0143] The process of establishing a digital twin database for slurry dewatering utilizes the source code template for building a data framework in the MySQL system and related data twinning methods. The input data stream includes real-time monitoring data and relevant physicochemical properties of the slurry. For real-time import and export of the twin database, a real-time data transmission system and method are established, including using the generated binlog files and the Maxwell tool to monitor and receive real-time data updates, thereby obtaining real-time changes in the collected data.

[0144] Technologies for improving the efficiency of digital twins include strategies and methods for simplifying numerical simulation models, cleaning and optimizing simulation data, forming predictive models, optimizing visualization data transmission, and optimizing rendering. Specifically, this includes simplifying the model by selecting representative models during slurry simulation; processing numerical simulation results, including methods for degrading the data model and the mathematical model; selecting deep learning strategies and methods for generating training experience models, including strategies and algorithms related to training with LSTM and AM and constraining them with PINN; and methods and strategies for improving visualization efficiency in the digital twin visualization process, such as model data processing and rendering.

[0145] The visualization control route design for slurry dewatering utilizes a digital twin platform on a PC to visualize the 3D model and adjust related parameters. A visualization panel allows for manual intervention and control experiments as needed. Control commands received from the web frontend are transmitted as binary data to the PID controller via a Node.js backend. Relevant slurry raw material data and real-time twin data monitoring are loaded into a mathematical model, and the data is processed via a WebSocket server to achieve adaptive control of the PID controller.

Claims

1. A digital twin method for a slurry dewatering process, characterized in that, Includes the following steps: S1: Real-time data acquisition and control of slurry dewatering equipment: Sensors are installed at the feed inlet and discharge outlet of the slurry dewatering equipment, and multi-module data of the sensors are constructed in real time; Feed or chemical and temperature PID controllers are installed at the feed inlet or chemical inlet of the slurry dewatering equipment. S2: Construct an industrial test database: Construct an industrial test database using multi-module sensor data and relevant fluid particle state data obtained from a numerical simulation platform; S3: Establish a virtual mapping model for real-time data transmission and visualization: Based on the relevant fluid particle state data obtained from the numerical simulation platform, a virtual mapping model is established through a reduced-order model and a neural network training model to achieve real-time data transmission and visualization. S4: Establish a geometric macroscopic model mounting platform for the equipment: Import the sensor multi-module data and slurry dewatering process equipment from the industrial test database into the geometric model mounting platform through Unity 3D to establish the geometric macroscopic model mounting platform; Preliminary numerical simulations are performed in industrial simulation software. The training model is imported and dynamically mapped to a virtual mapping model to achieve preliminary digital twinning and obtain digital twin results. Based on the output of the trained model, the flow state of coal slurry water is simulated in real time using the particle system and fluid simulation module in Unity. The fluid trajectory, velocity distribution, and particle motion and distribution are dynamically displayed. A particle system is created, including particle attributes, quantity, and life cycle. Each particle is assigned a direction vector and velocity. Using Unity's physics engine simulation algorithm, the flow field is meshed, and the flow velocity and pressure field of each mesh are stored. Stokes equations are solved, and the relevant parameters and boundary conditions of the fluid are adjusted based on the prediction results of the training model and the dynamic data collected in real time from the database, so as to simulate the dewatering flow process more accurately and in real time. S5: Digital Visualization and Control: Visualize the digital twin results in ECharts, utilize PID control to regulate the main equipment using a twin data adaptive control model and digital panel control; send relevant data from the Kafka message queue system to the adaptive model of the adaptive control process; based on the mathematical model of slurry dewatering efficiency, calculate relevant data from the twin data and send it to the EChart chart; set automatic or manual control; send control commands to the WebSocket server; the WebSocket server performs logical processing on the data for the PID controller to achieve adaptive control; calculate the Compression Dewatering Index (CDI) based on the monitored dewatering data; by adjusting the pressure and filter cake thickness, dewatering efficiency can be optimized and the CDI increased. （1） In equation (1), It is the change in filter cake thickness. It is the change in pressure.

2. The digital twin method for slurry dewatering process according to claim 1, characterized in that, The sensors installed at the feed inlet include: a raw material tank temperature sensor, a feed flow sensor, and a liquid pH sensor; the sensors installed on the main body of the slurry dewatering equipment at the feed inlet include: a filter press pressure sensor, a filter press temperature sensor, a filter press liquid level sensor, a filter press displacement sensor, a filter press vibration sensor, and a filter press torque sensor; the sensors installed at the discharge outlet include feed and discharge flow sensors, slurry and post-filter coal cake humidity sensors, coal cake density sensors, and a filter press effect image sensor; the PID controller includes a PID controller for the feed inlet and chemical inlet, a PID controller for the feed inlet and chemical inlet temperature, a PID controller for the main body pressure of the dewatering equipment, a PID controller for the main body timing of the dewatering equipment, and a PID controller for the main body temperature of the dewatering equipment.

3. The digital twin method for slurry dewatering process according to claim 1, characterized in that, The industrial test database is established using a MySQL relational database management system to create a digital twin database for slurry solid-liquid separation. This database includes basic data of the physical entity of the slurry dewatering process, various model data from the virtual entity, and simulation data from process simulation, behavioral simulation, evaluation, and analysis based on these models. Before data import and export, a relevant database framework is established using SQL. The slurry solid-liquid separation digital twin database comprises two types of data: sensor multi-module data and relevant fluid particle state data obtained from the numerical simulation platform. The database table structures for FluidSimulationResults and SensorData are designed to store fluid numerical simulation results and sensor multi-module data, respectively.

4. The digital twin method for slurry dewatering process according to claim 1, characterized in that, The specific steps for establishing a virtual mapping model, transmitting data in real time, and realizing visualization are as follows: The control and simulation of the physical and mechanical components of the equipment and the visualization of sensor data are achieved using the binlog binary format ROW file of MySQL software. The incremental data capture tool Maxwell is used to monitor the binlog file in real time, and the output endpoint of Maxwell, the Kafka message queue system, is configured. The API provided by the message queue client library is used to receive data changes. A Kafka consumer is written in Java to receive sensor data changes sent from Maxwell to Kafka, and the parsed JSON data changes are sent back to Kafka. Real-time data transmission is then performed to Unity. For phase entity displacement, adaptation to digital space is achieved using C# for spatial and temporal model matching. Data related to stress and phase interactions are visualized in digital space using C# scripts for synchronous real-time color rendering, which visually represents the stress and flow velocity state of the phases. A digital twin model of fluid dynamics is constructed by coupling the reduced-order model ROM with a neural network NN.

5. The digital twin method for slurry dewatering process according to claim 1, characterized in that, The specific steps for establishing the equipment's geometric macroscopic model and mounting platform are as follows: The parameters of the slurry dewatering process equipment and the initial sensor multi-module data are imported into the equipment theoretical geometric model mounting platform, and preliminary digital twinning is carried out in industrial simulation software. The theoretical geometric models of the equipment were created and adjusted in digital space using 3D modeling software. This included the establishment of steady-state and dynamic models. Each dynamic model was adjusted and controlled using corresponding control programs based on the collected physical data or processed data. In digital space, 3D modeling tools were used to construct physical models of the equipment, and the models were simplified to improve efficiency. C# control scripts were written for these equipment models in the Unity 3D environment to realize the dynamic updating and management of relevant data information such as feed flow rate and pressure in the models. The establishment of the macroscopic geometric model of the equipment included the geometric creation of various components of the equipment, the main pressure shaft of the filter press dewatering equipment, and hardware devices and monitoring data of the sample container that have little impact on the monitoring of the dewatering effect. Firstly, monitor the data visualization transmission process by creating a C# script in Unity to connect to the Kafka message queue system and receive data changes from Maxwell; secondly, control the Unity model visualization by parsing the received data changes into a format that Unity can understand and updating the model state in Unity as needed; and thirdly, visualize the solid and liquid phases of the slurry on the visualization platform. Secondly, visualize the data obtained from the numerical simulation, save the model in ONNX format after training, and load and run it using Unity's Barracuda plugin; set up a TCP server, create a TCP server script in Unity (Python) to receive data sent by the client; create a TCP client script in Unity to send data to the server; load the ONNX model using Barracuda, and send the model's output to the client via TCP. This process utilizes Unity's computational shaders to enhance the visualization and dynamic simulation of particles and flow fields. A custom shader is created, and its parameters are adjusted to modify the rich fluid flow and particle gradients. Unity's UI system is used to create a visualization interface that displays the model's output results. Specifically, a UI Text component is created to display the model's output, and a RawImage component is created to display the image of the input data. Vertex colors are changed in real time to represent the pressure field, and the temperature field is represented by material texture maps, thus achieving the visualization of numerical simulation data and dynamically mapping the model's output results into the virtual model. Set up a parameter control panel in Unity to input and adjust the data flow rate, pressure, and particle distribution data of the above training model in real time, so as to achieve real-time parameter control of the model; The Huffman lossless compression algorithm is used to convert the data into binary data and then into a compressed format. Data differential technology is used to improve transmission efficiency. Unity's Level of Detail and Draw Call technologies are used to render the device and related static models using Draw Call technology. The fluid and particle model details are dynamically adjusted according to the camera's distance using Level of Detail technology, and multiple LOD levels are set for rendering.

6. The digital twin method for slurry dewatering process according to claim 1, characterized in that, The steps of visualizing the digital twin results in ECharts and using PID control to regulate the main body of the equipment, including the twin data adaptive regulation model and digital panel regulation, are as follows: The results obtained from the digital twin are visualized in ECharts, and PID control is used to regulate the main body of the equipment. A user interface is built using JavaScript via the web, embedding ECharts charts to display the data, and a timer is used to periodically retrieve data from the backend and update the charts. Control commands are input into the user interface, and these commands need to be transmitted to the backend server via communication. The backend server Node.js application receives the frontend requests, processes the corresponding logic, and converts them into a binary format that the PID controller can accept. Then, these control commands are sent to the PID controller for control via serial communication.

7. A system for implementing the digital twin method for slurry dewatering process according to any one of claims 1 to 6, characterized in that, include: Real-time data acquisition and control unit for slurry dewatering process equipment: Sensors are installed at the inlet and outlet of the slurry dewatering process equipment, and multi-module data of the sensors are constructed in real time; PID controllers for feeding or chemical and temperature are installed at the inlet or chemical inlet of the slurry dewatering process equipment. Industrial test database construction unit: The industrial test database is constructed using multi-module sensor data and relevant fluid particle state data obtained from the numerical simulation platform; Virtual mapping model establishment and real-time data transmission and visualization unit: Based on the relevant fluid particle state data obtained from the numerical simulation platform, a virtual mapping model is established through a reduced-order model and a neural network training model to transmit data in real time and achieve visualization. Establishment of Equipment Geometry Macro Model Mounting Platform Unit: Import sensor multi-module data and slurry dewatering process equipment from the industrial test database into the equipment theoretical geometry model mounting platform through Unity 3D to establish the equipment geometry macro model mounting platform; Preliminary numerical simulations are performed in industrial simulation software. The training model process is imported, and the output results of the trained model are dynamically mapped to the virtual mapping model to achieve preliminary digital twinning and obtain digital twin results. Visualization and Control Unit: Visualizes the digital twin results in ECharts, and uses PID control to regulate the main body of the equipment, using a twin data adaptive control model and digital panel control.

Images