Intelligent copper blowing system based on multi-mode perception

Abstract

The invention provides an intelligent copper blowing system based on multi-modal sensing. The intelligent copper blowing system comprises a multi-modal sensing system, an intelligent decision and control module, a real-time simulation system, an air supply system and a flexible sealing smoke hood. The multi-mode sensing system collects melt components in the furnace, temperature field distribution, equipment operation states and environmental pollutant concentration data in real time. And the intelligent decision and control module predicts a flow field and pollutant diffusion path through a real-time simulation system based on the sensing data, and generates a dynamic control instruction. The air supply system adjusts the blast pressure and the oxygen concentration according to instructions, the flexible sealing smoke hood dynamically adjusts the expansion amount of the sealing apron board, and smoke leakage is restrained. The smelting efficiency can be improved, pollutant emission is reduced, and the service life of equipment is prolonged.

Description

Technical Field

[0001] The present invention relates to the technical field of converter blowing, and more specifically, the present invention relates to an intelligent copper blowing system based on multi-modal perception. Background Art

[0002] Converter blowing is a key process in the metallurgical industry, directly affecting the efficiency and quality of metal smelting. The traditional converter blowing process mainly relies on manual experience and fixed parameter control, and it is difficult to respond in real time to the complex physical and chemical changes in the furnace. In the prior art, the monitoring means for the composition of the melt in the furnace, the temperature field distribution, the equipment status, and the concentration of environmental pollutants are relatively single, lacking the ability of multi-modal data fusion and intelligent analysis. In addition, the existing control systems often cannot dynamically adjust the blowing parameters, resulting in low smelting efficiency, high energy consumption, and difficult control of pollutant emissions.

[0003] In the process of implementing the embodiments of the present invention, the inventors found that there are at least the following problems or defects in the prior art: the accuracy of traditional monitoring means is insufficient, and it is impossible to obtain multi-dimensional data in the furnace in real time; the control system lacks intelligent decision-making ability and it is difficult to achieve precise dynamic adjustment; the control effect of pollutant emissions is not good, and it is difficult to meet the increasingly strict environmental protection requirements. Summary of the Invention

[0004] The present invention provides an intelligent copper blowing system based on multi-modal perception, including:

[0005] A multi-modal perception system, an intelligent decision-making and control module, a real-time simulation system, a blowing system, and a flexible sealed fume hood;

[0006] The multi-modal perception system collects data on the composition of the melt in the furnace, the temperature field distribution, the equipment operation status, and the concentration of environmental pollutants in real time;

[0007] The intelligent decision-making and control module, based on the output data of the multi-modal perception system, predicts the flow field and the pollutant diffusion path through the real-time simulation system, and generates dynamic control instructions;

[0008] The blowing system adjusts the blowing pressure and oxygen concentration according to the control instructions;

[0009] The flexible sealed fume hood dynamically adjusts the telescopic amount of the sealing skirt according to the control instructions to inhibit flue gas leakage.

[0010] Further, the multi-modal perception system includes:

[0011] An in-furnace composition monitoring sub-module, a temperature field monitoring sub-module, an equipment status monitoring sub-module, and an environmental monitoring sub-module;

[0012] The in-furnace composition monitoring sub-module receives the melt radiation spectrum based on a high-temperature resistant fiber optic spectrometer, and analyzes the concentrations of Cu, Fe, and S elements through a multi-scale convolutional neural network; the temperature field monitoring sub-module scans the surface of the melt at the furnace mouth through an infrared thermal imager, and reconstructs the temperature field based on the dynamically calibrated Planck radiation law; the equipment status monitoring sub-module collects the tuyere vibration signal through a piezoelectric sensor, and monitors the erosion thickness of the refractory material in combination with an ultrasonic thickness measurement array; the environmental monitoring sub-module tracks the SO2 escape path through a laser gas analyzer, and calculates the SO2 concentration field based on a three-dimensional inversion algorithm.

[0013] Further, the in-furnace composition monitoring sub-module performs the following steps:

[0014] Receive the radiation spectrum of the melt in the 200 - 1100 nm band through a high-temperature resistant fiber optic spectrometer, and synchronously collect the nitrogen gas curtain flow data to compensate for the soot interference;

[0015] Perform baseline correction and noise filtering on the original spectral data, and extract the characteristic peak intensity sequence;

[0016] Based on the multi-scale convolutional neural network, fuse the local spectral features and the global wavelength correlation, and calculate the concentrations of Cu, Fe, and S elements. The formula is as follows:

[0017]

[0018] In the formula, C Cu is the concentration of Cu element in the melt; is the intensity of the i-th characteristic wavelength (510 nm, 650 nm, 890 nm); w i is the weight coefficient output by the convolutional neural network; k1, k2, k3 are dynamically calibrated parameters; ΔQ N2 is the nitrogen gas curtain flow deviation;

[0019] Cross-validate the element concentration data with the flow field prediction results of the real-time simulation system. If the deviation exceeds 5%, trigger the self-calibration mode of the spectrometer;

[0020] Synchronize the calibrated concentration data to the intelligent decision-making and control module through the edge computing node.

[0021] Further, the temperature field monitoring sub-module performs the following steps:

[0022] Scan the surface of the melt at the furnace mouth through an infrared thermal imager with a resolution of 640×480, and generate a radiation energy distribution map every 10 seconds;

[0023] Based on the dynamic blackbody source to calibrate the emissivity parameter, and calculate the temperature field in combination with the Planck radiation law. The formula is as follows:

[0024]

[0025] Where T(x,y) is the temperature at the coordinate (x,y); h is Planck's constant; c is the speed of light; λ is the working wavelength of the infrared thermal imager; k B is the Boltzmann constant; ∈(λ) is the surface emissivity of the melt after dynamic calibration; L meas (x, y) is the measured radiation intensity;

[0026] The temperature field cloud map is integrated with the composition data in the furnace, and the melt viscosity distribution map is generated based on the viscosity-temperature correlation model. The formula is as follows:

[0027]

[0028] Where v(T) is the melt viscosity; A, B, T0 are material characteristic constants;

[0029] When the local temperature gradient exceeds 200℃ / m, the air pressure adaptive adjustment instruction of the air distribution system is triggered, and the adjustment amount calculation formula is:

[0030]

[0031] In the formula, ΔP is the wind pressure adjustment amount; K is the thermal gradient sensitivity coefficient; is the temperature gradient.

[0032] Furthermore, the device status monitoring submodule performs the following steps:

[0033] The vibration signal of the air outlet is collected by a piezoelectric sensor in the frequency range of 0.5-10kHz, and the blast pressure data is recorded synchronously;

[0034] Perform wavelet packet decomposition on the vibration signal and extract the energy accumulation value A in the 3-5kHz frequency band clog , the formula is as follows:

[0035]

[0036] In the formula, k k (τ) is the wavelet packet coefficient of the ...th sub-band;

[0037] If A clog >0.8V and the blast pressure fluctuation rate exceeds 15%, it is judged as slag accumulation at the tuyere, triggering the high-pressure nitrogen pulse backflush sequence;

[0038] The ultrasonic thickness measurement array emits a 5MHz pulse wave and calculates the thickness of the refractory material based on the echo time difference:

[0039]

[0040] In the formula, δ corr= 0.023·T wall is the temperature compensation term;

[0041] When the erosion thickness is less than 50 mm, generate a maintenance work order and optimize the local tuyere blowing parameters to reduce the thermal stress impact.

[0042] Furthermore, the environmental monitoring sub-module performs the following steps:

[0043] Emit a 1.53 μm wavelength laser through a tunable diode laser absorption spectroscopy (TDLAS) to scan the SO2 diffusion path along the axis of the hood;

[0044] Based on the absorption spectrum intensity distribution, use a three-dimensional inversion algorithm to calculate the SO2 concentration field. The formula is as follows:

[0045]

[0046] In the formula, T ref is the reference temperature; T(x, y, z) is the real-time temperature field data;

[0047] Compare the SO2 concentration field with the CFD-DEM prediction results of the real-time simulation system. If the deviation exceeds 10%, optimize the PID control parameters for the negative pressure of the hood;

[0048] When the local SO2 concentration exceeds 50 ppm, trigger the instruction to expand the sealing skirt and increase the speed of the induced draft fan to 120%.

[0049] Furthermore, the real-time simulation system performs the following steps:

[0050] Extract 100 groups of main modes from the high-fidelity CFD database based on proper orthogonal decomposition (POD) to construct a low-dimensional flow field approximation space;

[0051] Couple the dynamic change of the melt viscosity through an improved Realizable k-ε turbulence model to predict the melt flow rate and the wall heat load. The control equations are as follows:

[0052]

[0053]

[0054] In the formula, G b is the buoyancy term; Y M is the compressible turbulence correction term; v is the kinematic viscosity; S is the modulus of the strain rate tensor;

[0055] Use the CFD-DEM coupling method to simulate the movement of flue gas particles and calculate the SO2 diffusion path and the wall deposition amount;

[0056] Fuse the simulation results with the multi-modal perception data to generate a dynamic control instruction set and feedback it to the intelligent decision-making module.

[0057] Further, the CFD-DEM coupling method includes:

[0058] Calculate the inter-particle contact forces and motion trajectories by the discrete element method (DEM), and the formula is as follows:

[0059]

[0060] In the formula, k n and k t are the normal and tangential stiffness coefficients; γ n and γ t are the damping coefficients; δ n and δ t are the normal and tangential displacements;

[0061] Based on the local porosity, dynamically switch the Gidaspow drag force model to calculate the particle-fluid interaction forces;

[0062] Map the particle motion data to the CFD grid and update the source term in the gas-phase momentum equation:

[0063]

[0064] In the formula, δ(·) is the Dirac function; N p is the number of particles in the grid;

[0065] Solve the two-way coupling equation by implicit iteration and output the high-precision flue gas diffusion simulation results.

[0066] Further, the intelligent decision-making and control module performs the following steps:

[0067] Construct a multi-objective reinforcement learning model with the melting efficiency, environmental protection index, and equipment life as the optimization objectives, and the reward function is designed as follows:

[0068]

[0069] In the formula, α, β, and γ are the weight coefficients; Q melt is the real-time melting amount; C thr is the emission threshold; d crit is the critical erosion thickness;

[0070] Generate a collaborative control strategy for the blast pressure, oxygen concentration, and feeding amount through the dynamic programming algorithm, and the formula is as follows:

[0071]

[0072] In the formula, K p and Kd is the gain coefficient; ΔT is the temperature field uniformity index;

[0073] Transmit the control instruction to the air supply system and the flexible sealed hood, and synchronously update the boundary conditions of the real-time simulation system.

[0074] Furthermore, the flexible sealed hood performs the following steps:

[0075] Drive the silicone rubber sealing skirt to expand and contract through a pneumatic actuator, with a response time of less than 5 seconds, to fill the dynamic gap at the furnace mouth;

[0076] Combined with the pressure difference gradient inside and outside the hood, use an adaptive PID controller to adjust the speed of the induced draft fan, and the control equation is as follows:

[0077]

[0078] In the formula, e(t) is the pressure difference deviation; S is the expansion and contraction amount of the sealing skirt; K f is the feedforward compensation coefficient;

[0079] Real-time monitor the SO2 leakage rate. If it exceeds 3 ppm / m 3 , trigger the secondary expansion of the skirt and increase the nitrogen curtain flow rate to 8 L / min.

[0080] According to the above embodiments of the present invention, it has at least the following beneficial effects: The present invention can improve the intelligent level of the converter blowing process. Through the multi-modal perception system, multi-dimensional data such as the melt composition, temperature field distribution, equipment status, and environmental pollutant concentration in the furnace are collected in real time. Combined with the dynamic optimization algorithm of the intelligent decision-making and control module, precise adjustment of the blowing parameters is achieved. This multi-modal data fusion and intelligent analysis ability can effectively improve the smelting efficiency, reduce energy consumption, while reducing manual intervention and enhancing production stability.

[0081] In addition, the present invention can predict the flow field and pollutant diffusion path through the real-time simulation system. Combined with the dynamic adjustment function of the flexible sealed hood, it can significantly suppress flue gas leakage and reduce pollutant emissions. The adaptive adjustment function of the air supply system can optimize the blast pressure and oxygen concentration, extend the service life of the equipment, and reduce the maintenance cost. Overall, the present invention can meet the urgent needs of modern metallurgical industry for efficient, environmentally friendly, and intelligent production. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] By reading the following detailed description with reference to the accompanying drawings, the above and other objects, features, and advantages of the exemplary embodiments of the present invention will become readily understood. In the drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, wherein:

[0083] Figure 1Schematic diagram of the intelligent copper smelting system based on multimodal perception provided by an embodiment of the present invention;

[0084] Figure 2 Schematic diagram of a converter provided by an embodiment of the present invention. Detailed implementation manners

[0085] The principles and spirit of the present invention will be described below with reference to several exemplary embodiments. It should be understood that these embodiments are provided only to enable those skilled in the art to better understand and then implement the present invention, rather than limiting the scope of the present invention in any way. On the contrary, these embodiments are provided to make the present invention more thorough and complete, and to be able to fully convey the scope of the present invention to those skilled in the art.

[0086] Those skilled in the art know that the embodiments of the present invention can be implemented as a system, device, equipment, method, or computer program product. Therefore, the present invention can be specifically implemented in the following forms: completely hardware, completely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software.

[0087] It should be noted that any number of elements in the drawings is for illustration rather than limitation, and any naming is only for distinction and does not have any limiting meaning.

[0088] Refer to the following Figure 1 , Figure 1 Schematic diagram of the intelligent copper smelting system based on multimodal perception provided by an embodiment of the present invention. As Figure 1 shown, an intelligent copper smelting system based on multimodal perception includes:

[0089] A multimodal perception system 101, an intelligent decision-making and control module 102, a real-time simulation system 103, a blast system 104, and a flexible sealed hood 105;

[0090] The multimodal perception system collects data on the composition of the melt in the furnace, the temperature field distribution, the operating state of the equipment, and the concentration of environmental pollutants in real time;

[0091] Based on the output data of the multimodal perception system, the intelligent decision-making and control module predicts the flow field and the pollutant diffusion path through the real-time simulation system and generates dynamic control instructions;

[0092] The blast system adjusts the blast pressure and oxygen concentration according to the control instructions;

[0093] The flexible sealed hood dynamically adjusts the telescopic amount of the sealing skirt according to the control instructions to inhibit flue gas leakage.

[0094] It should be noted that the present invention relates to an intelligent copper smelting system based on multi-modal perception. The core lies in the real-time collection of multi-dimensional data such as the composition of the melt in the furnace, the temperature field distribution, the operating status of the equipment, and the concentration of environmental pollutants through a multi-modal perception system. The multi-modal perception system refers to a monitoring system that can simultaneously obtain various types of data, including the composition in the furnace, temperature, equipment status, and environmental pollutants. The intelligent decision-making and control module is based on this data, predicts the flow field and pollutant diffusion path through a real-time simulation system, and generates dynamic control instructions to achieve precise control of the smelting process. The air supply system adjusts the blast pressure and oxygen concentration according to the control instructions, while the flexible sealed fume hood dynamically adjusts the telescopic amount of the sealing skirt according to the instructions to effectively suppress flue gas leakage.

[0095] Specifically, the multi-modal perception system includes multiple sub-modules, which are respectively used to monitor the composition in the furnace, the temperature field, the equipment status, and the concentration of environmental pollutants. The in-furnace composition monitoring sub-module receives the melt radiation spectrum through a high-temperature-resistant fiber optic spectrometer and combines the nitrogen curtain flow data to compensate for the soot interference, thereby accurately analyzing the concentrations of elements such as copper, iron, and sulfur. The temperature field monitoring sub-module scans the surface of the melt at the furnace mouth through an infrared thermal imager and reconstructs the temperature field in combination with the dynamically calibrated Planck radiation law. The equipment status monitoring sub-module collects the tuyere vibration signal through a piezoelectric sensor and monitors the erosion thickness of the refractory material in combination with an ultrasonic thickness measurement array. The environmental monitoring sub-module tracks the escape path of sulfur dioxide through a laser gas analyzer and calculates its concentration field based on a three-dimensional inversion algorithm.

[0096] Preferably, when the in-furnace composition monitoring sub-module collects spectral data, it can fuse local spectral features and global wavelength correlations through a multi-scale convolutional neural network to more accurately calculate the element concentration. After the temperature field monitoring sub-module generates the temperature field cloud map, it can fuse it with the in-furnace composition data and generate a melt viscosity distribution map based on the viscosity-temperature correlation model to provide a basis for the air pressure adjustment of the air supply system. When the equipment status monitoring sub-module detects slag accumulation at the tuyere, it can trigger a high-pressure nitrogen pulse and purge sequence to ensure the normal operation of the equipment. When the environmental monitoring sub-module detects that the local sulfur dioxide concentration exceeds the standard, it can trigger an expansion instruction for the sealing skirt and an increase in the speed of the induced draft fan to further reduce pollutant emissions.

[0097] In some embodiments, the multi-modal perception system includes:

[0098] an in-furnace composition monitoring sub-module, a temperature field monitoring sub-module, an equipment status monitoring sub-module, and an environmental monitoring sub-module;

[0099] The in-furnace component monitoring sub-module receives the melt radiation spectrum based on a high-temperature resistant fiber optic spectrometer, and analyzes the concentrations of Cu, Fe, and S elements through a multi-scale convolutional neural network; the temperature field monitoring sub-module scans the surface of the melt at the furnace mouth through an infrared thermal imager, and reconstructs the temperature field based on the dynamically calibrated Planck radiation law; the equipment status monitoring sub-module collects the tuyere vibration signal through a piezoelectric sensor, and monitors the erosion thickness of the refractory material in combination with an ultrasonic thickness measurement array; the environmental monitoring sub-module tracks the SO2 escape path through a laser gas analyzer, and calculates the SO2 concentration field based on a three-dimensional inversion algorithm.

[0100] It should be noted that the multi-modal perception system in the present invention includes multiple sub-modules, which are respectively used to monitor the in-furnace components, temperature field, equipment status, and concentration of environmental pollutants. The in-furnace component monitoring sub-module receives the melt radiation spectrum through a high-temperature resistant fiber optic spectrometer, and compensates for the soot interference by combining the nitrogen gas curtain flow data, so as to accurately analyze the concentrations of elements such as copper, iron, and sulfur. The temperature field monitoring sub-module scans the surface of the melt at the furnace mouth through an infrared thermal imager, and reconstructs the temperature field by combining the dynamically calibrated Planck radiation law. The equipment status monitoring sub-module collects the tuyere vibration signal through a piezoelectric sensor, and monitors the erosion thickness of the refractory material in combination with an ultrasonic thickness measurement array. The environmental monitoring sub-module tracks the escape path of sulfur dioxide through a laser gas analyzer, and calculates its concentration field based on a three-dimensional inversion algorithm.

[0101] Specifically, the in-furnace component monitoring sub-module receives the melt radiation spectrum through a high-temperature resistant fiber optic spectrometer in the wavelength range of 200 to 1100 nanometers, and compensates for the interference of soot on the spectrum by combining the nitrogen gas curtain flow data. The temperature field monitoring sub-module uses an infrared thermal imager to scan the surface of the melt at the furnace mouth with a resolution of 640×480, generates a radiation energy distribution map every 10 seconds, calibrates the emissivity parameter through a dynamic blackbody source, and calculates the temperature field by combining the Planck radiation law. The equipment status monitoring sub-module collects the tuyere vibration signal through a piezoelectric sensor in the frequency band of 0.5 to 10 kHz, and monitors the erosion thickness of the refractory material by combining the ultrasonic thickness measurement array to emit a 5 MHz pulse wave. The environmental monitoring sub-module emits a laser with a wavelength of 1.53 microns through a tunable laser gas analyzer, scans the sulfur dioxide diffusion path on the axis of the hood, and calculates its concentration field based on the absorption spectrum intensity distribution.

[0102] Preferably, when collecting spectral data, the in-furnace composition monitoring sub-module can fuse local spectral features and global wavelength correlation through a multi-scale convolutional neural network, so as to calculate the element concentration more accurately. After generating the temperature field cloud map, the temperature field monitoring sub-module can fuse it with the in-furnace composition data and generate a melt viscosity distribution map based on the viscosity-temperature correlation model, thereby providing a basis for the air pressure adjustment of the air supply system. When detecting slag accumulation at the tuyere, the equipment status monitoring sub-module can trigger a high-pressure nitrogen pulse and purge sequence to ensure the normal operation of the equipment. When detecting that the local sulfur dioxide concentration exceeds the standard, the environmental monitoring sub-module can trigger the expansion instruction of the sealing skirt and the increase of the induced draft fan speed to further reduce pollutant emissions.

[0103] In some embodiments, the in-furnace composition monitoring sub-module performs the following steps:

[0104] Receive the radiation spectrum of the melt in the 200-1100nm band through a high-temperature resistant fiber optic spectrometer, and synchronously collect the nitrogen curtain flow data to compensate for the soot interference;

[0105] Perform baseline correction and noise filtering on the original spectral data, and extract the characteristic peak intensity sequence;

[0106] Based on the multi-scale convolutional neural network, fuse the local spectral features and the global wavelength correlation, and calculate the concentrations of Cu, Fe, and S elements. The formula is as follows:

[0107]

[0108] In the formula, C Cu is the concentration of Cu element in the melt; is the intensity of the i-th characteristic wavelength (510nm, 650nm, 890nm); w i is the weight coefficient output by the convolutional neural network; k1, k2, and k3 are dynamically calibrated parameters; ΔQ N2 is the nitrogen curtain flow deviation;

[0109] Cross-validate the element concentration data with the flow field prediction results of the real-time simulation system. If the deviation exceeds 5%, trigger the self-calibration mode of the spectrometer;

[0110] Synchronize the calibrated concentration data to the intelligent decision-making and control module through the edge computing node.

[0111] It should be noted that the in-furnace composition monitoring sub-module in the present invention receives the radiation spectrum of the melt in the wavelength band of 200 to 1100 nanometers through a high-temperature-resistant fiber optic spectrometer, and combines the nitrogen gas curtain flow rate data to compensate for the interference of soot on the spectrum. The high-temperature-resistant fiber optic spectrometer is a spectral analysis device that can work in a high-temperature environment and can accurately capture the spectral information of the melt radiation. The nitrogen gas curtain flow rate data is used to compensate for the interference of soot on the spectrum to ensure the accuracy of the spectral data. By performing baseline correction and noise filtering on the original spectral data, extracting the characteristic peak intensity sequence, and combining a multi-scale convolutional neural network to fuse local spectral features and global wavelength correlation, the concentrations of elements such as copper, iron, and sulfur are calculated.

[0112] Specifically, when the in-furnace composition monitoring sub-module collects spectral data, it first performs baseline correction and noise filtering on the original spectral data to remove background noise and interference signals and extract the characteristic peak intensity sequence. The multi-scale convolutional neural network is a deep learning model that can simultaneously capture the local features and global correlation of spectral data, thereby more accurately calculating the element concentration. The calculated element concentration data will be cross-validated with the flow field prediction results of the real-time simulation system. If the deviation exceeds 5%, the self-calibration mode of the spectrometer will be triggered to ensure the accuracy of the data. The calibrated concentration data is synchronized to the intelligent decision-making and control module through the edge computing node for generating dynamic control instructions.

[0113] Preferably, when the in-furnace composition monitoring sub-module calculates the element concentration, it can fuse the spectral data of multiple characteristic wavelengths through a multi-scale convolutional neural network, thereby improving the calculation accuracy. The self-calibration mode of the spectrometer can automatically adjust the measurement parameters of the spectrometer when data deviation is detected to ensure the reliability of the data. The edge computing node can process and transmit the calibrated concentration data in real time to ensure that the intelligent decision-making and control module can timely obtain the latest in-furnace composition information, thereby generating more accurate control instructions.

[0114] In some embodiments, the temperature field monitoring sub-module performs the following steps:

[0115] Scan the surface of the melt at the furnace mouth with an infrared thermal imager at a resolution of 640×480, and generate a radiation energy distribution map every 10 seconds;

[0116] Based on the dynamic blackbody source to calibrate the emissivity parameter, and calculate the temperature field in combination with Planck's radiation law. The formula is as follows:

[0117]

[0118] In the formula, T(x,y) is the temperature at the coordinate (x,y); h is Planck's constant; c is the speed of light; λ is the working wavelength of the infrared thermal imager; k Bis the Boltzmann constant; ∈(λ) is the melt surface emissivity after dynamic calibration; L meas (x,y) is the measured radiation intensity;

[0119] Fuse the temperature field cloud map with the in-furnace composition data, and generate a melt viscosity distribution map based on the viscosity-temperature correlation model. The formula is as follows:

[0120]

[0121] In the formula, μ(T) is the melt viscosity; A, B, and T0 are material characteristic constants;

[0122] When the local temperature gradient exceeds 200 °C / m, send an adaptive adjustment command for the air supply system wind pressure, and the calculation formula for the adjustment amount is:

[0123]

[0124] In the formula, ΔP is the wind pressure adjustment amount; K is the thermal gradient sensitivity coefficient; is the temperature gradient.

[0125] It should be noted that in the present invention, the temperature field monitoring sub-module scans the melt surface at the furnace mouth through an infrared thermal imager to generate a radiation energy distribution map, and combines a dynamic blackbody source to calibrate the emissivity parameter, and reconstructs the temperature field based on Planck's radiation law. The infrared thermal imager is a device that can capture the infrared radiation on the surface of an object and generates a radiation energy distribution map by scanning the melt surface. The dynamic blackbody source is used to calibrate the emissivity parameter of the melt surface to ensure the accuracy of temperature field calculation. Planck's radiation law describes the relationship between the radiation energy of an object and its temperature, and the temperature field of the melt surface can be reconstructed through this law. After the generated temperature field cloud map is fused with the in-furnace composition data, a melt viscosity distribution map is generated based on the viscosity-temperature correlation model, thereby providing a basis for the wind pressure adjustment of the air supply system.

[0126] Specifically, the temperature field monitoring sub-module uses an infrared thermal imager to scan the melt surface at the furnace mouth with a resolution of 640×480, and generates a radiation energy distribution map every 10 seconds. The dynamic blackbody source is used to calibrate the emissivity parameter of the melt surface in real time to ensure the accuracy of temperature field calculation. Based on Planck's radiation law, the temperature field of the melt surface is calculated through the radiation energy distribution map. After the generated temperature field cloud map is fused with the in-furnace composition data, a melt viscosity distribution map is generated. When the local temperature gradient exceeds 200 degrees Celsius per meter, an adaptive adjustment command for the air supply system wind pressure is triggered to ensure the uniformity of the melt temperature.

[0127] Preferably, when generating the temperature field cloud map, the temperature field monitoring sub-module can calibrate the emissivity parameter in real time through a dynamic blackbody source to ensure the accuracy of temperature field calculation. After the generated temperature field cloud map is fused with the in-furnace composition data, a melt viscosity distribution map can be generated based on the viscosity-temperature correlation model, thereby providing a basis for the air pressure regulation of the air supply system. When it is detected that the local temperature gradient exceeds the set threshold, an adaptive air pressure regulation command for the air supply system can be triggered to ensure the uniformity of the melt temperature and avoid local overheating or overcooling phenomena.

[0128] In some embodiments, the equipment status monitoring sub-module performs the following steps:

[0129] Collect the tuyere vibration signal through a piezoelectric sensor in the frequency band of 0.5 - 10 kHz and synchronously record the blast pressure data;

[0130] Perform wavelet packet decomposition on the vibration signal and extract the energy accumulation value A in the frequency band of 3 - 5 kHz clog , the formula is as follows:

[0131]

[0132] In the formula, W k (τ) is the wavelet packet coefficient of the k-th sub-band;

[0133] If A clog > 0.8V and the blast pressure volatility exceeds 15%, it is determined that there is slag accumulation at the tuyere, and a high-pressure nitrogen pulse back-blow sequence is triggered;

[0134] Emit a 5 MHz pulse wave through an ultrasonic thickness measurement array and calculate the refractory thickness based on the echo time difference:

[0135]

[0136] In the formula, δ corr = 0.023·T wall is the temperature compensation term;

[0137] When the erosion thickness is less than 50 mm, a maintenance work order is generated and the local blast parameters are optimized to reduce the thermal stress impact.

[0138] It should be noted that the equipment status monitoring sub-module in the present invention collects the vibration signal of the tuyere through a piezoelectric sensor, and combines an ultrasonic thickness measurement array to monitor the erosion thickness of the refractory material. The piezoelectric sensor is a device that can convert mechanical vibration into an electrical signal and is used to collect the vibration signal of the tuyere. The ultrasonic thickness measurement array calculates the erosion thickness of the refractory material by emitting ultrasonic pulses and measuring the echo time difference. By performing wavelet packet decomposition on the vibration signal, the energy accumulation value in a specific frequency band is extracted to determine whether there is slag accumulation at the tuyere. If slag accumulation at the tuyere is detected, the system will trigger a high-pressure nitrogen pulse and a purge sequence to ensure the normal operation of the equipment.

[0139] Specifically, the equipment status monitoring sub-module collects the vibration signal of the tuyere in the frequency band of 0.5 to 10 kHz through a piezoelectric sensor and synchronously records the blast pressure data. Wavelet packet decomposition is performed on the vibration signal, and the energy accumulation value in the frequency band of 3 to 5 kHz is extracted to determine whether there is slag accumulation at the tuyere. If the energy accumulation value exceeds the set threshold and the blast pressure volatility exceeds 15%, the system determines that there is slag accumulation at the tuyere and triggers a high-pressure nitrogen pulse and a purge sequence. The ultrasonic thickness measurement array emits ultrasonic pulses of 5 MHz and calculates the erosion thickness of the refractory material based on the echo time difference. When the erosion thickness is less than 50 mm, the system will generate a maintenance work order and optimize the local blast parameters to reduce the thermal stress impact.

[0140] Preferably, when the equipment status monitoring sub-module detects slag accumulation at the tuyere, it can remove the slag through a high-pressure nitrogen pulse and a purge sequence to ensure the normal operation of the equipment. When the ultrasonic thickness measurement array monitors the erosion thickness of the refractory material, it can improve the measurement accuracy through a temperature compensation term. When it is detected that the erosion thickness of the refractory material is less than the set threshold, the system can generate a maintenance work order and optimize the local blast parameters to reduce the thermal stress impact and extend the service life of the equipment.

[0141] In some embodiments, the environmental monitoring sub-module performs the following steps:

[0142] Emitting a 1.53 μm wavelength laser through a tunable diode laser absorption spectroscopy (TDLAS) to scan the SO2 diffusion path along the axis of the hood;

[0143] Based on the absorption spectrum intensity distribution, a three-dimensional inversion algorithm is used to calculate the SO2 concentration field, and the formula is as follows:

[0144]

[0145] In the formula, T ref is the reference temperature; T(x, y, z) is the real-time temperature field data;

[0146] Compare the SO2 concentration field with the CFD-DEM prediction results of the real-time simulation system. If the deviation exceeds 10%, optimize the PID control parameters of the hood negative pressure.

[0147] When the local SO2 concentration exceeds 50 ppm, trigger the sealing skirt expansion instruction and increase the induced draft fan speed to 120%.

[0148] It should be noted that in the present invention, the environmental monitoring sub-module tracks the escape path of sulfur dioxide through a tunable diode laser absorption spectrometer, and based on the absorption spectral intensity distribution, uses a three-dimensional inversion algorithm to calculate the SO2 concentration field. A tunable diode laser absorption spectrometer is a device that can emit laser light of a specific wavelength. By scanning the sulfur dioxide diffusion path on the axis of the hood, the absorption spectral intensity distribution is obtained. The three-dimensional inversion algorithm is a mathematical method for calculating the gas concentration field based on the absorption spectral intensity distribution, which can accurately reflect the distribution of sulfur dioxide in the hood. The calculated SO2 concentration field is compared with the prediction results of the real-time simulation system. If the deviation exceeds 10%, the system will optimize the hood negative pressure control parameters to ensure effective control of pollutant emissions.

[0149] Specifically, the environmental monitoring sub-module emits a laser with a wavelength of 1.53 microns through a tunable diode laser absorption spectrometer to scan the sulfur dioxide diffusion path on the axis of the hood, and obtains the absorption spectral intensity distribution. Based on the absorption spectral intensity distribution, a three-dimensional inversion algorithm is used to calculate the SO2 concentration field. The calculated SO2 concentration field is compared with the prediction results of the real-time simulation system. If the deviation exceeds 10%, the system will optimize the hood negative pressure control parameters to ensure effective control of pollutant emissions. When the local sulfur dioxide concentration exceeds 50 ppm, the system will trigger the sealing skirt expansion instruction and increase the induced draft fan speed to 120% to further reduce pollutant emissions.

[0150] Preferably, when calculating the SO2 concentration field, the environmental monitoring sub-module can combine the three-dimensional inversion algorithm with the real-time temperature field data to improve the calculation accuracy. When it is detected that the local sulfur dioxide concentration exceeds the set threshold, the system can trigger the sealing skirt expansion instruction and increase the induced draft fan speed to ensure effective control of pollutant emissions. The optimized hood negative pressure control parameters can further improve the sealing performance of the hood and reduce sulfur dioxide leakage.

[0151] In some embodiments, the real-time simulation system performs the following steps:

[0152] Extract 100 sets of main modes from the high-fidelity CFD database based on proper orthogonal decomposition (POD) to construct a low-dimensional flow field approximation space;

[0153] Predict the melt flow rate and wall heat load by coupling the dynamic change of melt viscosity with an improved Realizable k-ε turbulence model. The governing equations are as follows:

[0154]

[0155] In the equations, G b is the buoyancy term; Y M is the compressible turbulence correction term; ν is the kinematic viscosity; S is the modulus of the strain rate tensor;

[0156] Use the CFD-DEM coupling method to simulate the movement of flue gas particles and calculate the SO2 diffusion path and wall deposition amount;

[0157] Fuse the simulation results with multi-modal perception data to generate a dynamic control instruction set and feedback it to the intelligent decision-making module.

[0158] It should be noted that the real-time simulation system in the present invention extracts the main modes from the high-fidelity computational fluid dynamics database based on proper orthogonal decomposition to construct a low-dimensional flow field approximation space. Proper orthogonal decomposition is a mathematical method used to extract the main characteristic modes from high-dimensional data, thereby constructing a low-dimensional approximation space. The high-fidelity computational fluid dynamics database contains a large amount of detailed flow field data. Through proper orthogonal decomposition, 100 groups of main modes can be extracted for constructing a low-dimensional flow field approximation space. The real-time simulation system couples the dynamic change of melt viscosity through an improved turbulence model to predict the melt flow rate and wall heat load, and simulates the movement of flue gas particles by the computational fluid dynamics-discrete element method coupling method to calculate the sulfur dioxide diffusion path and wall deposition amount.

[0159] Specifically, the real-time simulation system extracts 100 groups of main modes from the high-fidelity computational fluid dynamics database based on proper orthogonal decomposition to construct a low-dimensional flow field approximation space. The improved turbulence model can couple the dynamic change of melt viscosity to predict the melt flow rate and wall heat load. The computational fluid dynamics-discrete element method coupling method is used to simulate the movement of flue gas particles and calculate the sulfur dioxide diffusion path and wall deposition amount. After the simulation results are fused with multi-modal perception data, a dynamic control instruction set is generated and fed back to the intelligent decision-making module to optimize the control strategy of the blowing process.

[0160] Preferably, when constructing the low-dimensional flow field approximation space, the real-time simulation system can extract more main modes through proper orthogonal decomposition to improve the accuracy of flow field prediction. The improved turbulence model can better couple the dynamic change of melt viscosity to improve the prediction accuracy of the melt flow rate and wall heat load. The computational fluid dynamics-discrete element method coupling method can more accurately simulate the movement of flue gas particles and optimize the calculation results of the sulfur dioxide diffusion path and wall deposition amount. The generated dynamic control instruction set can be fed back to the intelligent decision-making module in real time to ensure the precise control of the blowing process.

[0161] In some embodiments, the CFD-DEM coupling method includes:

[0162] Calculating the inter-particle contact forces and motion trajectories by the discrete element method (DEM), with the formula as follows:

[0163]

[0164] where k n and k t are the normal and tangential stiffness coefficients; γ n and γ t are the damping coefficients; δ n and δ t are the normal and tangential displacements;

[0165] Based on the local porosity, dynamically switch the Gidaspow drag force model to calculate the particle-fluid interaction forces;

[0166] Map the particle motion data to the CFD grid and update the source term in the gas-phase momentum equation:

[0167]

[0168] where δ(·) is the Dirac function; N p is the number of particles in the grid;

[0169] Solve the two-way coupling equation through implicit iteration and output high-precision flue gas diffusion simulation results.

[0170] It should be noted that the computational fluid dynamics-discrete element method coupling method in the present invention calculates the inter-particle contact forces and motion trajectories by the discrete element method, and combines the local porosity to dynamically switch the particle-fluid interaction model. The discrete element method is a numerical method for simulating the interaction and motion trajectories between particles, which can accurately calculate the contact forces and motion trajectories between particles. The local porosity refers to the distribution density of particles in the fluid, and dynamically switching the particle-fluid interaction model can adjust the interaction forces between particles and the fluid according to the local porosity. By mapping the particle motion data to the computational fluid dynamics grid and updating the source term in the gas-phase momentum equation, high-precision flue gas diffusion simulation is achieved.

[0171] Specifically, the computational fluid dynamics-discrete element method coupling method calculates the contact forces and motion trajectories between particles by the discrete element method, and combines the local porosity to dynamically switch the particle-fluid interaction model. The particle motion data is mapped to the computational fluid dynamics grid, and the source term in the gas-phase momentum equation is updated, so as to more accurately simulate the flue gas diffusion process. By solving the two-way coupling equation through implicit iteration, high-precision flue gas diffusion simulation results are output, providing reliable data support for the intelligent decision-making and control module.

[0172] Preferably, when calculating the contact force between particles by the computational fluid dynamics - discrete element method coupling method, the simulation accuracy of the particle motion trajectory can be improved by adjusting the normal and tangential stiffness coefficients and damping coefficients. The dynamic switching of the local porosity can adjust the particle - fluid interaction model according to the particle distribution density to ensure the accuracy of the simulation results. By implicitly iteratively solving the bidirectional coupling equation, the accuracy of the flue gas diffusion simulation can be further improved, providing more reliable data support for the intelligent decision - making and control module.

[0173] In some embodiments, the intelligent decision - making and control module performs the following steps:

[0174] Construct a multi - objective reinforcement learning model with the melting efficiency, environmental protection indicators, and equipment life as the optimization objectives. The reward function is designed as follows:

[0175]

[0176] In the formula, α, β, and γ are weight coefficients; Q melt is the real - time melting amount; C thr is the emission threshold; d crit is the critical erosion thickness;

[0177] Generate a coordinated control strategy for the blast pressure, oxygen concentration, and feeding amount through the dynamic programming algorithm. The formula is as follows:

[0178]

[0179] In the formula, K p , K d are gain coefficients; ΔT is the temperature field uniformity index;

[0180] Transmit the control instructions to the air supply system and the flexible sealed fume hood, and synchronously update the boundary conditions of the real - time simulation system.

[0181] It should be noted that the intelligent decision - making and control module in the present invention constructs a multi - objective reinforcement learning model with the melting efficiency, environmental protection indicators, and equipment life as the optimization objectives to generate a coordinated control strategy. The multi - objective reinforcement learning model is a machine learning model that can optimize multiple objectives simultaneously. By designing the reward function, the relationship between the melting efficiency, environmental protection indicators, and equipment life is balanced. The design of the reward function takes into account factors such as the real - time melting amount, sulfur dioxide emission concentration, and refractory erosion thickness to ensure that the system reduces pollutant emissions and extends the equipment service life while melting efficiently. A coordinated control strategy for the blast pressure, oxygen concentration, and feeding amount is generated through the dynamic programming algorithm, and the control instructions are transmitted to the air supply system and the flexible sealed fume hood to synchronously update the boundary conditions of the real - time simulation system.

[0182] Specifically, the intelligent decision-making and control module optimizes the smelting efficiency, environmental protection indicators, and equipment life through a multi-objective reinforcement learning model. The design of the reward function takes into account factors such as the real-time smelting volume, sulfur dioxide emission concentration, and refractory erosion thickness. The dynamic programming algorithm is used to generate a coordinated control strategy for the blast pressure, oxygen concentration, and feeding amount, ensuring that while the system smelts efficiently, pollutant emissions are reduced and the equipment life is extended. The generated control instructions are transmitted to the air supply system and the flexible sealed fume hood, and the boundary conditions of the real-time simulation system are updated synchronously to ensure the accuracy of the simulation results.

[0183] Preferably, when optimizing the smelting efficiency, the intelligent decision-making and control module can balance the relationship between the smelting efficiency, environmental protection indicators, and equipment life by adjusting the weight coefficients of the reward function. The dynamic programming algorithm can dynamically adjust the blast pressure, oxygen concentration, and feeding amount according to real-time data to ensure the optimal control of the system under different working conditions. The generated control instructions can be transmitted to the air supply system and the flexible sealed fume hood in real time to ensure the timeliness and accuracy of the system response. Synchronously updating the boundary conditions of the real-time simulation system can further improve the reliability of the simulation results and provide more accurate data support for intelligent decision-making.

[0184] In some embodiments, the flexible sealed fume hood performs the following steps:

[0185] Drive the silicone rubber sealing skirt to expand and contract through a pneumatic actuator, with a response time of less than 5 seconds, to fill the dynamic gap at the furnace mouth;

[0186] Combined with the pressure difference gradient inside and outside the fume hood, an adaptive PID controller is used to adjust the speed of the induced draft fan, and the control equation is as follows:

[0187]

[0188] In the formula, e(t) is the pressure difference deviation; S is the expansion and contraction amount of the sealing skirt; K f is the feedforward compensation coefficient;

[0189] Real-time monitor the SO2 leakage rate. If it exceeds 3 ppm / m 3 , trigger the secondary expansion of the skirt and increase the nitrogen curtain flow rate to 8 L / min.

[0190] It should be noted that in the flexible sealing hood of the present invention, the pneumatic actuator drives the expansion and contraction of the silicone rubber sealing skirt, with a response time of less than 5 seconds, which can effectively fill the dynamic gap at the furnace mouth. The pneumatic actuator is a mechanical device driven by air pressure, which can quickly and accurately control the expansion and contraction of the sealing skirt. The silicone rubber sealing skirt has good high-temperature resistance and elastic properties, which can adapt to the dynamic changes at the furnace mouth and ensure the sealing effect of the hood. Combining the pressure difference gradient inside and outside the hood, an adaptive proportional-integral-derivative controller is used to adjust the speed of the induced draft fan to ensure the stability of the negative pressure inside the hood and reduce the leakage of flue gas. The sulfur dioxide leakage rate is monitored in real time. If it exceeds the set threshold, the secondary expansion of the skirt and the increase of the nitrogen curtain flow rate are triggered to further reduce pollutant emissions.

[0191] Specifically, the flexible sealing hood drives the expansion and contraction of the silicone rubber sealing skirt through a pneumatic actuator, with a response time of less than 5 seconds, which can quickly fill the dynamic gap at the furnace mouth. The adaptive proportional-integral-derivative controller adjusts the speed of the induced draft fan according to the pressure difference gradient inside and outside the hood to ensure the stability of the negative pressure inside the hood and reduce the leakage of flue gas. The sulfur dioxide leakage rate is monitored in real time. If it exceeds 3 ppm per cubic meter, the secondary expansion of the skirt and the increase of the nitrogen curtain flow rate to 8 liters per minute are triggered to further reduce pollutant emissions.

[0192] Preferably, when adjusting the expansion and contraction amount of the sealing skirt of the flexible sealing hood, a rapid response can be achieved through a pneumatic actuator to ensure the timely filling of the dynamic gap at the furnace mouth. The adaptive proportional-integral-derivative controller can dynamically adjust the speed of the induced draft fan according to the pressure difference gradient inside and outside the hood to ensure the stability of the negative pressure inside the hood. When monitoring the sulfur dioxide leakage rate in real time, if the detected leakage rate exceeds the set threshold, the secondary expansion of the skirt and the increase of the nitrogen curtain flow rate can be triggered to ensure the effective control of pollutant emissions.

[0193] As Figure 2 shown. A schematic diagram of a converter provided by an embodiment of the present invention. The main structure of the converter includes components such as a furnace body 1, tuyeres 2, a charging port, and a flue gas outlet 3. The furnace body 1 is circular and is the main part of the converter, which is used to accommodate the melt to be blown. The tuyeres 2 are installed on the side of the furnace body 1 and are used to inject oxygen into the furnace to promote the oxidation reaction of the melt. The charging port and the flue gas outlet 3 are located at the top of the furnace body 1 and are used to add the raw materials to be blown and discharge the flue gas. In practical applications, the multi-modal perception system will monitor in real time data such as the composition of the melt in the furnace, the temperature field distribution, the operating state of the equipment, and the concentration of environmental pollutants. For example, the concentrations of elements such as Cu, Fe, and S in the melt are monitored by a high-temperature resistant fiber optic spectrometer, and the temperature field distribution on the surface of the melt at the furnace mouth is monitored by an infrared thermal imager. The intelligent decision-making and control module generates dynamic control instructions by predicting the flow field and pollutant diffusion path through a real-time simulation system based on these perception data. The air supply system adjusts the blast pressure and oxygen concentration according to the instructions, and the flexible sealing hood dynamically adjusts the expansion and contraction amount of the sealing skirt according to the instructions to inhibit the leakage of flue gas.

[0194] Specifically, the equipment status monitoring sub-module is installed at the tuyere 2 position of the furnace body 1. It collects the tuyere vibration signals through piezoelectric sensors, combines with the ultrasonic thickness measurement array to monitor the erosion thickness of refractory materials, and monitors the operation status of the equipment in real time, promptly discovers equipment failures and triggers maintenance instructions. The environmental monitoring sub-module is installed near the flue gas outlet 3. It emits laser with adjustable wavelength through Tunable Diode Laser Absorption Spectroscopy (TDLAS), scans the diffusion path of the hood axis, calculates the SO2 concentration field based on the absorption spectrum intensity distribution, monitors the pollutant concentration in the flue gas in real time, and provides data support for environmental protection control. The air supply system is installed on the side of the furnace body 1 and leads to the tuyere 2, and is used to supply air into the furnace. It adjusts the blast pressure and oxygen concentration according to the instructions of the intelligent decision-making and control module to optimize the oxygen supply during the smelting process. The flexible sealing hood is installed on the top of the furnace body 1 and covers the flue gas outlet 3. It drives the expansion and contraction of the silicone rubber sealing skirt through a pneumatic actuator, dynamically adjusts the expansion and contraction amount of the sealing skirt, suppresses flue gas leakage, effectively reduces pollutant emissions, and improves the environmental protection effect.

[0195] The above-mentioned various embodiments of the present invention have the following beneficial effects: The present invention can improve the intelligent level of the converter smelting process. Through the multi-modal perception system, it can collect multi-dimensional data such as the melt composition, temperature field distribution, equipment status, and environmental pollutant concentration in the furnace in real time. Combining with the dynamic optimization algorithm of the intelligent decision-making and control module, it can achieve precise adjustment of smelting parameters. This multi-modal data fusion and intelligent analysis ability can effectively improve the smelting efficiency, reduce energy consumption, while reducing manual intervention and enhancing production stability. The real-time simulation system can predict the flow field and pollutant diffusion path. Combining with the dynamic adjustment function of the flexible sealing hood, it can significantly suppress flue gas leakage and reduce pollutant emissions. The adaptive adjustment function of the air supply system can optimize the blast pressure and oxygen concentration, extend the service life of the equipment, and reduce the maintenance cost.

[0196] In addition, the present invention can improve the monitoring accuracy of the composition, temperature field, and pollutant concentration in the furnace through technical means such as multi-scale convolutional neural network and three-dimensional inversion algorithm. The equipment status monitoring sub-module can detect the slag accumulation at the tuyere and the erosion of refractory materials in real time, and promptly trigger maintenance and adjustment instructions to ensure the normal operation of the equipment. The intelligent decision-making and control module can optimize the smelting efficiency, environmental protection indicators, and equipment life based on the multi-objective reinforcement learning model, and generate collaborative control strategies. The flexible sealing hood can quickly respond to the dynamic gap changes at the furnace mouth, and combine with the adaptive proportional-integral-derivative controller to adjust the speed of the induced draft fan to ensure the stability of the negative pressure inside the hood and reduce pollutant emissions. Overall, the present invention can meet the urgent needs of modern metallurgical industry for efficient, environmental protection, and intelligent production.

[0197] The above description is only some preferred embodiments of the present invention and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present invention is not limited to the technical solutions formed by the specific combination of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the technical solutions formed by mutually replacing the above features with the technical features (but not limited to) having similar functions disclosed in the embodiments of the present invention.

Claims

1. An intelligent copper blowing system based on multimodal sensing, characterized in that: include: Multimodal perception system, intelligent decision-making and control module, real-time simulation system, air supply system and flexible sealed smoke hood; The multimodal sensing system collects data on melt composition, temperature field distribution, equipment operation status and environmental pollutant concentration in the furnace in real time; The intelligent decision-making and control module generates dynamic control instructions based on the output data of the multimodal perception system and predicts the flow field and pollutant diffusion path through a real-time simulation system; The air supply system adjusts the blast pressure and oxygen concentration according to the control instructions; The flexible sealed smoke hood dynamically adjusts the expansion and contraction amount of the sealing skirt plate according to the control instruction to suppress smoke leakage.

2. The system according to claim 1, characterized in that The multimodal perception system comprises: Furnace composition monitoring submodule, temperature field monitoring submodule, equipment status monitoring submodule and environment monitoring submodule; The furnace composition monitoring submodule receives the melt radiation spectrum based on a high-temperature resistant fiber optic spectrometer, and analyzes the concentrations of Cu, Fe, and S elements through a multi-scale convolutional neural network; the temperature field monitoring submodule scans the melt surface at the furnace mouth through an infrared thermal imager, and reconstructs the temperature field based on the dynamically calibrated Planck radiation law; the equipment status monitoring submodule collects the tuyere vibration signal through a piezoelectric sensor, and monitors the erosion thickness of the refractory material in combination with an ultrasonic thickness measurement array; the environment monitoring submodule tracks the SO2 escape path through a laser gas analyzer, and calculates the SO2 concentration field based on a three-dimensional inversion algorithm.

3. The system according to claim 2, characterized in that The furnace composition monitoring submodule performs the following steps: The radiation spectrum of the melt in the 200-1100nm band is received by a high-temperature resistant optical fiber spectrometer, and the nitrogen air curtain flow data is simultaneously collected to compensate for smoke interference; Perform baseline correction and noise filtering on the original spectral data to extract the characteristic peak intensity sequence; Based on the multi-scale convolutional neural network, the local spectral features and the global wavelength correlation are fused to calculate the concentrations of Cu, Fe, and S elements. The formula is as follows: In the formula, C Cu is the concentration of Cu element in the melt; is the intensity of the i-th characteristic wavelength; w i is the weight coefficient of the convolutional neural network output; k1, k2, k3 are dynamic calibration parameters; ΔQ N2 is the nitrogen air curtain flow deviation; Cross-validate the element concentration data with the flow field prediction results of the real-time simulation system. If the deviation exceeds the preset value, trigger the spectrometer self-calibration mode. The calibrated concentration data is synchronized to the intelligent decision-making and control module through the edge computing node.

4. The system according to claim 2, characterized in that The temperature field monitoring submodule performs the following steps: The infrared thermal imager scans the melt surface at the furnace mouth with a resolution of 640×480, and generates a radiation energy distribution map every 10 seconds; Based on the dynamic blackbody source to calibrate the emissivity parameters, the temperature field is calculated in combination with Planck's radiation law. The formula is as follows: Where T(x,y) is the temperature at the coordinate (x,y); h is Planck's constant; c is the speed of light; λ is the working wavelength of the infrared thermal imager; k B is the Boltzmann constant; ∈(λ) is the surface emissivity of the melt after dynamic calibration; L meas (x, y) is the measured radiation intensity; The temperature field cloud map is integrated with the composition data in the furnace, and the melt viscosity distribution map is generated based on the viscosity-temperature correlation model. The formula is as follows: Where, μ(T) is the melt viscosity; A, B, T0 are material characteristic constants; When the local temperature gradient exceeds 200℃ / m, the air pressure adaptive adjustment instruction of the air distribution system is triggered, and the adjustment amount calculation formula is: In the formula, ΔP is the wind pressure adjustment amount; K is the thermal gradient sensitivity coefficient; is the temperature gradient.

5. The system according to claim 2, characterized in that The equipment status monitoring submodule performs the following steps: The vibration signal of the air outlet is collected by a piezoelectric sensor in the frequency range of 0.5-10kHz, and the blast pressure data is recorded synchronously; The vibration signal is decomposed by wavelet packets to extract the energy accumulation value of the 3-5kHz frequency band. The formula is as follows: Where W k (τ) is the wavelet packet coefficient of the kth sub-band, A clog is the energy accumulation value, τ is the time variable; If A clog >0.8V and the blast pressure fluctuation rate exceeds 15%, it is judged as slag accumulation at the tuyere, triggering the high-pressure nitrogen pulse backflush sequence; The ultrasonic thickness measurement array emits a 5MHz pulse wave and calculates the thickness of the refractory material based on the echo time difference: In the formula, δ corr is the temperature compensation term; v is the propagation speed of ultrasonic waves in refractory materials; When the erosion thickness is less than 50 mm, a maintenance work order is generated and the local air blowing parameters are optimized to reduce thermal stress shock.

6. The system according to claim 2, characterized in that The environment monitoring submodule performs the following steps: The tunable laser gas analyzer emits 1.53μm wavelength laser to scan the SO2 diffusion path along the hood axis; Based on the absorption spectrum intensity distribution, the three-dimensional inversion algorithm is used to calculate the SO2 concentration field. The formula is as follows: Where, T ref is the reference temperature; T(x,y,z) is the real-time temperature field data; C SO2 (x,y,z) is the SO2 concentration field at the left side (x,y,z); L is the optical path length; I0(x,y,z) is the incident light intensity; I(x,y,z) is the transmitted light intensity; Compare the SO2 concentration field with the CFD-DEM prediction results of the real-time simulation system. If the deviation exceeds 10%, optimize the hood negative pressure PID control parameters. When the local SO2 concentration exceeds 50ppm, the sealing skirt expansion instruction is triggered and the induced draft fan speed is increased to 120%.

7. The system according to claim 1, characterized in that The real-time simulation system performs the following steps: Based on proper orthogonal decomposition, 100 main modes are extracted from the high-fidelity CFD database to construct a low-dimensional flow field approximation space. The melt flow rate and wall heat load are predicted by coupling the dynamic change of melt viscosity with the improved Realizable k-ε turbulence model. The control equation is as follows: In the formula, G b is the buoyancy term; Y M is the compressible turbulence correction term; ν is the kinematic viscosity; S is the strain rate tensor modulus; ρ is the fluid density; k is the turbulent kinetic energy; μ t is the turbulent viscosity; μ is the dynamic viscosity; σ k , σ ∈ is the turbulence model constant; G k is the kinetic energy generated by the average velocity gradient;∈ is the turbulent heat consumption rate; The CFD-DEM coupling method is used to simulate the movement of flue gas particles and calculate the SO2 diffusion path and wall deposition amount; The simulation results are integrated with the multimodal perception data to generate a dynamic control instruction set and feed it back to the intelligent decision-making module.

8. The system according to claim 7, characterized in that The CFD-DEM coupling method includes: The contact force and motion trajectory between particles are calculated by discrete element method, and the formula is as follows: In the formula, k n , k t are the normal and tangential stiffness coefficients; γ n , γ t is the damping coefficient; δ n , δ t are normal and tangential displacements; F c,ij is the contact force between particles i and j; is the rate of change of normal and tangential directions; n and t are the normal and tangential unit vectors; The particle-fluid interaction force is calculated by dynamically switching the Gidaspow drag model based on the local porosity; The particle motion data is mapped to the CFD grid and the source term in the gas phase momentum equation is updated as follows: Where δ(·) is the Dirac function; N p is the number of particles in the grid; S u is the source term of the energy equation; F d,i is the drag force on particle i; The bidirectional coupling equations are solved implicitly and iteratively to output high-precision smoke diffusion simulation results.

9. The system according to claim 1, characterized in that The intelligent decision and control module performs the following steps: A multi-objective reinforcement learning model is constructed, with smelting efficiency, environmental protection indicators and equipment life as optimization objectives, and the reward function is designed as follows: In the formula, α, β, γ are weight coefficients; Q melt is the real-time smelting amount; C thr is the emission threshold; d crit is the critical erosion thickness; Q max is the theoretical maximum melting amount; C SO2 is the real-time concentration of SO2; C thr is the emission threshold; Δd is the real-time erosion thickness of the refractory material; The coordinated control strategy of blast pressure, oxygen concentration and feed amount is generated by dynamic programming algorithm. The formula is as follows: In the formula, K p , K d is the gain coefficient; ΔT is the temperature field uniformity index; ΔP 风压 is the blast pressure adjustment amount; The control instructions are transmitted to the air supply system and the flexible sealed smoke hood, and the boundary conditions of the real-time simulation system are updated synchronously.

10. The system according to claim 1, characterized in that The flexible sealed fume hood performs the following steps: The pneumatic actuator drives the silicone rubber sealing skirt to expand and contract, with a response time of less than 5 seconds to fill the dynamic gap at the furnace mouth; Combined with the pressure difference gradient inside and outside the hood, the adaptive PID controller is used to adjust the speed of the induced draft fan. The control equation is as follows: Where, e(t) is the pressure difference deviation; S is the expansion and contraction of the sealing skirt; K f is the feedforward compensation coefficient; u(t) is the speed control value of the induced draft fan; The SO2 leakage rate is monitored in real time. If it exceeds the preset leakage rate threshold, the skirt plate will be expanded twice and the nitrogen curtain flow rate will be increased.

Images