Self-adapting energy-saving control system for stirring of high-viscosity chemical materials

Abstract

The application discloses a self-adaptive energy-saving control system for stirring of high-viscosity chemical materials, and particularly relates to the technical field of energy-saving control, wherein the state estimation module is arranged to deduce the most real state of the system in real time through a process model and a nonlinear state transfer function; the digital twin model is arranged to continuously update itself using the latest data through data assimilation technology, so that the digital twin model is always consistent with the physical entity; the energy consumption optimization module is arranged to define an energy consumption target function, a stirring effect target function and a stirring time target function, and the NSGA-II multi-objective optimization algorithm is applied to find a Pareto optimal solution set, and the optimal energy-saving path under the current specific conditions is recalculated and generated every time a decision is made; the intelligent control module is arranged to generate stirring speed control instructions by model predictive control, and to generate material temperature control instructions by fuzzy logic control; the application realizes energy saving under the premise of ensuring control quality, and realizes predictive operation and maintenance.

Description

Technical Field

[0001] This invention relates to the field of energy-saving control technology, and more specifically, to an adaptive energy-saving control system for stirring highly viscous chemical materials. Background Technology

[0002] In the chemical industry, the mixing of high-viscosity materials is a critical and energy-intensive unit operation. The high viscosity of the materials results in poor flowability and low heat and mass transfer efficiency. In order to achieve uniform mixing and reaction temperature control, traditional mixing systems often consume a lot of energy. Therefore, it is necessary to implement energy-saving control for the mixing process of high-viscosity chemical materials.

[0003] Existing energy-saving control systems for stirring high-viscosity chemical materials collect data on motor power, mixing vessel temperature, and stirring speed using sensors. Based on historical data, an energy consumption-speed-viscosity mapping model is constructed. When the real-time motor power exceeds a set threshold, overmixing or equipment malfunction is identified, and the stirring speed is automatically reduced to a safe level. Temperature control typically employs PID control, generating temperature control commands based on the deviation between the set and measured temperatures. When power or temperature parameters exceed safe ranges, an alarm or shutdown is triggered. This system has a simple structure and low implementation cost, and can provide certain energy-saving and protective functions for production processes with stable operating conditions and minimal changes in material properties.

[0004] However, existing systems still have shortcomings: First, their rule base is static and fixed, and cannot adaptively respond to dynamic changes in material rheology, production formula adjustments, or environmental disturbances; second, they can only achieve energy saving in the waste avoidance mode, and cannot perform dynamic global optimization, so they do not know whether the current state is the most energy-efficient possible; third, in order to avoid overshoot or oscillation, PID parameters are usually set conservatively, resulting in a slow adjustment process, and requiring more time and energy to recover stability when disturbances occur. Summary of the Invention

[0005] In order to overcome the above-mentioned defects of the prior art, the present invention provides an adaptive energy-saving control system for stirring high-viscosity chemical materials, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an adaptive energy-saving control system for stirring highly viscous chemical materials, comprising: Real-time status sensing module: Acquires motor parameters, temperature parameters, mechanical parameters, process parameters, and material parameters in real time through sensors during the stirring process; The data preprocessing module verifies the integrity of the acquired raw parameter data packets, then applies the Z-score method to detect and process outliers, eliminates noise based on parameter characteristic matched filtering algorithm, and finally performs time alignment on parameters with different sampling frequencies. State estimation module: Constructs observation vector and state estimation vector and initializes state estimation vector, applies process model and nonlinear state transition function to predict state and update predicted state covariance, applies observation model to verify state estimation and update execution state, evaluates estimation quality and adaptively adjusts process noise covariance and observation noise covariance. Digital twin model update module: Receives preprocessed sensor parameters and state estimation results, updates digital twin model parameters and synchronizes digital twin model state through data assimilation technology, and applies the updated digital twin model to predict the future state of the system under different control strategies; Energy consumption optimization module: Defines the objective function and sets the constraints, applies the NSGA-II multi-objective optimization algorithm to find the Pareto optimal solution set, extracts the optimal solution from the Pareto front based on the preset decision rules, and outputs the optimal setpoint of the parameters and the expected performance index to the intelligent control module; Fault diagnosis module: Receives preprocessed sensor parameters and state estimation results and extracts multi-dimensional fault features. Then, it performs fault detection to determine the fault type, fault probability correction value and confidence level, assesses the fault severity, and generates safety action instructions based on the fault detection results and fault severity assessment results. Intelligent control module: Receives the optimal setpoint, expected performance index and current state estimate, generates stirring speed control command using model predictive control, generates material temperature control command using fuzzy logic control, and outputs control command vector to actuator drive module; Actuator drive module: Converts received safety action commands and control commands into actuator signals and executes them.

[0007] The technical effects and advantages of this invention are as follows: 1. To address the poor adaptability of existing systems, this invention incorporates a state estimation module that uses a process model and nonlinear state transition function to deduce the most realistic system state in real time. It also adaptively adjusts for process and observation noise, filtering out interference and accurately capturing subtle changes in operating conditions. Furthermore, this invention employs a digital twin model that continuously updates itself with the latest data using data assimilation technology, ensuring it remains consistent with the physical entity. As material properties change, the model itself also "evolves," guaranteeing a deep understanding of the current system state and laying the foundation for adaptive control. Finally, to address the limited energy-saving effects of existing systems, this invention incorporates energy consumption optimization. The module clearly defines the objective functions of reducing energy consumption, improving mixing efficiency, and reducing processing time. It applies the NSGA-II multi-objective optimization algorithm to perform a global search in a large parameter space to find the Pareto optimal solution set. At each decision, the optimal energy-saving path under the current specific conditions is recalculated and generated. To address the shortcomings of existing systems in the contradiction between control quality and energy saving, this invention sets up an intelligent control module that uses model predictive control to generate stirring speed control commands and fuzzy logic control to generate material temperature control commands. The combination of the two enables the system to collaboratively and intelligently handle multivariate control problems, achieving energy saving while ensuring control quality.

[0008] 2. This invention realizes the linkage between the state estimation module and the fault diagnosis module. The result of state estimation is not only used for control, but also as the input for fault diagnosis. This improves the earlyness and accuracy of fault diagnosis, because the diagnosis is based on the judgment of the true state of the system, rather than the original signal that may be contaminated by noise. This constitutes an active defense system of perception-health assessment-safety decision-making. Attached Figure Description

[0009] Figure 1 This is a system structure block diagram of the present invention.

[0010] Figure 2 This is a diagram illustrating the method steps of the present invention. Detailed Implementation

[0011] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0012] like Figure 1As shown, this embodiment provides an adaptive energy-saving control system for stirring highly viscous chemical materials, including a real-time state perception module, a data preprocessing module, a state estimation module, a digital twin model update module, an energy consumption optimization module, a fault diagnosis module, an intelligent control module, an actuator drive module, and a database. The real-time state perception module is connected to the data preprocessing module, the data preprocessing module is connected to the state estimation module, the digital twin model update module, and the fault diagnosis module, the state estimation module is connected to the digital twin model update module, the intelligent control module, the energy consumption optimization module, and the fault diagnosis module, the digital twin model update module is connected to the energy consumption optimization module, the energy consumption optimization module is connected to the intelligent control module, the fault diagnosis module and the intelligent control module are both connected to the actuator drive module, and all modules in the system are connected to the database.

[0013] The real-time status sensing module acquires motor parameters, temperature parameters, mechanical parameters, process parameters, and material parameters of the stirring process in real time through sensors. Furthermore, the motor parameters include motor torque, speed, and input power; the temperature parameters include material temperature and bearing temperature; the mechanical parameters are mechanical vibration signals; the process parameters are inlet and outlet temperatures; and the material parameters are the average particle size of the material.

[0014] In this embodiment, it should be specifically noted that: a strain gauge torque sensor can be selected to collect motor torque, installed at the motor end, with advantages including direct measurement, high accuracy, and strong anti-interference; an incremental encoder can be selected to collect motor speed, installed at the motor shaft end, with advantages including high resolution, fast response, and vibration resistance; a three-phase power sensor can be selected to collect motor input power, installed at the motor power input end, with advantages including direct measurement of active power and high accuracy; and a platinum resistance thermometer can be selected to collect material temperature, installed on the inner wall of the reactor, with advantages including high accuracy and stability. Good performance and corrosion resistance; embedded thermocouples can be selected to collect bearing temperature, installed inside the bearing housing, with advantages including fast response, high temperature resistance, and easy installation; IEPE accelerometers can be selected to collect mechanical vibration signals, installed in the bearing housing or motor housing, with advantages including wide frequency response, high sensitivity, and impact resistance; resistance temperature detectors or thermocouples can be selected to collect inlet and outlet temperatures, installed on the pipe wall, with advantages including multi-point measurement and temperature distribution monitoring; online laser particle size analyzers can be selected to collect material particle size, installed in the circulation pipeline, with advantages including real-time monitoring and representative sampling.

[0015] The data preprocessing module verifies the integrity of the acquired raw parameter data packet, then applies the Z-score method to detect and process outliers, eliminates noise based on parameter characteristic matching filtering algorithm, and finally performs time alignment on parameters with different sampling frequencies. In this embodiment, the specific execution flow of data preprocessing is as follows: Set the sliding window size and initialize the filter parameters; Receive the raw parameter data packet and execute the following steps in a loop: Check the integrity of the data packet, verify that all necessary parameters are complete, confirm the continuity of the data time series, detect whether there is data loss or transmission interruption, and proceed to the next step if the data integrity verification is successful. Parse the data packets to obtain the current value of each signal, update the sliding window data for each signal, calculate the mean and standard deviation of the sliding window, mark data points with an absolute Z-score greater than 3 as outliers, perform interpolation if the current value is an outlier; otherwise, retain the current value. For a single isolated outlier, linear interpolation of the preceding and following normal values ​​is used for replacement. For a continuous sequence of outliers, expert rule judgment is initiated. If it is determined to be an actual working condition anomaly, it is retained and marked. If it is determined to be a sensor malfunction, it is replaced with the historical trend prediction value. If there are no outliers, this step is skipped. The processed motor torque is filtered using an adaptive Kalman filter, the processed motor speed is filtered using a second-order low-pass Butterworth filter, the processed motor input power is filtered using a moving window weighted average filter, the processed material temperature and inlet / outlet temperature are filtered using an exponential weighted moving average filter and temperature gradient limitation, the processed bearing temperature is filtered using a double exponential smoothing filter, the processed mechanical vibration signal is denoised using wavelet thresholding and envelope analysis, and the processed material particle size is filtered using a Bayesian update filter. Time alignment is performed on parameters with different sampling frequencies, high-frequency parameters are downsampled, and low-frequency parameters are upsampled by interpolation. Use τ in sequence a ω d a P a a a T m a T b a T in a T out a d p a This indicates the pre-processed motor torque, motor speed, motor input power, vibration signal, material temperature, bearing temperature, inlet temperature, outlet temperature, and average particle size of the material. The processed data is timestamped and marked with data processing information before being stored in a local database. The processed data packets are sent to the state estimation module, the digital twin model update module, and the fault diagnosis module.

[0016] In this embodiment, it should be specifically noted that the Z-score formula is as follows: In the formula, x, μ, and σ are the data value at any point within the sliding window, the mean of the sliding window, and the standard deviation of the sliding window, respectively.

[0017] The state estimation module constructs observation vectors and state estimation vectors and initializes the state estimation vectors. It applies process models and nonlinear state transition functions to predict the state and update the predicted state covariance. It applies observation models to verify the state estimation and update the execution state. It evaluates the estimation quality and adaptively adjusts the process noise covariance and observation noise covariance. Furthermore, the state estimation module includes a state vector initialization unit, a derived feature calculation unit, a state prediction unit, an observation update unit, an estimation quality assessment unit, an adaptive parameter adjustment unit, a state estimation output unit, and a model learning optimization unit. The state vector initialization unit defines a state estimation vector and loads the historical process database to obtain initial parameters for similar operating conditions to initialize the extended Kalman filter parameters. The state estimation vector includes material apparent viscosity, mixing uniformity, effective shear rate, material concentration distribution, and heat transfer coefficient. The Kalman filter parameters include state covariance matrix, process noise covariance matrix, and observation noise covariance matrix. In this embodiment, it should be specifically noted that the apparent viscosity of the material, the mixing uniformity, the effective shear rate, the material concentration distribution, and the heat transfer coefficient are represented by η, U, γ, Cm, and h, respectively; and the state covariance matrix, the process noise covariance matrix, and the observation noise covariance matrix are represented by AP, AQ, and AR, respectively.

[0018] The derived feature extraction unit calculates derived feature parameters based on partially preprocessed parameters. The derived feature parameters include power number, Reynolds number, torque change rate, and temperature gradient. In this embodiment, it should be specifically noted that the power number N p The calculation formula is: In the formula P a ρ m ω b d b The parameters are, in order, the motor input power after pretreatment, the material density, the stirring speed, and the stirring paddle diameter. The formula for calculating the stirring speed is: In the formula ω a b a The parameters are, in order, the pre-processed motor speed and the reduction ratio of the reducer; the formula for calculating the Reynolds number Re is: In the formula ηpreviousη is the estimated apparent viscosity of the material at the previous moment. When performing the first state estimation, η... previous The initial value of the apparent viscosity of the material is given; the formula for calculating the torque change rate dτ / dt is: In the formula τ current τ previous Δt represents the motor torque at the current moment, the motor torque at the previous moment, and the sampling interval, respectively; the formula for calculating the temperature gradient ∇T is: In the formula L T The characteristic length of the flow path.

[0019] The state prediction unit retrieves the material apparent viscosity, mixing uniformity, effective shear rate, material concentration distribution, heat transfer coefficient state estimate and control input parameters from the previous moment, applies a nonlinear state transition function to perform state prediction, then calculates the Jacobian matrix of the nonlinear state transition function with respect to the state estimate vector, and updates the predicted state covariance matrix based on the Jacobian matrix, the state covariance matrix from the previous moment and the process noise covariance matrix from the previous moment. In this embodiment, it should be specifically noted that the state estimation vector is x=[η,U,γ,C]. m ,h] ᵀ The estimated values ​​of the material's apparent viscosity, mixing uniformity, effective shear rate, material concentration distribution, and heat transfer coefficient at the previous moment are x. k-1 =[η k-1 U k-1 ,γ k-1 C m,k-1 ,h k-1 ] ᵀ The control input parameter at the previous moment was c. k-1 =[ω b,k-1 ,T set,k-1 ]ᵀ, where ω b,k-1 T is the stirring speed at time k-1. set,k-1 The temperature is set at time k-1, and the state covariance matrix at the previous time step is AP. k-1 The process noise covariance matrix at the previous time step is AQ. k-1 .

[0020] In this embodiment, it should be specifically noted that the expression for state prediction using the nonlinear state transition function is as follows: , where x k|k-1 w k-1 The values ​​are, in order, the predicted value of the state estimation vector at time k and the process noise; for a specific state estimation vector, the state transition function for the apparent viscosity of the material is: In the formula η k|k-1 α T T ref αγ γ ref w η The parameters are, in order, the predicted apparent viscosity of the material at time k, the viscosity temperature coefficient, the reference temperature, the viscosity shear thinning coefficient, the reference shear rate, and the viscosity process noise. The state transition function for mixing uniformity is: In the formula, U k|k-1 k m ω b,k-1 w U The values ​​are, in order: predicted mixing uniformity at time k, mixing rate constant, stirring speed at time k-1, and mixing process noise. The state transition function for the effective shear rate is: In the formula γ k|k-1 k γ w γ The values ​​are, in order, the predicted effective shear rate at time k, the shear rate conversion coefficient, and the shear rate process noise. The state transition function for the material concentration distribution is: In the formula, C m,k|k-1 k c 、∇C、w C The predicted concentration distribution, concentration diffusion coefficient, concentration gradient, and concentration process noise at time k are given in sequence. The state transition function for the heat transfer coefficient is: In the formula h k|k-1 k h η ref t a τ h w h The values ​​are, in order: predicted heat transfer coefficient at time k, viscosity influence coefficient on heat transfer, reference viscosity, time elapsed in the current mixing stage, heat transfer response time constant, and process noise of the heat transfer coefficient.

[0021] In this embodiment, it is specifically necessary to explain that the Jacobian matrix F of the nonlinear state transition function f with respect to the state estimation vector x is calculated. k-1 The specific formula is: The specific formula for updating the predicted state covariance matrix based on the Jacobian matrix, the state covariance matrix of the previous time step, and the process noise covariance matrix of the previous time step is as follows: In the formula, matrix Fk-1 ᵀ Let Fk-1 be the transpose of matrix Fk-1.

[0022] The observation update unit retrieves the observation vector at the current time, applies a nonlinear observation function to calculate the observation prediction value based on the predicted state, calculates the difference between the actual observation value and the observation prediction value, calculates the Jacobian matrix and the optimal Kalman gain matrix of the observation function with respect to the state vector, and then uses the Kalman gain and observation residuals to correct the state estimate and update the state covariance matrix. In this embodiment, it should be specifically noted that the observation vector is z a =[τ a ,P a ,T m a ,∇T,d p a ] ᵀ The observation vector at the current time is z. k a =[τk a ,Pk a ,T m,k a ,∇Tk,d p,k a ] ᵀ The expression for calculating the observed predicted value based on the predicted state using a nonlinear observation function is as follows: , where z k c g(x) k|k-1 ), v k The parameters are, in order, the predicted value observed at time k, the predicted value calculated using the state prediction value, and the process noise; for a specific observation vector, the torque observation prediction formula is: In the formula τ k c k τ ω b,k n τ v τ The parameters are, in order: torque observation prediction value at time k, torque geometric constant, stirring speed at time k, torque rheological index, and torque observation noise. The power observation prediction formula is as follows: In the formula P k c NP(Re k ), ρ m d b v P The parameters are, in order: power observation prediction at time k, power number based on Reynolds number, material density, impeller diameter, and power observation noise. The temperature observation prediction formula is as follows: In the formula, T m,k c T m,k-1 a P ka m a c p h k|k-1 A c v T The parameters are as follows: predicted temperature at time k, material temperature at time k-1, motor input power at time k, material mass, specific heat capacity of the material, predicted heat transfer coefficient at time k, heat transfer area, and temperature observation noise. The formula for predicting the temperature gradient is: In the formula, ∇T k c k ∇T U k|k-1 V f v ∇T The predicted values ​​for temperature gradient at time k, temperature gradient coefficient, predicted mixing uniformity at time k, reactor volume, and temperature gradient observation noise are listed in order. The formula for predicting the average particle size of the material is as follows: In the formula d p,k c d p,0 k d γ k|k-1 t b U k|k-1 v d The values ​​are, in order: the observed and predicted average particle size of the material at time k, the initial particle diameter, the particle breakage rate constant, the predicted effective shear rate at time k, the stirring time, the predicted mixing uniformity at time k, and the observed noise of the average particle size of the material.

[0023] In this embodiment, it should be specifically noted that the difference between the observed actual value and the observed predicted value is y. k The specific formula is as follows: ; Calculate the Jacobian matrix G of the observation function g with respect to the state vector x. k The specific formula is: ; Optimal Kalman gain matrix K y,k The specific calculation formula is as follows: In the formula, matrix G k ᵀ For matrix G k The transpose of the matrix, matrix (G) k AP k|k-1 G k ᵀ +AR k ) -1 For matrix (G) k AP k|k-1 G k ᵀ +AR kThe inverse matrix of ); the specific formula for state estimation using Kalman gain and observation residual correction is: Update the state covariance matrix AP k The specific formula is: In the formula, I is the identity matrix, AP k|k-1 To predict the state covariance matrix, matrix (IK) y,k G k ) ᵀ For matrix (IK) y,k G k The transpose of matrix K, matrix K y,k ᵀ For matrix K y,k The transpose of .

[0024] The estimation quality assessment unit extracts the standard deviation of the state estimate from the updated state covariance matrix to define the estimation uncertainty. Then, it performs the innovation sequence test to check whether the observation residuals are statistically reasonable. It calculates the NEES test for the overall consistency of the state estimate and tests the physical reasonableness and trend consistency. Then, it performs a state estimate consistency assessment based on innovation consistency, NEES consistency, physical consistency, and trend consistency. Based on the state estimate consistency score, it classifies the state estimate into confidence levels.

[0025] In this embodiment, it is specifically noted that in state estimation, uncertainty can be represented by the standard deviation of the estimation error. In the Kalman filter algorithm, the covariance matrix of the state estimation is calculated, and its diagonal elements are the variances of each state estimate. The standard deviation is the square root of the variance. Therefore, the standard deviation of the state estimate can be extracted from the updated state covariance matrix to define the estimation uncertainty.

[0026] In this embodiment, it is specifically necessary to explain the state estimation consistency score C. Y The consistency score for new information is C. innovation NEES Consistency Score C NEES Physical consistency score C physical Trend consistency score C trend In turn, it is correlated with the corresponding information consistency weight w y1 NEES Consistency Weight w y2 Physical rationality weight w y3 Trend consistency weight w y4 The formula for summing the products is as follows: Information consistency weight w y1 NEES Consistency Weight w y2 Physical rationality weight w y3 The trend consistency weight wy4 takes a typical value; the specific formula for the information consistency score is: In the formula n z r k,i r max The dimensions of the observation vector, the standardized innovation of the i-th observation, and the maximum allowable value of the standardized innovation are, in order; the specific formula for the NEES consistency score is: In the formula n x ε k ε max The parameters are, in order: the dimension of the state vector, the NEES value at time K, and the maximum allowable deviation; the specific formula for the physical consistency score is: In the formula, I(physical,i) is the physical rationality indicator function of the i-th state quantity. If the state quantity is within a reasonable physical range, it is 1; otherwise, it is 0. The specific formula for the trend consistency score is: In the formula, Δx i Δx i,expected Δx i,max These are, in order: the estimated change of the i-th state vector at the current time, the expected change of the i-th state vector according to the physical model, and the allowable maximum deviation of the i-th state vector change. When the state estimation consistency score is C Y A consistency score ≥ 0.8 indicates high confidence, and a state estimation consistency score > 0.8 indicates a high degree of confidence. Y A consistency score ≥0.6 indicates medium confidence, and when the state estimation consistency score is C... Y A value less than 0.6 indicates low confidence.

[0027] The adaptive parameter adjustment unit adjusts the observation noise covariance based on the innovation sequence and adjusts the process noise covariance based on the rate of change of state. The state estimation output unit packages the state estimation results, which include all state estimates, estimation uncertainty, estimation quality flag, and timestamp. Then, it transmits the apparent viscosity and mixing uniformity state estimates, estimation quality flag, and timestamp to the energy consumption optimization module, transmits all state estimates to the intelligent control module, transmits all state estimates, estimation uncertainty, and timestamp to the digital twin module, and transmits the complete state estimation results to the fault diagnosis module. The model learning optimization unit is used to analyze state estimation errors, learn model parameters online, and accumulate process knowledge.

[0028] The digital twin model update module receives the preprocessed sensor parameters and state estimation results, updates the digital twin model parameters and synchronizes the digital twin model state through data assimilation technology, and applies the updated digital twin model to predict the future state of the system under different control strategies. Furthermore, the digital twin model update module includes a data receiving unit, a model parameter calibration unit, a model state synchronization unit, a model verification and correction unit, a prediction and scenario analysis unit, and a data output unit; The data receiving unit is used to receive preprocessed sensor parameters and state estimation results; The model parameter calibration unit updates the viscosity model parameters, heat transfer model parameters, power consumption model, and hybrid model parameters in the digital twin model through data assimilation technology. The model state synchronization unit sets the initial state of the digital twin model to the current state of the physical system, thereby synchronizing the fluid dynamics state and the energy balance state. The model verification and correction unit runs simulation in the digital twin model, compares the simulation results of the digital twin model with the subsequent real-time sensor parameters and state estimates to calculate the model error, and triggers the model correction mechanism when the model error exceeds the threshold, i.e., adaptive adjustment. The prediction and scenario analysis unit uses the updated digital twin model to predict the time and energy consumption required to achieve the target mixing degree under different control strategies, predict the viscosity change trajectory and mixing uniformity change trajectory under different speed settings, and predict the viscosity change trajectory and reaction rate change trajectory under different temperature settings. The data output unit saves the updated model parameters and status for the next update, and transmits the prediction results and scenario analysis results to the energy consumption optimization module.

[0029] The energy consumption optimization module defines the objective function and sets the constraints, applies the NSGA-II multi-objective optimization algorithm to find the Pareto optimal solution set, extracts the optimal solution from the Pareto front based on the preset decision rules, and outputs the optimal parameter setting value and expected performance index to the intelligent control module. Furthermore, the objective function includes an energy consumption objective function, a stirring effect objective function, and a stirring time objective function. The expression for the energy consumption objective function is as follows: In the formula, E and P(t) a t f The total energy consumption, the motor input power at time t, and the total stirring time are, in order. The expression for the objective function of the stirring effect is: , where U(t) f Let be the mixing uniformity at the end time. The expression for the objective function of stirring time is: The constraints include rotational speed constraints, temperature constraints, torque constraints, mixing uniformity constraints, and viscosity change rate constraints. The expression for the rotational speed constraint is: In the formula b a ω is the reduction ratio of the reducer. b,min ωb (t), ω b,max ω d,min a ω d (t)a、ω d,max a The values ​​are, in order, the minimum allowable stirring speed, the stirring speed at time t, the maximum allowable stirring speed, the minimum allowable motor speed, the motor speed at time t, and the maximum allowable motor speed. The temperature constraint expression is: In the formula, T m,min a T m (t) a T m,max a T b,min a T b (t) a T b,max a The parameters are, in order, the minimum allowable material temperature, the material temperature at time t, the maximum allowable material temperature, the minimum allowable bearing temperature, the bearing temperature at time t, and the maximum allowable bearing temperature. The torque constraint expression is as follows: In the formula, τ(t) a τ max a The motor torque and maximum permissible torque at time t are given in sequence. The expression for the mixing uniformity constraint is as follows: In the formula, U(t) f U represents the mixing uniformity at the end time. target To achieve the desired mixing uniformity, the viscosity change rate constraint expression is as follows: In the formula η max The maximum allowable viscosity change rate; the preset decision rule prioritizes the solution with the lowest energy consumption, if energy consumption is the same, prioritizes the solution with the best mixing effect, and if both energy consumption and mixing effect are the same, prioritizes the solution with the shortest stirring time; the optimal parameter setting value includes the optimal stirring speed setting value ω. b,opt and the optimal material temperature setpoint T m,opt The expected performance index is the expected energy consumption E corresponding to the optimal solution. opt Expected mixing uniformity U opt and expected mixing time t opt .

[0030] The fault diagnosis module receives the preprocessed sensor parameters and state estimation results and extracts multi-dimensional fault features. Then, it performs fault detection to determine the fault type, fault probability correction value and confidence level, assesses the fault severity, and generates safety action instructions based on the fault detection results and fault severity assessment results.

[0031] Furthermore, the fault diagnosis module includes a data receiving unit, a multi-dimensional fault feature extraction unit, a fault detection unit, a fault severity assessment unit, a safety interlock decision unit, and a decision output unit. The data receiving unit is used to receive preprocessed sensor parameters and state estimation results; The multi-dimensional fault feature extraction unit extracts device-level features, process-level features, and system-level features based on the received data. The equipment layer features include the root mean square value, peak factor, kurtosis, characteristic frequency amplitude ratio, and sideband modulation index of mechanical vibration features; the bearing temperature rise rate and temperature distribution non-uniformity of temperature features; and the mechanical efficiency index and torque fluctuation coefficient of torque-power characteristics. In this embodiment, it is specifically necessary to explain the root mean square value X of the mechanical vibration characteristics. RMS The calculation formula is: x i Let N be the i-th sampled value of the vibration signal, and N be the number of sampling points; peak factor C f The calculation formula is: ,max(∣x i |) represents the maximum absolute value of the vibration signal; the formula for calculating kurtosis is: x e The mean value of the vibration signal; the formula for calculating the characteristic frequency amplitude ratio Rf is: A bpf It is the amplitude of the bearing's passing frequency, A overall It is the overall amplitude of the entire frequency band; the formula for calculating the sideband modulation index (MI) is: A sideband It is the amplitude of the sideband frequency, A carrier It is the amplitude of the carrier frequency; the formula for calculating the bearing temperature rise rate βT is: T b,t a It is the bearing temperature at the current moment, T. b,t−1 a The bearing temperature is the temperature at the previous moment, and Δt is the sampling time interval; the formula for calculating the temperature non-uniformity UT is: , σ Tm It is the standard deviation of the material temperature at multiple temperature measurement points, μ Tm This is the average temperature at these temperature measurement points; the mechanical efficiency index η a The calculation formula is: , τ a ω d a P a The parameters, in order, are the pre-processed motor torque, motor speed, and motor input power; torque fluctuation coefficient C. τ The calculation formula is: , τ max τ min μ τ These represent the maximum, minimum, and average torque values ​​within a given time window, in that order.

[0032] The process layer features include viscosity-shear rate deviation, rheological index change rate, mixing uniformity change rate, mixing stability index, actual heat transfer coefficient deviation, and temperature response time constant. The system-level features include energy consumption per unit output, energy efficiency degradation index, torque-viscosity correlation coefficient, multi-parameter principal component score, short-term process capability coefficient, actual process capability coefficient, equipment health coefficient, and predicted remaining service life of equipment. The fault detection unit standardizes the extracted multi-dimensional fault features and inputs them into a pre-trained multi-label fault classification model for inference, obtaining the original fault probability output for each fault type. Based on the working condition, operating status, and estimation uncertainty, the unit dynamically corrects the preset fault judgment probability threshold. Based on the fault correlation matrix, the unit corrects the original fault probability for each fault type. Based on the feature importance vector, estimation uncertainty, and corrected fault probability, the unit calculates the confidence vector. When the corrected fault probability is greater than the dynamic correction threshold for fault judgment probability and the confidence is greater than the confidence threshold, the unit determines that the fault type has occurred and outputs the determined fault type, the corrected fault probability value, and the confidence. In this embodiment, it should be specifically noted that the method used by the fault detection unit to standardize the extracted multi-dimensional fault features is the same as the standardization method used when training the multi-label fault classification model; the original fault probability vector P=[P1,P2,...,P...] m ], where m is the number of fault types and θ is the basic threshold for fault determination probability. i base =[θ1 base ,θ2 base ,...,θ m base The corrected fault probability vector P'=[P1',P2',...,P m Fault correlation matrix W, feature importance vector I'=[I1',I2',...,I] n '] (representing the importance of each feature to the model's decision), confidence vector C g =[C g,1 C g,2 ,...,C g,m The specific formula for dynamically adjusting the basic threshold for fault determination probability based on operating conditions, running status, and estimated uncertainties is as follows: θ ia α operating β stage w0, U est The parameters are, in order: dynamic correction threshold for fault determination probability, operating condition adjustment coefficient, operating phase adjustment coefficient, uncertainty weight, and the maximum estimated uncertainty output by the state estimation module. The specific formula for correcting the original fault probability for each fault type based on the fault correlation matrix is ​​as follows: W ji The confidence vector is calculated based on the influence of fault j on fault i, taking into account the feature importance vector, the estimated uncertainty, and the corrected fault probability. S i It is an important subset of features for fault i.

[0033] The fault severity assessment unit retrieves the severity weights of the features for each fault type, normalizes the features of each fault type based on the absolute boundary values ​​allowed by the process, multiplies the normalized value of the fault type feature with the feature severity weight, and sums the results to obtain the fault feature severity score. It then multiplies the fault probability correction value of the fault type with the confidence level to obtain the basic fault severity score. Finally, it weights and sums the fault feature severity score and the basic fault severity score to obtain the fault severity score. In this embodiment, the specific steps for calculating the weight of each fault type feature on the severity of the fault include: calculating the structural importance of each feature under each fault type based on fault tree analysis (FTA); obtaining subjective weights using the analytic hierarchy process (AHP); calculating objective weights based on historical data using the entropy weight method; calculating dynamic weights using grey relational analysis; combining the above weights to obtain a comprehensive weight, and then verifying and dynamically updating it.

[0034] In this embodiment, it is specifically noted that each fault type feature is normalized based on the absolute boundary value allowed by the process. For features with larger feature values, the severity of the fault is higher (positively correlated features), including the root mean square value of vibration, peak factor, kurtosis, characteristic frequency amplitude ratio, sideband modulation index, bearing temperature rise rate, temperature distribution non-uniformity, torque fluctuation coefficient, viscosity-shear rate deviation, actual heat transfer coefficient deviation, energy consumption per unit output, and energy efficiency degradation index. The normalization formula used is as follows: f(x), x, x abs,min x abs,max The values ​​are, in order: normalized value of characteristic parameter, characteristic parameter, minimum absolute value of characteristic parameter allowed by process, and maximum absolute value of characteristic parameter allowed by process. For characteristics with smaller values ​​indicating higher fault severity (negatively correlated characteristics), including mechanical efficiency index, mixed stability index, short-term process capability coefficient, actual process capability coefficient, equipment health coefficient, and predicted remaining service life of equipment, the normalization formula used is: For characteristic values ​​within a preset range, the condition is considered normal; the greater the deviation from the range, the higher the severity of the fault (range deviation characteristic). These characteristics include the rheological index change rate, mixing homogeneity change rate, temperature response time constant, torque-viscosity correlation coefficient, and multi-parameter principal component scores. The normalization formula used is: ;x set,min x set,max The values ​​are, in order, the minimum and maximum values ​​of the characteristic parameters within the normal range.

[0035] In this embodiment, it should be specifically noted that for the i-th fault type, its fault characteristic severity score S g,i The specific calculation formula is as follows: In the formula n t,i f i (x j ),w t,ij The following are, in order: the number of feature parameters for the i-th fault type, the normalized value of the j-th feature parameter for the i-th fault type, and the feature severity weight of the j-th feature parameter for the i-th fault type; for the i-th fault type, its basic fault severity score S base,i The specific calculation formula is as follows: P i '、C g,i The values ​​are, in order, the fault probability correction value and the confidence level for the i-th fault type; for the i-th fault type, its fault severity score S. t,i The specific calculation formula is as follows: w g w base The weights are, in order, fault feature weights and fault basis weights, typically taken as w. g =0.6、w base =0.4.

[0036] The safety interlocking decision unit matches the response mechanism level and generates safety action instructions based on the fault detection results and fault severity assessment results. The response mechanism levels, from low to high, are: early warning level, alarm level, interlocking level, and emergency level. The decision output unit sends the generated safety action instructions to the actuator driver module.

[0037] The intelligent control module receives the optimal setpoint, expected performance index and current state estimate, uses model predictive control to generate stirring speed control command, uses fuzzy logic control to generate material temperature control command, and outputs control command vector to the actuator drive module. The actuator drive module converts the received safety action commands and control commands into actuator signals and executes them.

[0038] The database is used to store data information for all modules in the system.

[0039] In this embodiment, it should be noted that the preset values ​​and thresholds used are selected based on actual needs, and no specific value restrictions are imposed here.

[0040] like Figure 2 As shown, this embodiment provides an adaptive energy-saving control method for stirring highly viscous chemical materials, including the following steps: The motor parameters, temperature parameters, mechanical parameters, process parameters, and material parameters of the stirring process are acquired in real time through sensors. The integrity of the acquired raw parameter data packet is verified. Then, the Z-score method is applied to detect and process outliers. Noise is eliminated based on the parameter characteristics matched filtering algorithm. Finally, the parameters with different sampling frequencies are time aligned. Construct observation vectors and state estimation vectors and initialize state estimation vectors; apply process model and nonlinear state transition function to predict state and update predicted state covariance; apply observation model to verify state estimation and update execution state; evaluate estimation quality and adaptively adjust process noise covariance and observation noise covariance. The system receives preprocessed sensor parameters and state estimation results, updates the digital twin model parameters and synchronizes the digital twin model state through data assimilation technology, and applies the updated digital twin model to predict the future state of the system under different control strategies. Define the objective function and set the constraints, apply the NSGA-II multi-objective optimization algorithm to find the Pareto optimal solution set, extract the optimal solution from the Pareto front based on the preset decision rules, and output the optimal parameter settings and expected performance indicators. It receives the optimal setpoint, expected performance index and current state estimate, uses model predictive control to generate stirring speed control command, uses fuzzy logic control to generate material temperature control command, and outputs control command vector; The system receives preprocessed sensor parameters and state estimation results and extracts multi-dimensional fault features. Then, it performs fault detection to determine the fault type, fault probability correction value, and confidence level, assesses the fault severity, and generates safety action instructions based on the fault detection results and fault severity assessment results. The received safety action commands and control commands are converted into actuator signals and executed.

[0041] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An adaptive energy-saving control system for stirring highly viscous chemical materials, characterized in that: include: Real-time status sensing module: Acquires motor parameters, temperature parameters, mechanical parameters, process parameters, and material parameters in real time through sensors during the stirring process; The data preprocessing module verifies the integrity of the acquired raw parameter data packets, then applies the Z-score method to detect and process outliers, eliminates noise based on parameter characteristic matched filtering algorithm, and finally performs time alignment on parameters with different sampling frequencies. State estimation module: Constructs observation vector and state estimation vector and initializes state estimation vector, applies process model and nonlinear state transition function to predict state and update predicted state covariance, applies observation model to verify state estimation and update execution state, evaluates estimation quality and adaptively adjusts process noise covariance and observation noise covariance. Digital twin model update module: Receives preprocessed sensor parameters and state estimation results, updates digital twin model parameters and synchronizes digital twin model state through data assimilation technology, and applies the updated digital twin model to predict the future state of the system under different control strategies; Energy consumption optimization module: Defines the objective function and sets the constraints, applies the NSGA-II multi-objective optimization algorithm to find the Pareto optimal solution set, extracts the optimal solution from the Pareto front based on the preset decision rules, and outputs the optimal setpoint of the parameters and the expected performance index to the intelligent control module; Fault diagnosis module: Receives preprocessed sensor parameters and state estimation results and extracts multi-dimensional fault features. Then, it performs fault detection to determine the fault type, fault probability correction value and confidence level, assesses the fault severity, and generates safety action instructions based on the fault detection results and fault severity assessment results. Intelligent control module: Receives the optimal setpoint, expected performance index and current state estimate, generates stirring speed control command using model predictive control, generates material temperature control command using fuzzy logic control, and outputs control command vector to actuator drive module; Actuator drive module: Converts received safety action commands and control commands into actuator signals and executes them.

2. The adaptive energy-saving control system for stirring highly viscous chemical materials according to claim 1, characterized in that: The real-time status sensing module's motor parameters include motor torque, speed, and input power; temperature parameters include material temperature and bearing temperature; mechanical parameters are mechanical vibration signals; process parameters are inlet and outlet temperatures; and material parameters are the average particle size of the material.

3. The adaptive energy-saving control system for stirring highly viscous chemical materials according to claim 1, characterized in that: The state estimation module includes a state vector initialization unit, a derived feature calculation unit, a state prediction unit, and an observation update unit. The state vector initialization unit defines a state estimation vector and loads the historical process database to obtain initial parameters for similar operating conditions to initialize the extended Kalman filter parameters. The state estimation vector includes material apparent viscosity, mixing uniformity, effective shear rate, material concentration distribution, and heat transfer coefficient. The Kalman filter parameters include state covariance matrix, process noise covariance matrix, and observation noise covariance matrix. The derived feature extraction unit calculates derived feature parameters based on partially preprocessed parameters. The derived feature parameters include power number, Reynolds number, torque change rate, and temperature gradient. The state prediction unit retrieves the material apparent viscosity, mixing uniformity, effective shear rate, material concentration distribution, heat transfer coefficient state estimate and control input parameters from the previous moment, applies a nonlinear state transition function to perform state prediction, then calculates the Jacobian matrix of the nonlinear state transition function with respect to the state estimate vector, and updates the predicted state covariance matrix based on the Jacobian matrix, the state covariance matrix from the previous moment and the process noise covariance matrix from the previous moment. The observation update unit retrieves the observation vector at the current time, applies a nonlinear observation function to calculate the observation prediction value based on the predicted state, calculates the difference between the actual observation value and the observation prediction value, calculates the Jacobian matrix and the optimal Kalman gain matrix of the observation function with respect to the state vector, and then uses the Kalman gain and observation residuals to correct the state estimate and update the state covariance matrix.

4. The adaptive energy-saving control system for stirring high-viscosity chemical materials according to claim 3, characterized in that: The state estimation module further includes an estimation quality assessment unit, an adaptive parameter adjustment unit, a state estimation output unit, and a model learning optimization unit; The estimation quality assessment unit extracts the standard deviation of the state estimate from the updated state covariance matrix to define the estimation uncertainty. Then, it performs the innovation sequence test to check whether the observation residuals are statistically reasonable. It calculates the NEES test for the overall consistency of the state estimate and tests the physical reasonableness and trend consistency. Then, it performs a state estimate consistency assessment based on innovation consistency, NEES consistency, physical consistency, and trend consistency. Based on the state estimate consistency score, it classifies the state estimate into confidence levels. The adaptive parameter adjustment unit adjusts the observation noise covariance based on the innovation sequence and adjusts the process noise covariance based on the rate of change of state. The state estimation output unit packages the state estimation results, which include all state estimates, estimation uncertainty, estimation quality flag, and timestamp. Then, it transmits the apparent viscosity and mixing uniformity state estimates, estimation quality flag, and timestamp to the energy consumption optimization module, transmits all state estimates to the intelligent control module, transmits all state estimates, estimation uncertainty, and timestamp to the digital twin module, and transmits the complete state estimation results to the fault diagnosis module. The model learning optimization unit is used to analyze state estimation errors, learn model parameters online, and accumulate process knowledge.

5. The adaptive energy-saving control system for stirring highly viscous chemical materials according to claim 1, characterized in that: The digital twin model update module includes a data receiving unit, a model parameter calibration unit, a model state synchronization unit, a model verification and correction unit, a prediction and scenario analysis unit, and a data output unit. The data receiving unit is used to receive preprocessed sensor parameters and state estimation results; The model parameter calibration unit updates the viscosity model parameters, heat transfer model parameters, power consumption model, and hybrid model parameters in the digital twin model through data assimilation technology. The model state synchronization unit sets the initial state of the digital twin model to the current state of the physical system, thereby synchronizing the fluid dynamics state and the energy balance state. The model verification and correction unit runs simulation in the digital twin model, compares the simulation results of the digital twin model with the subsequent real-time sensor parameters and state estimates to calculate the model error, and triggers the model correction mechanism when the model error exceeds the threshold, i.e., adaptive adjustment. The prediction and scenario analysis unit uses the updated digital twin model to predict the time and energy consumption required to achieve the target mixing degree under different control strategies, predict the viscosity change trajectory and mixing uniformity change trajectory under different speed settings, and predict the viscosity change trajectory and reaction rate change trajectory under different temperature settings. The data output unit saves the updated model parameters and status for the next update, and transmits the prediction results and scenario analysis results to the energy consumption optimization module.

6. The adaptive energy-saving control system for stirring high-viscosity chemical materials according to claim 1, characterized in that: The objective functions of the energy consumption optimization module include an energy consumption objective function, a stirring effect objective function, and a stirring time objective function. The expression for the energy consumption objective function is as follows: In the formula, E and P(t) a t f The total energy consumption, the motor input power at time t, and the total stirring time are, in order. The expression for the objective function of the stirring effect is: , where U(t) f Let be the mixing uniformity at the end time. The expression for the objective function of stirring time is: The constraints include rotation speed constraints, temperature constraints, torque constraints, mixing uniformity constraints, and viscosity change rate constraints. The preset decision rule prioritizes the solution with the lowest energy consumption; if energy consumption is the same, prioritize the solution with the best mixing effect; if energy consumption and mixing effect are the same, prioritize the solution with the shortest stirring time. The optimal parameter settings include the optimal stirring speed setting and the optimal material temperature setting. The expected performance indicators are the expected energy consumption, expected mixing uniformity, and expected stirring time corresponding to the optimal solution.

7. The adaptive energy-saving control system for stirring highly viscous chemical materials according to claim 1, characterized in that: The fault diagnosis module includes a data receiving unit, a multi-dimensional fault feature extraction unit, and a fault detection unit. The data receiving unit is used to receive preprocessed sensor parameters and state estimation results; The multi-dimensional fault feature extraction unit extracts device-level features, process-level features, and system-level features based on the received data. The equipment layer features include the root mean square value, peak factor, kurtosis, characteristic frequency amplitude ratio, and sideband modulation index of mechanical vibration features; the bearing temperature rise rate and temperature distribution non-uniformity of temperature features; and the mechanical efficiency index and torque fluctuation coefficient of torque-power characteristics. The process layer features include viscosity-shear rate deviation, rheological index change rate, mixing uniformity change rate, mixing stability index, actual heat transfer coefficient deviation, and temperature response time constant. The system-level features include energy consumption per unit output, energy efficiency degradation index, torque-viscosity correlation coefficient, multi-parameter principal component score, short-term process capability coefficient, actual process capability coefficient, equipment health coefficient, and predicted remaining service life of equipment. The fault detection unit standardizes the extracted multi-dimensional fault features and inputs them into a pre-trained multi-label fault classification model for inference, obtaining the original fault probability output for each fault type. Based on the operating conditions, running status, and estimated uncertainty, the unit dynamically corrects the preset fault judgment probability threshold. Based on the fault correlation matrix, the unit corrects the original fault probability for each fault type. Based on the feature importance vector, estimated uncertainty, and corrected fault probability, the unit calculates the confidence vector. When the corrected fault probability is greater than the dynamic correction threshold for fault judgment probability and the confidence is greater than the confidence threshold, the unit determines that the fault type has occurred and outputs the determined fault type, the corrected fault probability value, and the confidence.

8. The adaptive energy-saving control system for stirring high-viscosity chemical materials according to claim 7, characterized in that: The fault diagnosis module includes a fault severity assessment unit, a safety interlock decision unit, and a decision output unit. The fault severity assessment unit retrieves the severity weights of the features for each fault type, normalizes the features of each fault type based on the absolute boundary values ​​allowed by the process, multiplies the normalized value of the fault type feature with the feature severity weight, and sums the results to obtain the fault feature severity score. It then multiplies the fault probability correction value of the fault type with the confidence level to obtain the basic fault severity score. Finally, it weights and sums the fault feature severity score and the basic fault severity score to obtain the fault severity score. The safety interlocking decision unit matches the response mechanism level and generates safety action instructions based on the fault detection results and fault severity assessment results. The response mechanism levels, from low to high, are: early warning level, alarm level, interlocking level, and emergency level. The decision output unit sends the generated safety action instructions to the actuator driver module.

Images