A roll press control system and method based on online simulation and deep learning

Abstract

The present invention discloses a control system and method for a roller press based on online simulation and deep learning, belonging to the technical field of energy saving of roller presses. By combining online monitoring technology, CAE dynamic simulation technology and deep learning technology, the present invention conducts intelligent control and management of the roller press system, ensures the stable operation of the roller press, improves the production efficiency and service life of the roller press, reduces the grinding energy consumption, and improves the management level. In addition, the present invention provides a complete control system and method for optimizing the production capacity and life of the roller press, integrating the advantages of real-time and high-efficiency online monitoring, the advantage of accurately capturing the micro-particle dynamics by CAE dynamic simulation technology, and the advantage of rapid response of deep learning technology, and quickly calculates and instantaneously responds to control the actual complex and changeable working conditions of the roller press.

Description

Technical Field

[0001] The present invention belongs to the technical field of roller press energy saving, and specifically, relates to a roller press control system and method based on online simulation and deep learning. Background Art

[0002] Since the emergence of roller press technology in the 1980s, as an energy-saving and reliable crushing device, it has been increasingly used in the cement and mining industries. Compared with other crushing technologies, roller press technology has the advantages of high flexibility, good wear resistance, high output and low energy consumption. The principle of roller press crushing is that the loose cement particles move downward under the pressure of the upper particles, and are squeezed at the roller inlet to form a compact material layer. After the high pressure between the roller presses, the cement particles generate a large strong interaction force and are crushed or cracked for subsequent crushing.

[0003] However, in the actual production process, due to factors such as long-term operation of the equipment, complex and changeable working conditions, and improper operation and maintenance, the production capacity of the roller press is easily reduced and the service life is shortened, which brings great economic losses to the enterprise. At present, the management of the roller press mainly relies on regular shutdown and maintenance and manual experience judgment. This method is not only inefficient, but also difficult to accurately and timely discover potential problems of the equipment. Therefore, how to achieve the optimal management of the production capacity and life of the roller press and improve the operation efficiency and reliability of the equipment has become a technical problem to be solved in this field. In the prior art, various sensors are installed at key parts of the equipment to collect the operation status data of the equipment in real time, and big data machine learning analysis is performed to perform empirical analysis and intelligent control. Online monitoring technology has the advantages of strong real-time performance and large data volume, and can timely detect abnormal status and fault warning information of the equipment. However, purely relying on big data analysis cannot consider the essential dynamic mechanism of equipment operation, and it is difficult to fully reflect the operation status and performance of the equipment, resulting in very limited optimization and control effects. Numerical simulation technology can predict the service life and production capacity of the equipment, optimize the operation parameters, and improve the production efficiency and operation stability of the equipment by establishing a dynamic simulation model of the roller press and performing simulation analysis. Numerical simulation technology can provide accurate performance prediction and optimization solutions to make up for the shortcomings of online monitoring technology. However, numerical simulation has the problem of untimely calculation. Especially during the operation of the equipment, the calculation complexity of real-time simulation is high, and it is difficult to quickly provide optimization suggestions and decision support. It is currently mainly used in the scientific research and development stage or the problem diagnosis stage, and it is impossible to achieve real-time calculation to guide production. In order to solve the above problems, the present invention provides the following technical solutions. Summary of the invention

[0004] The object of the present invention is to provide a roller press control system and method based on online simulation and deep learning, which solves the problems in the prior art that the working condition adjustment of the intelligent control system of the roller press ignores real-time changes, the historical data can only reflect the past operating conditions, and it is difficult to adapt to the real-time changes of the current working conditions. The changing factors in the production process (such as raw material quality, environmental conditions, equipment status, etc.) may cause the historical data to become invalid, and the production parameters cannot be adjusted in time; there is a lack of prediction ability, and the historical data analysis cannot effectively predict future production trends and potential problems, making it difficult to make forward-looking decisions and preventive maintenance; relying solely on sensor monitoring data is costly, and the particle dynamics data inside the roller press cannot be obtained.

[0005] The object of the present invention can be achieved by the following technical solutions:

[0006] A roller press control system based on online simulation and deep learning, including a sensor module, a deep learning module, a numerical simulation module, a control module, an actuator, and a human-machine interface;

[0007] The sensor module collects the operation status data of the equipment in real time and transmits this data to the control module, the deep learning module, and the numerical simulation module;

[0008] The deep learning module processes and analyzes the real-time data and the numerical simulation results, identifies the operation mode and the fault mode, and provides optimization suggestions and warning information to the control module;

[0009] The numerical simulation module conducts simulation analysis based on the real-time data, captures the microscopic particle dynamics data inside the roller press, predicts the service life and production capacity of the equipment, and transmits the simulation results to the deep learning module and the control module;

[0010] The control module makes decisions and optimizes the control according to the deep learning results and the simulation results, generates control instructions and sends them to the actuator;

[0011] The actuator adjusts the operation parameters of the roller press according to the control instructions;

[0012] The human-machine interface provides an operation interface.

[0013] A roller press control method based on online simulation and deep learning, including the following steps:

[0014] S1. Establish a numerical simulation model

[0015] Based on the particle dynamics simulation software, use DEM to model and simulate the particle dynamics behavior and the particle-geometry interaction behavior, set the boundary conditions according to the actual working parameters of the roller press, and submit the calculation;

[0016] Analyze the results of the discrete element method (DEM) particle dynamics simulation of the roller press, and explore the changes in the particle size of the particles entering and leaving the rollers of the roller press, as well as the effects of the power and wear of the roller press;

[0017] S2. Verification and adjustment of simulation results

[0018] Adjust the simulation model by comparing the simulation results with the actual production data;

[0019] S3. Output simulation data

[0020] After the simulation model is verified, the simulation results are output to the control module and the deep learning module;

[0021] S4. Establish a deep learning analysis and prediction model

[0022] The deep learning module analyzes the production data and simulation results of the plant area to establish a deep learning analysis and prediction model for predicting the key performance indicators of the roller press;

[0023] S5. Verification and adjustment of the deep learning analysis and prediction model

[0024] Based on the actual production data and simulation results of the plant area, verify the deep analysis and prediction model, and use the Bayesian optimization method to adjust the model parameters;

[0025] S6. Output analysis data

[0026] After the deep learning analysis and prediction model is verified, the analysis results are output to the control module, and the control module uses these data to provide optimization decision support.

[0027] Advantages of the present invention:

[0028] 1. By combining the online monitoring technology, CAE dynamic simulation technology and deep learning technology, the present invention conducts intelligent control and management of the roller press system, ensuring the stable operation of the roller press, improving the production efficiency and service life of the roller press, reducing the grinding energy consumption, and improving the management level.

[0029] 2. The present invention provides a complete set of optimization management control systems and methods for the production capacity and life of the roller press, integrating the advantages of online monitoring in real time and high efficiency, the advantages of CAE dynamic simulation technology in accurately capturing microscopic particle dynamics, and the advantages of deep learning technology in rapid response, and quickly calculating and instantaneously responding to control the actual complex and changeable working conditions of the roller press. Brief description of the drawings

[0030] The present invention will be further described below with reference to the accompanying drawings.

[0031] Figure 1 It is a schematic diagram of the framework structure of a roller press control system based on online simulation and deep learning;

[0032] Figure 2 It is a schematic flow chart of a control method for a roller press based on online simulation and deep learning;

[0033] Figure 3 It is a schematic flow chart for optimizing the roller press system. Specific embodiments

[0034] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the protection scope of the present invention.

[0035] A control system for a roller press based on online simulation and deep learning includes a sensor module, a deep learning module, a numerical simulation module, a control module, an actuator, and a human-machine interface;

[0036] As Figure 1 shown, during operation, the sensor module collects the operation status data of the equipment in real time and transmits this data to the control module, the deep learning module, and the numerical simulation module;

[0037] The deep learning module processes and analyzes the real-time data and the numerical simulation results, identifies the operation mode and the fault mode, and provides optimization suggestions and warning information to the control module.

[0038] The numerical simulation module performs simulation analysis based on the real-time data, captures the microscopic particle dynamics data inside the roller press, predicts the service life and production capacity of the equipment, and transmits the simulation results to the deep learning module and the control module.

[0039] The control module makes decisions and optimizes the control according to the deep learning results and the simulation results, generates control instructions and sends them to the actuator.

[0040] The actuator adjusts the operation parameters of the roller press according to the control instructions to ensure that the equipment operates under the optimal working conditions.

[0041] The human-machine interface provides an interaction interface between the operator and the control module, displays the real-time data, the analysis results, and the simulation results, and receives operation instructions.

[0042] Through the collaborative work of these modules, the system can achieve intelligent control and management of the roller press, ensure the stable operation of the roller press, and improve the production efficiency and the service life of the equipment.

[0043] The present invention provides a control method for a roller press based on online simulation and deep learning, as Figure 2 andFigure 3 As shown in the figure, it includes the following steps:

[0044] S1. Establish a numerical simulation model. Based on particle dynamics simulation software such as Rocky and EDEM, the discrete element method (DEM) is used to model and simulate the particle dynamics behavior and particle-geometry interaction behavior. Set boundary conditions, material parameters, particle parameters, particle-particle and particle-geometry interaction parameters, etc. according to the actual working parameters of the roller press, and submit the calculation. This model can simultaneously achieve: (1) particle breakage simulation; (2) calculation simulation of equipment operating power; (3) simulation and prediction of equipment wear; (4) simulation of particle-particle and particle-wall interaction.

[0045] Analyze the results of the DEM particle dynamics simulation of the roller press, extract the particle size distribution data graphs of the incoming and outgoing rollers and the operating power data graphs of each equipment (moving roller, fixed roller and booster device), and draw the contact force nephogram of particles in the feeding pipe, the streamline graph of particle flow and the volume reduction and stress nephogram of the equipment wall wear.

[0046] Based on the particle dynamics simulation technology, by jointly adjusting the feeding position, shape, incoming material distribution and the cooperation of the hydraulic system of the roller press, explore the influence of the roller press on the particle size change of the incoming and outgoing rollers, the power and wear of the roller press, and then optimize the roller press system.

[0047] S2. Verify and adjust the simulation results. Based on the actual production data of the plant area, verify the simulation results. By comparing the simulation results and the actual production data, adjust the simulation model to improve the accuracy and reliability of the model.

[0048] During the model verification process, use the Bayesian optimization method to adjust the simulation model parameters:

[0049]

[0050] Among them, θ is the model parameter; D is the observed data; p(θ∣D) is the posterior probability distribution of the parameter, which is calculated by Bayes' theorem:

[0051]

[0052] S3. Output the simulation data. After the simulation model is verified, output the simulation results to the control module and the deep learning module. The control module uses the simulation data to provide decision support, and the deep learning module uses the simulation data for further analysis.

[0053] S4. Establish a deep learning analysis and prediction model. The deep learning module analyzes the plant production data and simulation results to establish a deep learning analysis and prediction model. This model is used to predict key performance indicators such as the operating status, production capacity, and lifespan of the roller press.

[0054] S5. Validate and adjust the deep learning analysis and prediction model. Based on the actual plant production data and simulation results, verify the analysis and prediction model. By comparing the prediction results of the analysis and prediction model with the actual data, use the Bayesian optimization method to adjust the model parameters to improve the prediction accuracy of the model.

[0055] S6. Output analysis data. After the deep learning analysis and prediction model is tested, output the analysis results to the control module. The control module uses this data to provide optimization decision support, including adjusting the operating conditions, whether to perform equipment maintenance, ensuring that the roller press operates under optimal conditions, and improving production efficiency and the service life of the equipment.

[0056] The control module is the core part of the entire system, responsible for receiving the data from the sensor module, the simulation results of the numerical simulation module, and the instructions from the deep learning module, and making decisions according to the preset control strategy. The control module interacts with the operator through the human-machine interface, receives operation instructions and displays the system operating status. The actuator actually adjusts the operating parameters of the roller press, such as the roll gap distance, the given pressure, and the feeding amount, according to the instructions of the control module, to ensure that the roller press operates under optimal conditions. The actuator includes components such as the hydraulic system, the motor, and the transmission device, and can quickly respond to control instructions to ensure the efficient operation of the system.

[0057] Among them, in order to achieve precise control and optimization of the roller press, a discrete-time state-space model of the roller press system is established using the Model Predictive Control (MPC) strategy as follows:

[0058] x k+1 = Ax k + Bu k + w k

[0059] y k = Cx k + v k

[0060] Among them, x k ∈R n : The system state vector, including the output, roll gap, and temperature of the roller press; u k ∈R m : The control input vector, including the feeding amount and the given pressure of the roller press; y k ∈R p: Output measurement vector, including sensor measurement data; A, B, C: System matrices, describing system dynamics; W k , v k : Process noise and measurement noise, assumed to be zero-mean Gaussian noise;

[0061] Within the prediction horizon N p , by optimizing the control input sequence U(k) = {u(k), u(k+1),..., u(k+N p -1)} to minimize the following objective function:

[0062]

[0063] where, N p : Prediction horizon length, y ref,k+i : Reference output at the (k+i)-th time; Δu k+i-1 = u k+i-1 - u k+i-2 : Increment of the control input; Weighted quadratic norm; Q, R: Positive definite weighting matrices, weighing the costs of tracking error and control increment respectively; Subsequently, a Quadratic Programming (QP) problem is formed based on this objective function:

[0064]

[0065] s.t. EΔU ≤ F

[0066] where, E, F: Matrices and vectors constructed from system constraint conditions. Optimization algorithms such as the interior point method or the active set method are used to solve this QP problem to obtain the optimal control sequence. In practical applications, only the first control input is executed, and then rolling optimization is performed. Since the model parameters of the roller press may change over time, an Adaptive Control strategy is adopted to improve the robustness of the system. The Recursive Least Squares (RLS) method is used to estimate the system model parameters online:

[0067] θ k = θ k+1 + L k (y k - φ T k θ k-1 )

[0068]

[0069] P k = P k-1 - L k φT k P k-1

[0070] where θ k : the estimated model parameter vector; φ k : the regression vector, composed of input and output data; P k : the covariance matrix. By updating the model parameters online, the controller can adapt to the dynamic changes of the system and maintain good control performance.

[0071] The sensor module is responsible for real-time monitoring of the operating status and environmental parameters of the roller press. The sensors include temperature sensors, pressure sensors, vibration sensors, moving roller displacement sensors, rotational speed sensors, weighing sensors, etc., which are installed at key parts such as the roller shaft, hydraulic cylinder and transmission device, and weighing bin of the roller press. The sensor module transmits the collected real-time data to the control module, numerical simulation module and deep learning module through the data bus. The real-time monitoring function of the sensor module ensures that the system can timely detect the abnormal status and fault warning information of the equipment, improving the reliability and safety of the equipment. To improve the accuracy and robustness of the sensor data, a Fusion Kalman Filter is used to fuse multi-source data. For the data of multiple sensors:

[0072] z k = Hx k + v k

[0073] where Z k : the fused measurement data; H: the observation matrix; measurement noise; v k : the measurement noise, assumed to be Gaussian white noise.

[0074] The human-machine interface provides a friendly and intuitive operation interface. Through the human-machine interface, the operator can view real-time monitoring data, system operating status, fault alarm information, numerical simulation results, numerical simulation predictions, deep learning analysis results, etc., and perform parameter settings and input of control instructions. The human-machine interface also has a data visualization function, which displays complex data results in the form of charts, graphs, etc., facilitating the operator's understanding and decision-making.

[0075] The deep learning module utilizes a CNN-LSTM neural network to process and analyze the real-time data collected by the sensor module and the numerical simulation results. Through data mining and machine learning algorithms, the deep learning module can analyze historical data, discover the rules and trends of equipment operation, identify the operation modes and fault characteristics of the roller press, and provide optimization suggestions. At the same time, based on the intelligent analysis of long-term data, it evolves the working conditions, processes new data in real time, predicts changes in working conditions, captures possible fault situations, issues early warnings, and assists on-site personnel in making decisions and optimizing control.

[0076] The numerical simulation module, based on the discrete element simulation technology, establishes a dynamic simulation model of the roller press and conducts simulation analysis. The simulation model takes into account factors such as the structural characteristics, material characteristics, and multi-body operation conditions of the roller press, and can accurately capture the dynamic behavior of microscopic particles. The numerical simulation module can predict the production capacity and usage of the equipment, optimize the operation parameters, and provide accurate performance predictions and optimization solutions. The simulation results are transmitted to the deep learning module and the control module in real time to provide decision-making support for them.

[0077] In the discrete element simulation, the motion of each particle is described by Newton's equations of motion:

[0078]

[0079]

[0080] where mi is the mass of particle i; is the translational velocity of the particle; and are the normal contact force and the tangential contact force between particle i and particle j, respectively; is the force exerted by the gas on particle i; Ii is the moment of inertia of the particle; is the angular velocity of the particle; is the vector pointing from the particle center to the contact point; μr is the rolling friction coefficient affected by particle i. The interaction calculation between particles uses the non-linear Hertz model:

[0081]

[0082]

[0083]

[0084]

[0085] where E′ is the equivalent Young's modulus, R′ is the equivalent radius, δn is the normal overlap, m′ is the equivalent mass, is the relative normal velocity, e is the restitution coefficient, G′ is the equivalent shear modulus, δt is the tangential overlap, is the relative tangential velocity, and μs is the sliding friction coefficient.

[0086] The Archard model is used to calculate the wear amount:

[0087]

[0088] In the formula, ΔV is the wear volume of the geometric body within a simulation time step; ΔW τ is the work done by the contact force of the particles on the surface of the geometric body within a time step, in J; C is the wear experience constant, and the smaller this value is, the more wear-resistant the geometric material is; F τ is the tangential component of the contact force between the particles and the surface of the geometric body; Δs τ is the relative displacement of the particles in the tangential plane within a time step.

[0089] The above content is only an example and illustration of the present invention. Those skilled in the art of this technology can make various modifications or supplements to the described specific embodiments or use similar methods to replace them. As long as they do not deviate from the invention or exceed the scope defined by this claim book, they shall fall within the protection scope of the present invention.

Claims

1. A roller press control system based on online simulation and deep learning, characterized in that: It includes sensor module, deep learning module, numerical simulation module, control module, actuator and human-machine interface; The sensor module collects the operating status data of the equipment in real time and transmits the data to the control module, deep learning module and numerical simulation module; The deep learning module processes and analyzes real-time data and numerical simulation results, identifies operating modes and failure modes, and provides optimization suggestions and early warning information to the control module; The numerical simulation module performs simulation analysis based on real-time data, captures the microscopic particle dynamics data inside the roller press, predicts the service life and production capacity of the equipment, and passes the simulation results to the deep learning module and the control module; The control module makes decisions and optimizes control based on the deep learning results and simulation results, and generates control instructions to send to the actuators; The actuator adjusts the operating parameters of the roller press according to the control instructions; The human-machine interface provides an operation interface; A roller press control method based on online simulation and deep learning according to a roller press control system includes the following steps: S1. Establishing numerical simulation model Based on the particle dynamics simulation software, DEM is used to model and simulate the particle dynamics behavior and particle-geometry interaction behavior, and the boundary conditions are set according to the actual working parameters of the roller press, and the calculation is submitted; The results of the DEM particle dynamics simulation of the roller press were analyzed to explore the influence of the roller press on the change of the particle size of the particles entering and exiting the roller, and the power and wear of the roller press. S2. Verification and adjustment of simulation results Adjust the simulation model by comparing the simulation results with the actual production data; S3. Output simulation data After the simulation model is verified, the simulation results are output to the control module and the deep learning module; S4. Establish a deep learning analysis and prediction model The deep learning module analyzes the plant production data and simulation results, and establishes a deep learning analysis and prediction model to predict the key performance indicators of the roller press; S5. Verification and adjustment of deep learning analysis and prediction models Based on the actual production data and simulation results of the plant, the deep analysis and prediction model was verified, and the model parameters were adjusted using the Bayesian optimization method; S6. Output analysis data After the deep learning analysis and prediction model is verified, the analysis results are output to the control module, which uses these data to provide optimization decision support; The discrete time state space model of the roller press system established using MPC is: x k+1 =Ax k +Bu k +w k y k =Cx k +v k Among them, x k ∈R n : system state vector, including the output, roll gap, and temperature of the roller press; u k ∈R m : Control input vector, including roller press feed rate and given pressure; y k ∈R p : Output measurement vector, including sensor measurement data; A, B, C: System matrix, describing the system dynamics characteristics; w k 、v k : Process noise and measurement noise, assumed to be zero-mean Gaussian noise; In the prediction domain N p By optimizing the control input sequence U(k)={u(k),u(k+1),…,u(k+N p -1)} minimize the following objective function: Among them, N p : Prediction time domain length, y ref,k+i : Reference output at the k+ith moment; Δu k+i-1 =u k+i-1 -u k+i-2 : Control the input increment; Weighted quadratic norm; Q, R: positive definite weighted matrices, weighing the cost of tracking error and control increment respectively; then a quadratic programming problem is formed based on the objective function: stEΔU≤F in, E, F: matrices and vectors constructed by system constraints. The optimization algorithm is used to solve the quadratic programming problem and obtain the optimal control sequence.

2. The roller press control system based on online simulation and deep learning according to claim 1, characterized in that: During the model validation process in step S2, the simulation model parameters are adjusted using the Bayesian optimization method: Where θ is the model parameter; D is the observed data; p(θ|D) is the posterior probability distribution of the parameter, calculated using Bayes’ theorem:

3. The roller press control system based on online simulation and deep learning according to claim 1, characterized in that: After obtaining the optimal control sequence, only the first control input is executed, followed by rolling optimization.

4. The roller press control system based on online simulation and deep learning according to claim 3 is characterized in that: The model parameters are also updated online by: Estimate system model parameters online using recursive least squares: i k =θ k+1 +L k (y k -f T k i k-1 ) P k =P k-1 -L k φ T k P k-1 Among them, θ k : estimated model parameter vector; φ k : regression vector, consisting of input and output data; P k : covariance matrix.

5. The roller press control system based on online simulation and deep learning according to claim 3 is characterized in that: The sensor module collects the running status data of the device in real time and uses a fusion Kalman filter to fuse the data of multiple sensors: z k =Hx k +v k Among them, z k : fused measurement data; H: observation matrix; measurement noise; v k : Measurement noise, assumed to be Gaussian white noise.

6. The roller press control system based on online simulation and deep learning according to claim 5, characterized in that: The numerical simulation module establishes a dynamic simulation model of the roller press and performs simulation analysis. In the discrete element simulation, the motion of each particle is described by Newton's equation of motion: Where mi is the mass of particle i; is the translational velocity of the particle; and are the normal contact force and tangential contact force between particles i and j, respectively; is the force exerted by the gas on particle i; Ii is the moment of inertia of the particle; is the angular velocity of the particle; is the vector pointing from the center of the particle to the contact point; μr is the rolling friction coefficient affected by particle i.

7. The roller press control system based on online simulation and deep learning according to claim 6, characterized in that: The interaction between particles is calculated using the nonlinear Hertz model: Among them, E' is the equivalent Young's modulus, R' is the equivalent radius, δ n is the normal overlap, m′ is the equivalent mass, is the relative normal velocity, e is the coefficient of restitution, G′ is the equivalent shear modulus, δ t is the tangential overlap, is the relative tangential velocity, μ s is the sliding friction coefficient.

8. The roller press control system based on online simulation and deep learning according to claim 7, characterized in that: The wear volume is calculated using the Archard model: Where ΔV is the wear volume of the geometric body in one simulation time step; is the work J done by the contact force of the particle on the surface of the geometric body in one time step; C is the wear empirical constant, the smaller the value, the more wear-resistant the geometric material; is the tangential component of the contact force between the particle and the surface of the geometric body; is the relative displacement of the particle on the tangential plane within one time step.

Images