Lime milk impurity removal system and method for calcium silicate production

Abstract

The invention relates to the technical field of industrial automation and chemical engineering data processing, in particular to a lime milk impurity removal system and method for calcium silicate production, and the method comprises the steps: collecting data in real time through a data collection module to generate a time sequence multi-dimensional feature vector; the impurity dynamic deduction module simulates an impurity movement track based on a three-dimensional geometric model and outputs prediction data; the multi-target collaborative optimization decision-making module generates a Pareto optimal process scheme by taking the vibration frequency of a vibrating screen, the retention time of a sedimentation tank, the aperture combination of a jumping screen and the adding amount of a flocculating agent as decision variables; the dynamic scheduling module converts the scheme into an executable instruction to control the operation of the physical equipment device; and the virtual-real interaction closed-loop correction module generates a deviation signal by comparing measured data and predicted data of the optical detection unit and the spectrum analyzer, and updates model parameters and weight coefficients by using a federated learning mechanism to form closed-loop correction.

Description

Technical Field

[0001] This invention relates to the fields of industrial automation and chemical engineering data processing technology, and in particular to a lime slurry purification system and method for calcium silicate production. Background Technology

[0002] In existing calcium silicate production, quality control begins with the high-purity screening of raw materials, as low impurity content helps reduce the formation of by-products, thereby improving the chemical homogeneity of the product. Temperature and pressure parameters during the reaction stage need to be precisely controlled to promote the orderly formation of calcium silicate crystals, thus optimizing the material's mechanical properties. Online monitoring technology tracks process variables in real time, and data feedback allows for timely adjustments to operating conditions, maintaining the stability of the production process. The final product undergoes multi-dimensional testing, including physical and chemical analysis, to verify its compliance with application standards, achieving reliable overall quality management.

[0003] Existing quality control technologies for calcium silicate production suffer from the following technical challenges: The purification processes for raw materials and process fluids are inefficient. For example, in the lime slurry preparation stage, unburned limestone and incompletely combusted carbon particles are difficult to completely separate through conventional sedimentation and simple filtration, and these fine particles can enter the synthesis process along with the lime slurry. Simultaneously, although dust impurities in recycled wastewater are treated by sedimentation, micron-sized suspended solids easily penetrate existing filtration units and accumulate in the system. Colloidal suspended matter in sodium silicate solutions also lacks online monitoring and efficient removal methods. These impurities, as causes of black spots, directly incorporate into the calcium silicate crystal structure, leading to fluctuations in the whiteness and purity of the finished product. Conventional detection methods relying on offline sampling are lagging and cannot provide real-time early warning and feedback control for impurity introduction, making it difficult to maintain product quality stability. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a lime slurry impurity removal system and method for calcium silicate production. This invention solves the technical problem that solid impurities and suspended solids in the raw materials (lime slurry, sodium silicate solution) and production wastewater of calcium silicate synthesis are difficult to effectively monitor and remove, resulting in black spot defects in the final product and making it impossible to stably determine product quality control indicators (such as whiteness and purity) through conventional testing methods.

[0005] To solve the above-mentioned technical problems, the specific contents of the present invention are as follows: In a first aspect, the present invention provides a lime slurry impurity removal system for calcium silicate production, comprising a physical device and a control device, wherein the control device establishes a communication connection with the physical device. The physical equipment includes a lime ash reactor, a vibrating screen, a primary sedimentation tank, a jumping screen, a two-stage sedimentation tank, a precision filter, a high-speed centrifuge, and a chemical flocculation reactor connected in sequence, as well as an optical detection unit and a spectrometer located at the outlet of the two-stage sedimentation tank; The control device includes a data acquisition module, an impurity dynamic simulation module, a multi-objective collaborative optimization decision-making module, a dynamic scheduling module, and a virtual-real interactive closed-loop correction module. The data acquisition module is configured to collect data on the particle size distribution, spectral characteristics, turbidity, and pH value of lime milk from the physical equipment, and to perform noise reduction and timestamp alignment on the collected data to generate a time-series multidimensional feature vector. The impurity dynamic simulation module is configured to receive time-series multidimensional feature vectors, simulate the movement trajectory of impurities based on the three-dimensional geometric model of the physical device, simulate the movement trajectory of impurities through computational fluid dynamics and discrete element method, and dynamically correct simulation parameters using long short-term memory network to output predicted data on impurity types, concentrations and evolution trends. The multi-objective collaborative optimization decision module is configured to receive prediction data and use the vibration frequency of the vibrating screen, the residence time of the sedimentation tank, the combination of screen apertures, and the amount of flocculant added as decision variables to generate the Pareto optimal process scheme using a multi-objective optimization algorithm. The dynamic scheduling module is configured to convert Pareto optimal process schemes into executable instructions to control the operation of each piece of equipment in the physical equipment unit. The virtual-real interaction closed-loop correction module is configured to compare the measured data and predicted data of the optical detection unit and the spectrometer, generate a deviation signal, update the model parameters of the impurity dynamic inference module through the federated learning mechanism, and adjust the weight coefficients of the multi-objective collaborative optimization decision module to form a closed-loop correction. The data acquisition module inputs its time-series multidimensional feature vectors into the impurity dynamic deduction module; the predicted data from the impurity dynamic deduction module is input into the multi-objective collaborative optimization decision-making module; the Pareto optimal process scheme from the multi-objective collaborative optimization decision-making module is input into the dynamic scheduling module; the control commands from the dynamic scheduling module drive the physical equipment; the measured data from the optical detection unit and the spectrometer are fed back to the virtual-real interaction closed-loop correction module; and the deviation signal from the virtual-real interaction closed-loop correction module is used to reverse adjust the impurity dynamic deduction module and the multi-objective collaborative optimization decision-making module, thus forming a collaborative network for bidirectional data flow.

[0006] Furthermore, the lime slurry impurity removal system for calcium silicate production of the present invention includes an optical sensing unit, a chemical sensing unit, and a data preprocessing unit in its data acquisition module. The optical sensing unit is configured to acquire particle size distribution and spectral characteristic data via a laser particle size analyzer and a hyperspectral imaging probe. The chemical sensing unit is configured to acquire turbidity and pH data via an online turbidity sensor and a pH electrode. The data preprocessing unit is configured to perform wavelet transform denoising and sliding window timestamp alignment on the data collected by the optical sensing unit and the chemical sensing unit to generate time-series multidimensional feature vectors. The data from the optical sensing unit and the chemical sensing unit are input to the data preprocessing unit in parallel, and the output of the data preprocessing unit is used as the input to the impurity dynamic inference module.

[0007] Furthermore, the lime slurry impurity removal system for calcium silicate production of the present invention includes an impurity dynamic deduction module comprising a model building layer, a simulation calculation layer, and a learning optimization layer; The model building layer is configured to create a three-dimensional geometric model of the physical device using three-dimensional scanning data; The simulation calculation layer is configured to simulate lime slurry flow and impurity movement by coupling computational fluid dynamics and discrete element method; The learning optimization layer is configured to utilize a long short-term memory network to fuse historical production data and dynamically correct simulation parameters. The three-dimensional geometric model of the model building layer provides mesh boundary conditions for the simulation calculation layer. The simulation results of the simulation calculation layer are input into the learning and optimization layer, and the output of the learning and optimization layer is used as prediction data.

[0008] Furthermore, the lime slurry impurity removal system for calcium silicate production of the present invention includes a multi-objective collaborative optimization decision module comprising a parameter encoding mechanism, a fitness evaluation process, and a Pareto solution screening strategy; The parameter encoding mechanism is configured to normalize decision variables into a hybrid encoded chromosome; The fitness evaluation process is configured with the fitness function being the impurity removal efficiency, energy cost, and reagent consumption. The Pareto solution selection strategy is configured to use non-dominated sorting and crowding calculation to select the optimal solution; The parameter encoding mechanism includes an input-output fitness evaluation process, and the results of the fitness evaluation process are used to generate a Pareto optimal process scheme through a Pareto solution screening strategy.

[0009] Furthermore, the lime slurry impurity removal system for calcium silicate production of the present invention includes a dynamic scheduling module comprising an instruction conversion unit and a real-time scheduling unit; The instruction conversion unit is configured to convert Pareto optimal process schemes into PLC executable instructions; The real-time scheduling unit is configured to monitor the operating status of the equipment through reinforcement learning algorithms and adaptively adjust the equipment parameters. The output of the instruction conversion unit drives the physical equipment, and the real-time scheduling unit collects equipment status data and feeds it back to the instruction conversion unit, forming a local closed loop.

[0010] Furthermore, the lime slurry impurity removal system for calcium silicate production of the present invention includes a virtual-real interactive closed-loop correction module comprising a deviation calculation mechanism and a federated learning process; The deviation calculation mechanism is configured to calculate the Mahalanobis distance between the measured data and the predicted data through principal component analysis. The federated learning process is configured to encrypt bias data and aggregate and update model parameters; Among them, the deviation signal of the deviation calculation mechanism triggers the federated learning process, and the output of the federated learning process simultaneously updates the impurity dynamic inference module and the multi-objective collaborative optimization decision module.

[0011] Furthermore, in the lime slurry impurity removal system for calcium silicate production of the present invention, after the predicted data of the impurity dynamic deduction module is input into the multi-objective collaborative optimization decision module, the multi-objective collaborative optimization decision module calls the digital twin model interface to simulate the parameter effect and feeds back the simulation results to the impurity dynamic deduction module to update the initial boundary conditions, thus forming a virtual trial-and-error closed loop.

[0012] Furthermore, in the lime slurry impurity removal system for calcium silicate production of the present invention, the equipment operation status data of the dynamic scheduling module is synchronized in real time to the impurity dynamic inference module to correct the flow field parameters. At the same time, the deviation signal of the virtual-real interactive closed-loop correction module adjusts the weight coefficient of the multi-objective collaborative optimization decision module, so that the system can adapt to raw material fluctuations.

[0013] 9. The lime slurry impurity removal system for calcium silicate production according to claim 1, characterized in that the sampling frequency of the data acquisition module is dynamically adjusted by the virtual-real interactive closed-loop correction module according to the magnitude of the deviation; when the sensor data is abnormal, a backup detection device is switched; and the model parameters of the impurity dynamic inference module and the multi-objective collaborative optimization decision-making module are shared across production lines through a federated learning mechanism to improve the robustness of the system.

[0014] Secondly, the present invention provides a method for removing impurities from lime slurry used in calcium silicate production, applied to the aforementioned lime slurry removal system for calcium silicate production, comprising: Step 1: Collect particle size distribution, spectral characteristics, turbidity and pH value data of lime slurry from the physical equipment device, and perform noise reduction and timestamp alignment on the collected data to generate a time-series multidimensional feature vector. The physical equipment device includes a lime slurry reactor, a vibrating screen, a primary sedimentation tank, a jumping screen, a two-stage sedimentation tank, a precision filter, a high-speed centrifuge and a chemical flocculation reactor connected in sequence, as well as an optical detection unit and a spectrometer set at the outlet of the two-stage sedimentation tank. Step 2: Receive time-series multidimensional feature vectors, simulate the movement trajectory of impurities based on the three-dimensional geometric model of the physical device using computational fluid dynamics and discrete element method, and dynamically correct simulation parameters using long short-term memory network to output predicted data on impurity types, concentrations and evolution trends. Step 3: Receive the predicted data, and use the vibration frequency of the vibrating screen, the residence time in the sedimentation tank, the combination of screen apertures, and the amount of flocculant added as decision variables to generate the Pareto optimal process scheme using a multi-objective optimization algorithm. Step 4: Convert the Pareto optimal process scheme into executable instructions to control the operation of each device in the physical equipment unit; Step 5: Compare the measured data and predicted data of the optical detection unit and the spectrometer to generate a deviation signal, and update the simulation parameters in Step 2 through the federated learning mechanism. At the same time, adjust the weight coefficients in Step 3 to form a closed-loop correction. In this process, the time-series multidimensional feature vector generated in step 1 is input into step 2, the predicted data output in step 2 is input into step 3, the Pareto optimal process scheme generated in step 3 is input into step 4, the control command in step 4 drives the physical equipment, and the measured data of the optical detection unit and the spectrometer are fed back to step 5. The deviation signal generated in step 5 adjusts the parameters in steps 2 and 3 in reverse, forming a collaborative network for bidirectional data flow, which is used to remove impurities in lime slurry to improve the whiteness and purity of calcium silicate products.

[0015] Beneficial effects of this invention; The beneficial effects of this invention are reflected in the following aspects: First, the data acquisition module collects real-time data on the particle size distribution, spectral characteristics, turbidity, and pH value of lime slurry. This data is then processed using wavelet transform denoising and sliding window timestamp alignment to generate a time-series multidimensional feature vector, enabling continuous and accurate monitoring of impurity states. Second, the impurity dynamic prediction module, based on a three-dimensional geometric model of the physical equipment, simulates the impurity trajectory using computational fluid dynamics and discrete element methods. It also utilizes a long short-term memory network to dynamically correct simulation parameters, outputting predicted data on impurity types, concentrations, and evolution trends, thus providing early warnings of impurity behavior. Third, the multi-objective collaborative optimization decision-making module uses the vibration frequency of the vibrating screen, the residence time in the sedimentation tank, the combination of screen apertures, and the flocculant dosage as decision variables, employing a multi-objective optimization algorithm. The system generates a Pareto-optimal process scheme, balancing impurity removal efficiency, energy consumption, and reagent consumption. A dynamic scheduling module converts this Pareto-optimal scheme into executable instructions, controlling the operation of each device in the physical equipment unit for precise control. A virtual-real interactive closed-loop correction module compares measured and predicted data from the optical detection unit and the spectrometer, generating a deviation signal. It then updates the model parameters of the impurity dynamic deduction module through a federated learning mechanism, while simultaneously adjusting the weight coefficients of the multi-objective collaborative optimization decision module to form a closed-loop correction. A collaborative network with bidirectional data flow enables the system to adapt to raw material fluctuations and equipment wear in real time. Through virtual trial and error and continuous optimization, it effectively removes impurities and eliminates black spot defects, thereby steadily improving the whiteness and purity of calcium silicate products. Attached Figure Description

[0016] To more clearly illustrate the technical solution of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on the drawings without creative effort.

[0017] Figure 1 A flowchart of a lime milk purification method for calcium silicate production provided in an embodiment of the present invention. Detailed Implementation

[0018] To make the technical solution of the present invention clearer, the present invention will be clearly and completely described below with reference to specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. 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. The present invention provided by various embodiments will be described in detail below with reference to the accompanying drawings. To better understand the purpose of the present invention, the present invention will be described in further detail below.

[0019] This invention provides a lime slurry impurity removal system for calcium silicate production, comprising: The data acquisition module uses laser particle size analyzers and hyperspectral imaging probes deployed at the outlet of the lime slurry reactor, the inlet and outlet of the vibrating screen, each stage of sedimentation tanks, and the advanced treatment unit to collect real-time particle size distribution and spectral characteristic data during the lime slurry flow process. Online turbidity sensors and pH electrodes simultaneously monitor changes in suspended solids concentration and pH. The collected raw data undergoes denoising processing using wavelet transform algorithms to eliminate high-frequency interference components. Then, a sliding window matching mechanism aligns the optical and chemical data with millisecond-level timestamps, generating a time-series multidimensional feature vector including a particle distribution histogram, spectral absorption curve, turbidity time-series curve, and pH gradient.

[0020] After receiving the time-series multidimensional feature vector, the impurity dynamic simulation module initiates computational fluid dynamics simulation based on the three-dimensional geometric model of the physical equipment. The simulation process uses the k-epsilon turbulence model to solve the Navier-Stokes equations, simulating the flow state of lime slurry within the equipment. The discrete element method (DEM) calculates the trajectories and collision energy losses of impurities such as carbon particles and burnt limestone using the Hertz-Mundling contact model. The learning optimization layer calls upon the mapping relationship between black spot formation and impurity concentration in the historical production database, dynamically corrects the simulation parameters through a long short-term memory network, and outputs predicted data on impurity types, concentrations, and evolution trends.

[0021] The multi-objective collaborative optimization decision-making module encodes process parameters such as vibrating screen vibration frequency, sedimentation tank residence time, screen aperture combination, and flocculant dosage as decision variables. The parameter encoding mechanism normalizes continuous variables into a hybrid encoding chromosome, and the fitness evaluation process uses impurity removal efficiency, energy cost, and reagent consumption as multi-objective optimization functions. By iteratively calling the digital twin model interface, the impurity removal effect under different parameter combinations is simulated, and a Pareto optimal process scheme is generated using a non-dominated sorting algorithm and a crowding calculation strategy. The optimization results are synchronously fed back to the impurity dynamic simulation module to update the initial boundary conditions.

[0022] The dynamic scheduling module converts the Pareto optimal process scheme into PLC executable instructions through the instruction conversion unit. These instructions specifically include the frequency adjustment signal for the vibrating screen motor, the opening control command for the sedimentation tank sludge discharge valve, the centrifuge speed setpoint, and the flow parameters for the flocculant metering pump. The real-time scheduling unit constructs an equipment operation status monitoring network using reinforcement learning algorithms. It adaptively adjusts the filter backwashing cycle based on the degree of filter cartridge clogging, dynamically optimizes the start-up and shutdown strategy based on centrifuge current fluctuations, and synchronizes the equipment status data to the digital twin model in real time.

[0023] The virtual-real interactive closed-loop correction module uses principal component analysis (PCA) to calculate the Mahalanobis distance between the transmittance, number of black spots, and predicted data collected by the optical detection unit, generating a deviation signal. This deviation signal triggers a federated learning mechanism; the encrypted deviation data is uploaded to an aggregation server for weighted averaging, updates the global model parameters, and is then distributed to the local digital twin model. Simultaneously, the deviation signal serves as a constraint input to a multi-objective optimization algorithm, dynamically adjusting the weighting coefficients for noise removal efficiency and energy consumption cost. The data quality assessment unit adaptively adjusts the sampling frequency based on the deviation magnitude and automatically switches to a backup detection device when sensor data is abnormal.

[0024] The modules establish a bidirectional interactive channel via a data bus. Preprocessed data from the data acquisition module provides high-quality input to the deduction module, while the predicted data from the deduction module serves as the evaluation benchmark for the optimization decision-making module. The optimization results drive physical equipment execution through the dynamic scheduling module. Deviation signals from the closed-loop correction module simultaneously affect parameter updates in the deduction module and weight adjustments in the optimization module, forming a closed-loop process from data acquisition to equipment control. This multi-module nonlinear interactive architecture enables the system to adapt to raw material fluctuations and equipment wear, continuously optimizing the impurity removal effect through virtual-real interaction.

[0025] During system operation, real-time data generated by the physical equipment is continuously input into the control device via the data acquisition module. The calculation results from the control device are fed back to the physical equipment via the dynamic scheduling module, forming an execution closed loop. Quality verification data from the optical detection unit and the spectral analyzer are compared with virtual space prediction results. The resulting deviation signals are updated online and shared across production lines via a federated learning mechanism. This data-model bidirectional driving mechanism overcomes the limitations of existing linear control architectures, achieving a stable improvement in the whiteness and purity of calcium silicate products.

[0026] The data acquisition module of this invention includes an optical sensing unit, a chemical sensing unit, and a data preprocessing unit. The optical sensing unit irradiates the lime slurry flow with a laser beam emitted by a laser particle size analyzer, calculates the particle size distribution based on the scattered light pattern, and simultaneously captures the reflection spectrum at different wavelengths using a hyperspectral imaging probe to identify the spectral characteristics of impurities. The chemical sensing unit measures the scattering intensity of light by suspended particles using an online turbidity sensor to derive the turbidity value, and detects changes in hydrogen ion concentration using a pH electrode. The data preprocessing unit receives the raw data from the optical and chemical sensing units, applies a wavelet transform algorithm to decompose the signal frequency domain components, filters out high-frequency noise, and then aligns the data streams from different sensors according to timestamps using a sliding window mechanism to generate a time-seriesd multidimensional feature vector. Data from the optical and chemical sensing units are input to the data preprocessing unit in parallel, and the output of the preprocessing unit serves as the input to the impurity dynamic inference module, ensuring data consistency and accuracy.

[0027] The impurity dynamic prediction module of this invention includes a model building layer, a simulation calculation layer, and a learning and optimization layer. The model building layer acquires point cloud data of the physical equipment using a 3D laser scanner to construct an accurate 3D geometric model, including internal flow paths and obstacle contours. The simulation calculation layer divides the computational grid based on the 3D geometric model, uses computational fluid dynamics to solve the flow equations to simulate the lime slurry velocity field, and combines the discrete element method to calculate the trajectory and collision behavior of impurity particles in the flow field. The learning and optimization layer accesses a historical production database, uses a long short-term memory network to analyze the temporal correlation between impurity concentration and black spot defects, and dynamically adjusts simulation parameters such as viscosity coefficient and collision recovery coefficient. The 3D geometric model of the model building layer provides mesh boundary conditions for the simulation calculation layer, the simulation results of the simulation calculation layer are input into the learning and optimization layer, and the output of the learning and optimization layer is used as prediction data, forming a closed loop from geometric modeling to parameter optimization.

[0028] This invention's multi-objective collaborative optimization decision module includes a parameter encoding mechanism, a fitness evaluation process, and a Pareto solution screening strategy. The parameter encoding mechanism normalizes continuous variables such as the vibration frequency of the vibrating screen and the residence time in the sedimentation tank, and encodes them together with discrete variables such as the combination of screen apertures into a chromosome string. The fitness evaluation process uses impurity removal efficiency, energy cost, and reagent consumption as multi-objective functions. Impurity removal efficiency is calculated based on predicted data to determine the impurity removal rate; energy cost is estimated based on equipment power; and reagent consumption is calculated based on the amount of flocculant used. The Pareto solution screening strategy uses a non-dominated sorting algorithm to compare the quality of solution sets and then selects the evenly distributed Pareto optimal solution through crowding degree calculation. The output of the parameter encoding mechanism is input to the fitness evaluation process, and the result of the fitness evaluation process generates a Pareto optimal process scheme through the Pareto solution screening strategy, achieving a multi-objective trade-off.

[0029] The dynamic scheduling module of this invention includes an instruction conversion unit and a real-time scheduling unit. The instruction conversion unit analyzes the Pareto optimal process scheme, converting the vibrating screen frequency into a pulse width modulation signal, the sedimentation tank residence time into a timer setting, the screen aperture combination into pneumatic valve control commands, and the flocculant dosage into a metering pump flow setpoint. The real-time scheduling unit constructs a state-action value function using a reinforcement learning algorithm, monitors equipment operating status such as vibrating screen amplitude and centrifuge current, and adaptively adjusts parameters to cope with load changes. The output of the instruction conversion unit drives the physical equipment devices, while the real-time scheduling unit collects equipment status data and feeds it back to the instruction conversion unit, forming a local closed-loop control and improving response speed.

[0030] This invention's virtual-real interactive closed-loop correction module includes a deviation calculation mechanism and a federated learning process. The deviation calculation mechanism projects the transmittance curve of the optical detection unit and the black dot distribution map of the spectral analyzer onto the feature space using principal component analysis, calculates the Mahalanobis distance between the measured data and the predicted data, and generates a deviation signal. The federated learning process encrypts the deviation data using an encryption algorithm, uploads it to an aggregation server for weighted averaging of model parameters, updates the global model, and then distributes it back locally. The deviation signal from the deviation calculation mechanism triggers the federated learning process. The output of the federated learning process simultaneously updates the model parameters of the impurity dynamic inference module and the weight coefficients of the multi-objective collaborative optimization decision module, achieving collaborative correction.

[0031] After the predicted data from the impurity dynamic simulation module of this invention is input into the multi-objective collaborative optimization decision-making module, the multi-objective collaborative optimization decision-making module calls the digital twin model interface to simulate the impurity removal effect under different process parameters, and feeds back the simulation results to the impurity dynamic simulation module to update the initial boundary conditions. This virtual trial-and-error closed loop allows for multiple iterative optimizations in the digital space, reducing the cost of physical trial and error and improving decision-making accuracy.

[0032] The dynamic scheduling module of this invention synchronizes the equipment operating status data to the impurity dynamic deduction module in real time, correcting flow field parameters such as flow velocity and pressure distribution. Simultaneously, the deviation signal from the virtual-real interactive closed-loop correction module adjusts the weighting coefficients of the multi-objective collaborative optimization decision-making module, prioritizing impurity removal efficiency or energy consumption cost. This bidirectional adjustment enables the system to adapt to fluctuations in the composition of lime slurry raw materials and maintain stable operation.

[0033] The sampling frequency of the data acquisition module in this invention is dynamically adjusted by the virtual-real interactive closed-loop correction module according to the magnitude of the deviation. The sampling rate is increased when the deviation is large and decreased when the deviation is small to save resources. When sensor data is abnormal, such as signal loss or drift, the system automatically switches to a backup detection device. The model parameters of the impurity dynamic inference module and the multi-objective collaborative optimization decision-making module are shared across production lines through a federated learning mechanism, aggregating data from multiple lines to improve the model's generalization ability, thereby enhancing the system's robustness.

[0034] Secondly, please refer to Figure 1 This invention provides a method for removing impurities from lime slurry used in calcium silicate production, applied to the lime slurry removal system used in calcium silicate production, comprising: Step 1: Collect particle size distribution, spectral characteristics, turbidity and pH value data of lime slurry from the physical equipment device, and perform noise reduction and timestamp alignment on the collected data to generate a time-series multidimensional feature vector. The physical equipment device includes a lime slurry reactor, a vibrating screen, a primary sedimentation tank, a jumping screen, a two-stage sedimentation tank, a precision filter, a high-speed centrifuge and a chemical flocculation reactor connected in sequence, as well as an optical detection unit and a spectrometer set at the outlet of the two-stage sedimentation tank. Step 2: Receive time-series multidimensional feature vectors, simulate the movement trajectory of impurities based on the three-dimensional geometric model of the physical device using computational fluid dynamics and discrete element method, and dynamically correct simulation parameters using long short-term memory network to output predicted data on impurity types, concentrations and evolution trends. Step 3: Receive the predicted data, and use the vibration frequency of the vibrating screen, the residence time in the sedimentation tank, the combination of screen apertures, and the amount of flocculant added as decision variables to generate the Pareto optimal process scheme using a multi-objective optimization algorithm. Step 4: Convert the Pareto optimal process scheme into executable instructions to control the operation of each device in the physical equipment unit; Step 5: Compare the measured data and predicted data of the optical detection unit and the spectrometer to generate a deviation signal, and update the simulation parameters in Step 2 through the federated learning mechanism. At the same time, adjust the weight coefficients in Step 3 to form a closed-loop correction. In this process, the time-series multidimensional feature vector generated in step 1 is input into step 2, the predicted data output in step 2 is input into step 3, the Pareto optimal process scheme generated in step 3 is input into step 4, the control command in step 4 drives the physical equipment, and the measured data of the optical detection unit and the spectrometer are fed back to step 5. The deviation signal generated in step 5 adjusts the parameters in steps 2 and 3 in reverse, forming a collaborative network for bidirectional data flow, which is used to remove impurities in lime slurry to improve the whiteness and purity of calcium silicate products.

[0035] This invention addresses the challenge of monitoring and removing solid impurities and suspended solids from lime slurry by constructing a collaborative system integrating physical equipment and control devices. The system first acquires real-time data on the particle size distribution, spectral characteristics, turbidity, and pH of the lime slurry through a data acquisition module. After wavelet transform denoising and sliding window timestamp alignment, a time-series multidimensional feature vector is generated, enabling continuous and accurate monitoring of impurity states. The impurity dynamic prediction module, based on a three-dimensional geometric model of the physical equipment, simulates impurity trajectories using computational fluid dynamics and discrete element methods. It then dynamically corrects simulation parameters using a long short-term memory network, outputting predicted data on impurity types, concentrations, and evolution trends, thus providing early warnings of impurity behavior. The multi-objective collaborative optimization decision-making module uses the vibration frequency of the vibrating screen, the residence time in the sedimentation tank, the combination of screen apertures, and the flocculant dosage as decision variables. Through a multi-objective optimization algorithm, it generates a Pareto optimal process scheme, balancing impurity removal efficiency, energy costs, and reagent consumption. The dynamic scheduling module converts the optimized scheme into executable instructions, controlling the operation of each device within the physical equipment to achieve precise operation. The virtual-real interactive closed-loop correction module compares the measured data and predicted data from the optical detection unit and the spectrometer, generates a deviation signal, updates the model parameters of the impurity dynamic inference module through a federated learning mechanism, and adjusts the weight coefficients of the multi-objective collaborative optimization decision module to form a closed-loop correction. This collaborative network with bidirectional data flow enables the system to adapt to raw material fluctuations and equipment wear in real time. Through virtual trial and error and continuous optimization, it effectively removes impurities and eliminates black spot defects, thereby steadily improving the whiteness and purity of calcium silicate products.

Claims

1. A lime slurry impurity removal system for calcium silicate production, characterized in that, It includes physical equipment and control devices, with the control devices establishing a communication connection with the physical equipment. The physical equipment includes a lime ash reactor, a vibrating screen, a primary sedimentation tank, a jumping screen, a two-stage sedimentation tank, a precision filter, a high-speed centrifuge, and a chemical flocculation reactor connected in sequence, as well as an optical detection unit and a spectrometer located at the outlet of the two-stage sedimentation tank; The control device includes a data acquisition module, an impurity dynamic simulation module, a multi-objective collaborative optimization decision-making module, a dynamic scheduling module, and a virtual-real interactive closed-loop correction module. The data acquisition module is configured to collect data on the particle size distribution, spectral characteristics, turbidity, and pH value of lime slurry from the physical equipment, and to perform noise reduction and timestamp alignment on the collected data to generate a time-series multidimensional feature vector. The impurity dynamic inference module is configured to receive time-series multidimensional feature vectors, simulate the movement trajectory of impurities based on the three-dimensional geometric model of the physical device through computational fluid dynamics and discrete element method, and dynamically correct simulation parameters using long short-term memory network to output predicted data on impurity type, concentration and evolution trend. The multi-objective collaborative optimization decision module is configured to receive prediction data and use the vibration frequency of the vibrating screen, the residence time of the sedimentation tank, the combination of screen apertures, and the amount of flocculant added as decision variables to generate the Pareto optimal process scheme using a multi-objective optimization algorithm. The dynamic scheduling module is configured to convert Pareto optimal process schemes into executable instructions to control the operation of each piece of equipment in the physical equipment unit. The virtual-real interaction closed-loop correction module is configured to compare the measured data and predicted data of the optical detection unit and the spectrometer, generate a deviation signal, update the model parameters of the impurity dynamic inference module through a federated learning mechanism, and adjust the weight coefficients of the multi-objective collaborative optimization decision module to form a closed-loop correction.

2. The lime slurry impurity removal system for calcium silicate production according to claim 1, characterized in that, Also includes: The time-series multidimensional feature vector of the data acquisition module is input into the impurity dynamic deduction module. The predicted data of the impurity dynamic deduction module is input into the multi-objective collaborative optimization decision module. The Pareto optimal process scheme of the multi-objective collaborative optimization decision module is input into the dynamic scheduling module. The control command of the dynamic scheduling module drives the physical equipment. The measured data of the optical detection unit and the spectrometer are fed back to the virtual-real interaction closed-loop correction module. The deviation signal of the virtual-real interaction closed-loop correction module is used to adjust the impurity dynamic deduction module and the multi-objective collaborative optimization decision module in reverse, forming a collaborative network for bidirectional data flow. The data acquisition module includes an optical sensing unit, a chemical sensing unit, and a data preprocessing unit; The optical sensing unit is configured to acquire particle size distribution and spectral characteristic data via a laser particle size analyzer and a hyperspectral imaging probe. The chemical sensing unit is configured to acquire turbidity and pH data via an online turbidity sensor and a pH electrode. The data preprocessing unit is configured to perform wavelet transform denoising and sliding window timestamp alignment on the data collected by the optical sensing unit and the chemical sensing unit to generate time-series multidimensional feature vectors. The data from the optical sensing unit and the chemical sensing unit are input to the data preprocessing unit in parallel, and the output of the data preprocessing unit is used as the input to the impurity dynamic inference module.

3. The lime slurry impurity removal system for calcium silicate production according to claim 1, characterized in that, The impurity dynamic inference module includes a model building layer, a simulation calculation layer, and a learning optimization layer; The model building layer is configured to create a three-dimensional geometric model of the physical device using three-dimensional scanning data; The simulation calculation layer is configured to simulate lime slurry flow and impurity movement by coupling computational fluid dynamics and discrete element method; The learning optimization layer is configured to utilize a long short-term memory network to fuse historical production data and dynamically correct simulation parameters. The three-dimensional geometric model of the model building layer provides mesh boundary conditions for the simulation calculation layer. The simulation results of the simulation calculation layer are input into the learning and optimization layer, and the output of the learning and optimization layer is used as prediction data.

4. The lime slurry impurity removal system for calcium silicate production according to claim 1, characterized in that, The multi-objective collaborative optimization decision-making module includes a parameter encoding mechanism, a fitness evaluation process, and a Pareto solution selection strategy; The parameter encoding mechanism is configured to normalize decision variables into a hybrid encoded chromosome; The fitness evaluation process is configured with the fitness function being the impurity removal efficiency, energy cost, and reagent consumption. The Pareto solution selection strategy is configured to use non-dominated sorting and crowding calculation to select the optimal solution; The parameter encoding mechanism includes an input-output fitness evaluation process, and the results of the fitness evaluation process are used to generate a Pareto optimal process scheme through a Pareto solution screening strategy.

5. The lime slurry impurity removal system for calcium silicate production according to claim 1, characterized in that, The dynamic scheduling module includes an instruction conversion unit and a real-time scheduling unit; The instruction conversion unit is configured to convert Pareto optimal process schemes into PLC executable instructions; The real-time scheduling unit is configured to monitor the operating status of the equipment through reinforcement learning algorithms and adaptively adjust the equipment parameters. The output of the instruction conversion unit drives the physical equipment, and the real-time scheduling unit collects equipment status data and feeds it back to the instruction conversion unit, forming a local closed loop.

6. The lime slurry impurity removal system for calcium silicate production according to claim 1, characterized in that, The virtual-real interaction closed-loop correction module includes a deviation calculation mechanism and a federated learning process; The deviation calculation mechanism is configured to calculate the Mahalanobis distance between the measured data and the predicted data through principal component analysis. The federated learning process is configured to encrypt bias data and aggregate and update model parameters; Among them, the deviation signal of the deviation calculation mechanism triggers the federated learning process, and the output of the federated learning process simultaneously updates the impurity dynamic inference module and the multi-objective collaborative optimization decision module.

7. The lime slurry impurity removal system for calcium silicate production according to claim 1, characterized in that, After the predicted data from the impurity dynamic simulation module is input into the multi-objective collaborative optimization decision module, the multi-objective collaborative optimization decision module calls the digital twin model interface to simulate the parameter effects and feeds back the simulation results to the impurity dynamic simulation module to update the initial boundary conditions, thus forming a virtual trial-and-error closed loop.

8. The lime slurry impurity removal system for calcium silicate production according to claim 1, characterized in that, The equipment operation status data of the dynamic scheduling module is synchronized to the impurity dynamic inference module in real time to correct the flow field parameters. At the same time, the deviation signal of the virtual-real interaction closed-loop correction module adjusts the weight coefficient of the multi-objective collaborative optimization decision module so that the system can adapt to raw material fluctuations.

9. The lime slurry impurity removal system for calcium silicate production according to claim 1, characterized in that, The sampling frequency of the data acquisition module is dynamically adjusted by the virtual-real interactive closed-loop correction module according to the magnitude of the deviation. When the sensor data is abnormal, the backup detection equipment is switched. The model parameters of the impurity dynamic inference module and the multi-objective collaborative optimization decision module are shared across production lines through a federated learning mechanism to improve the robustness of the system.

10. A method for removing impurities from lime slurry used in calcium silicate production, applied to the lime slurry removal system for calcium silicate production as described in any one of claims 1 to 9, characterized in that, include: Step 1: Collect particle size distribution, spectral characteristics, turbidity and pH value data of lime slurry from the physical equipment device, and perform noise reduction and timestamp alignment on the collected data to generate a time-series multidimensional feature vector. The physical equipment device includes a lime slurry reactor, a vibrating screen, a primary sedimentation tank, a jumping screen, a two-stage sedimentation tank, a precision filter, a high-speed centrifuge and a chemical flocculation reactor connected in sequence, as well as an optical detection unit and a spectrometer set at the outlet of the two-stage sedimentation tank. Step 2: Receive time-series multidimensional feature vectors, simulate the movement trajectory of impurities based on the three-dimensional geometric model of the physical device using computational fluid dynamics and discrete element method, and dynamically correct simulation parameters using long short-term memory network to output predicted data on impurity types, concentrations and evolution trends. Step 3: Receive the predicted data, and use the vibration frequency of the vibrating screen, the residence time in the sedimentation tank, the combination of screen apertures, and the amount of flocculant added as decision variables to generate the Pareto optimal process scheme using a multi-objective optimization algorithm. Step 4: Convert the Pareto optimal process scheme into executable instructions to control the operation of each device in the physical equipment unit; Step 5: Compare the measured data and predicted data of the optical detection unit and the spectrometer to generate a deviation signal, and update the simulation parameters in Step 2 through the federated learning mechanism. At the same time, adjust the weight coefficients in Step 3 to form a closed-loop correction. In this process, the time-series multidimensional feature vector generated in step 1 is input into step 2, the predicted data output in step 2 is input into step 3, the Pareto optimal process scheme generated in step 3 is input into step 4, the control command in step 4 drives the physical equipment, and the measured data of the optical detection unit and the spectrometer are fed back to step 5. The deviation signal generated in step 5 adjusts the parameters in steps 2 and 3 in reverse, forming a collaborative network for bidirectional data flow, which is used to remove impurities in lime slurry to improve the whiteness and purity of calcium silicate products.

Images