Process for dealkalizing and deeply reducing red mud through gradient temperature control treatment

Abstract

The invention discloses a process for dealkalizing and deeply reducing red mud through gradient temperature control treatment, which comprises the following steps: 1) accurately metering raw materials and grinding; 2) feeding the ground raw materials into a red mud stirrer, and stirring at a gradient rotating speed; 3) feeding the stirred and mixed raw materials into a granulator for granulation, and pressing the materials into particle materials with the particle size of 5-15mm; 4) feeding the particle material into a dealkalization solid-phase reaction device, and forming a particle material after a solid-phase reaction in a reducing atmosphere; (5) feeding the particle material after the solid-phase reaction into a cooling machine, and slowly cooling in a furnace by adopting a nitrogen direct circulating cooling method; 6) crushing the cooled material by a crusher until the particle size is less than 5mm; 7) performing magnetic separation on the ground powder by a high-gradient magnetic separator.According to the method, efficient dealkalization and iron element reduction separation are realized through high-temperature dealkalization pre-reduction and high-temperature deep reduction melt separation.

Description

Technical Field

[0001] This invention belongs to the field of industrial process control and solid waste resource utilization technology, specifically relating to a process for dealkalizing and deeply reducing red mud through cascade temperature control. Background Technology

[0002] Red mud is a solid waste generated during the alumina production process from bauxite. Its composition is complex, containing high levels of elements such as iron, aluminum, and silicon, as well as certain amounts of heavy metals and radioactive substances. With the continuous growth of global alumina production, red mud emissions are also increasing dramatically, placing enormous pressure on the environment. The large-scale accumulation of red mud not only occupies significant land resources but may also pollute water bodies and soil through rainwater leaching, impacting ecosystem health. Deep reduction of red mud for iron extraction is a complex physicochemical process involving high temperatures, multiple phases, and multi-field coupling, with a complex mechanism.

[0003] Red mud contains alkali, which (mainly sodium) exists in various forms, including soluble alkali and bound alkali. Existing red mud treatment processes cannot effectively remove alkali, nor can they achieve efficient alkali removal and iron reduction separation. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a process for dealkalizing and deep reducing red mud by step temperature control, which achieves efficient dealkalization and iron reduction separation through high-temperature dealkalization pre-reduction and high-temperature deep reduction melting.

[0005] To address the aforementioned technical problems, this invention provides a process for dealkalizing and deeply reducing red mud through a tiered temperature-controlled treatment method, comprising the following steps: 1) Precise measurement of raw materials and grinding process: Using pretreated red mud as the base material, a reducing agent with a carbon content of ≥70% is fixed, and a composite additive is selected as the additive. The mass ratio of red mud, reducing agent and composite additive is 100:20-30:5-10; the raw materials with the above mass ratio are ground into powder. 2) The ground raw materials are fed into a red mud mixer and stirred using a gradient speed to ensure that the material mixing variation coefficient is ≤8%; 3) Feed the mixed raw materials into a granulator to granulate them into granules with a particle size of 5-15mm. 4) The granular material is fed into the dealkali solid-phase reaction device. Under a reducing atmosphere, the reaction temperature is controlled at 850℃±20℃ and the reaction time is 2-4h to remove alkali metals from the material and form solid-phase granular material. 5) The particulate material after the solid-phase reaction is fed into a cooler and cooled by direct nitrogen circulation. The cooling temperature is controlled to gradually decrease from 900℃ to ≤150℃, and the cooling time is 1.5-3h; then the furnace is cooled slowly. 6) After cooling, the material is crushed by a crusher to a particle size of less than 5mm. The crushed finished material enters the grinding section to be ground to 200-400 mesh. 7) The powder after grinding is separated by a high gradient magnetic separator.

[0006] 2. The process for dealkalization and deep reduction of red mud by step temperature control as described in claim 1 is characterized in that: in step 1), the reducing agent is coke powder with a fixed carbon content of 75%, the composite additive is bentonite with a bonding strength of 0.6-0.8 MPa, and the mass ratio of red mud, coke powder and bentonite is 100:25-30:6-8.

[0007] 3. The process for dealkalization and deep reduction of red mud by cascade temperature control as described in claim 1 is characterized in that: in step 2), a vertical turbulent flow mixer is used, the mixing shaft is equipped with 3-5 layers of blades, the blade angle is adjustable from 30-60°, and the mixing is first carried out at a low speed of 200-300 r / min for 3-5 min, and then at a high speed of 300-500 r / min for 8-10 min to ensure that the material mixing variation coefficient is ≤8%; by adding some external water, the moisture content is adjusted to 14-16%.

[0008] 4. The process for dealkalization and deep reduction of red mud by step temperature control as described in claim 1 is characterized in that: in step 3), after granulation by a granulator, the prepared 5-15mm granular material is sent to a drum dryer with an internal temperature of 200-350℃ for drying pretreatment, and the moisture content is dried from <16% to <2%.

[0009] 5. The process for dealkalization and deep reduction of red mud by cascade temperature control as described in claim 1, characterized in that: in step 4), an online anti-ringing control method is used to automatically control the solid-phase reactor; the temperature detection unit of the multi-source sensing module collects the temperature signals of the inside and outside of the solid-phase reactor in real time; the process parameter acquisition unit collects the current and torque signals of the main drive motor and the flue gas atmosphere composition signal; the visual image detection unit collects the material flow state inside the solid-phase reactor and the visual image of the wall surface; the edge calculation and feature extraction module filters, reduces noise, and standardizes the signals collected by the multi-source sensing module, and extracts feature values ​​related to ring formation, including axial temperature gradient, radial temperature deviation, area and movement trend of high-temperature region, fluctuation variance of drive torque, current trend slope, residence time in the preset temperature range, change in texture roughness of the inner wall of the solid-phase reactor, and physical... The pixel ratio of the adhesion area; the ring formation risk assessment and decision-making module receives data from the edge computing and feature extraction modules. The risk fusion model uses a multivariate fusion algorithm based on weight allocation to calculate the ring formation risk index (CRI). The CRI is a continuous value between 0 and 100, with higher values ​​indicating greater ring formation risk. The control decision generation unit generates targeted control commands based on the CRI and the feature value with the greatest contribution. The execution intervention module continuously monitors or issues warnings based on the control commands, or adjusts the material ratio and particle size through the material composition and particle size adjustment unit, or adjusts the operating temperature, atmosphere, or main drive motor speed through the operating parameter adjustment unit. The analysis module evaluates the effectiveness of the intervention measures, and the storage module records successful cases for optimizing the parameters of the risk fusion model and the control commands of the control decision generation unit.

[0010] 6. The process for dealkalization and deep reduction of red mud by cascade temperature control as described in claim 5, characterized in that: the ring formation risk index adopts a multivariate fusion algorithm based on weight allocation, and its expression is: CRI=100×F fusion ([W t ×F t W v ×F v W p ×F p ];Θ) in: F fusion (·): Top-level fusion function, which combines the weighted feature values ​​of each dimension into a single risk scalar; F t ,F v ,F p : These represent the normalized feature values ​​of the three dimensions: temperature field, visual image, and process parameters, respectively, with each value ranging from [0,1]. Wt W v W p : These correspond to the dynamic weight coefficients of each dimension, satisfying W t +W v +W p =1; Θ: The set of internal parameters of the fusion function; Multiplier 100: Maps the results to an intuitive risk index range of 0-100; Temperature field characteristic value F t The calculation method is as follows: F t =min(1,(ω1*N(ΔT)+ω2*N(G)+ω3*N(τ)) / (ω1+ω2+ω3)) in: N(ΔT) = min(1, (ΔT - ΔT) th ) / ΔT scale ): Local hotspot temperature difference anomaly, ΔT is the temperature difference between the current hotspot and the regional average temperature. th The threshold for initiating interest, ΔT scale This is the scaling factor; N(G)=|G actual -G baseline | / G range : Axial temperature gradient anomaly index, G actual To calculate gradients in real time, G baseline Based on historical normal benchmarks, G range This is within the normal fluctuation range; N(τ)=(τ measured -τ min ) / (τ max -τ min ): External wall temperature conduction delay rate, τ measured τ represents the measured delay time. min and τ max For the minimum and maximum theoretical delays; ω1, ω2, ω3: Sub-feature weights, corresponding to the weights of local temperature, axial temperature, and outer wall temperature, respectively, preset parameters; Visual image feature value F v The calculation method is as follows: F v =SVM(I t ,I {t-1} ,I {t-2} ,...) A trained machine learning model was used to analyze a series of endoscopic images. The model output the probability of determining the "loop" state, which was then compressed to [0,1] by the Sigmoid function and used as F. vThe value directly reflects the strength of visual evidence of wall adhesion, and the machine learning model is a support vector machine (SVM). Process parameter characteristic value F p The calculation method is as follows: F p =max(N(σ T ),N(φ),N(Δt)) in: N(σ T )=σ Tcurrent / σ Tnormal The ratio of the variance in drive torque fluctuation reflects changes in mechanical load. N(φ) = |φ actual -φ optimal | / φ tolerance φ: Deviation of reduction potential reflects an abnormal chemical reaction environment; optimal For optimal setting, φ tolerance Allowable deviation; N(Δt)=(t residence -t setpoint ) / t range : The change in the residence time of the material within the preset temperature range.

[0011] After adopting the above process, the high-temperature dealkali pre-reduction + high-temperature deep reduction melting process in this process achieves efficient dealkali removal and iron element reduction separation, ensuring the long-term stable operation of the production line. Recycled alkali reuse: (1) Existence form; Bayer process red mud is a solid residue after alumina extraction from bauxite. Its alkali (mainly sodium) exists in various forms, which can be divided into soluble alkali and bound alkali: Free alkali: including unwashed NaOH, Na2CO3 and a small amount of Na2SO4, which are attached to the surface of red mud particles or pores and are easily dissolved by water washing. Adsorbed alkali: Na + Adsorbed on the surface of minerals such as silicates and iron oxides in red mud, they can be released through ion exchange or acid treatment. Sodium aluminosilicate (such as sodalite and nepheline): The alkali is trapped in the crystal structure of aluminosilicate and is the main form of alkali in red mud (accounting for more than 60% of the total). Alkali-containing iron minerals: Some sodium forms solid solutions or surface complexes with iron oxides (such as hematite) in red mud. Alkali metal silicates: Glassy phase substances such as sodium silicate (Na2SiO3) generated by high-temperature reaction. (2) Release mechanism of free alkali during medium-high temperature roasting and reduction: Medium-high temperature roasting (usually 800–1100°C) destroys the stable structure of alkali in red mud through physicochemical reaction, transforming it into volatile substances. Release: Chemical reaction process Dehydration and decomposition: Aluminum hydroxide, goethite, and other minerals in red mud decompose into oxides, releasing structural water.

[0012] Sodium carbonate decomposition: Na2CO3→Na2O+CO2 (>850°C).

[0013] 2) Destruction of mineral structure: Aluminosilicates (such as sodalite) react with additives (such as CaO and CaCO3) at high temperatures. Disrupting the [SiO4]-[AlO4] network releases Na2O: Na6[Al6Si6O24]·2Na2CO3+CaO→Ca2Al2SiO7+Na2O↑Na6[Al6Si6O24]·2Na2CO3+CaO→Ca2Al2SiO7+Na2O↑ Alkali release under reducing atmosphere: In the presence of reducing agents (carbon, CO, H2), iron oxides (Fe2O3) in red mud are reduced to Fe or FeO, promoting the decomposition of alkali-containing minerals. NaAlSiO4+C→Na↑+Al2O3+SiO2+CONaAlSiO4+C→Na↑+Al2O3+SiO2+CO Sodium oxide (Na2O) volatilizes at high temperatures or reacts with SiO2 to generate volatile Na2SiO3 vapor.

[0014] (3) Recycling 1) Wet scrubbing: Spraying the flue gas with dilute alkaline solution or water dissolves soluble sodium salts (NaOH, Na2CO3), generating a rich alkaline solution (Na... + Concentrations can reach 20–50 g / L.

[0015] 2) The washing liquid is recycled and concentrated and then returned to the Bayer process batching system to replace part of the fresh alkali solution.

[0016] 3. Multi-metal simultaneous recycling: Titanium metal is recycled simultaneously while iron is being recycled.

[0017] 4. Nitrogen quenching effectively reduces the iron oxidation rate: (1) Completely avoid secondary oxidation of metallic iron and significantly improve iron recovery rate; (2) Refine the metallic iron grains to improve magnetic separation efficiency and iron concentrate grade; (3) High heat recovery and utilization rate, significantly reducing energy consumption.

[0018] . Attached Figure Description

[0019] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings: Figure 1 This is the production flow chart of the present invention (before granulation); Figure 2This is a production flow chart of the present invention (after granulation); Figure 3 This is a schematic diagram of the magnetic separation process of the present invention; Figure 4 This is a raw material analysis table of red mud from one embodiment; Figure 5 This is an ideal elemental analysis table of the material after roasting and reduction, according to one embodiment; Figure 6 This is an ideal elemental analysis table of the material after roasting and reduction, as exemplified by one embodiment. Detailed Implementation

[0020] Referring to the accompanying drawings, this invention provides a process for dealkalizing and deeply reducing red mud through cascade temperature control. The purpose is to use solid-state roasting reduction technology combined with digital twin technology to achieve efficient dealkalization and separation of iron and titanium elements, thereby processing and fully utilizing the red mud.

[0021] refer to Figure 1 As shown in the figure, the flowchart of the entire system is illustrated. This process for dealkalizing and deeply reducing red mud through a stepped temperature-controlled treatment includes the following steps: 1) Precise measurement of raw materials and grinding process: Using pretreated red mud as the base material, a reducing agent with a carbon content of ≥70% is fixed, and a composite additive is selected as the additive. The mass ratio of red mud, reducing agent and composite additive is 100:20-30:5-10; the raw materials with the above mass ratio are ground into powder. 2) The ground raw materials are fed into a red mud mixer and stirred using a gradient speed to ensure that the material mixing variation coefficient is ≤8%; 3) Feed the mixed raw materials into a granulator to granulate them into granules with a particle size of 5-15mm. 4) The granular material is fed into the dealkali solid-phase reaction device. Under a reducing atmosphere, the reaction temperature is controlled at 850℃±20℃ and the reaction time is 2-4h to remove alkali metals from the material and form solid-phase granular material. 5) The particulate material after the solid-phase reaction is fed into a cooler and cooled by direct nitrogen circulation. The cooling temperature is controlled to gradually decrease from 900℃ to ≤150℃, and the cooling time is 1.5-3h; then the furnace is cooled slowly. 6) After cooling, the material is crushed by a crusher to a particle size of less than 5mm. The crushed finished material enters the grinding section to be ground to 200-400 mesh. 7) The powder after grinding is separated by a high gradient magnetic separator.

[0022] In step 1), the reducing agent is coke powder with a fixed carbon content of 75%, and the composite additive is bentonite with a bonding strength of 0.6-0.8 MPa. The mass ratio of red mud, coke powder and bentonite is 100:25-30:6-8.

[0023] In step 2), a vertical turbulent mixer is used, with the mixing shaft equipped with 3-5 layers of blades and the blade angle adjustable from 30 to 60°. First, the mixture is stirred at a low speed of 200-300 r / min for 3-5 minutes, and then stirred at a high speed of 300-500 r / min for 8-10 minutes to ensure that the material mixing variation coefficient is ≤8%. The moisture content is adjusted to 14-16% by adding some external water.

[0024] In step 3), after granulation by the granulator, the prepared 5-15mm granules are sent to a drum dryer with an internal temperature of 200-350℃ for drying pretreatment, and the moisture content is dried from <16% to <2%.

[0025] In step 4), an online anti-ringing control method is used to automatically control the solid-phase reactor. The temperature detection unit of the multi-source sensing module collects temperature signals from the inside and outside of the solid-phase reactor in real time. The process parameter acquisition unit collects the current and torque signals of the main drive motor and the flue gas atmosphere composition signal. The visual image detection unit collects the material flow state inside the solid-phase reactor and the visual image of the wall. The edge computing and feature extraction module filters, reduces noise, and standardizes the signals collected by the multi-source sensing module, and extracts feature values ​​related to ringing. The feature values ​​include axial temperature gradient, radial temperature deviation, area and movement trend of high-temperature region, fluctuation variance of drive torque, current trend slope, residence time in the preset temperature range, change in texture roughness of the inner wall of the solid-phase reactor, and pixel ratio of physical adhesion area. Ringing risk assessment and decision-making are then performed. The module receives data from the edge computing and feature extraction modules. The risk fusion model uses a multivariate fusion algorithm based on weight allocation to calculate the ring formation risk index (CRI). The CRI is a continuous value between 0 and 100, with higher values ​​indicating greater ring formation risk. The control decision generation unit generates targeted control commands based on the CRI and the feature value with the greatest contribution. The intervention execution module continuously monitors or issues warnings based on the control commands, or adjusts the material ratio and particle size through the material composition and particle size adjustment unit, or adjusts the operating temperature, atmosphere, or main drive motor speed through the operating parameter adjustment unit. The analysis module evaluates the effectiveness of the intervention measures, and the storage module records successful cases for optimizing the parameters of the risk fusion model and the control commands of the control decision generation unit.

[0026] The clustering risk index employs a multivariate fusion algorithm based on weight allocation, and its expression is as follows: CRI=100×Ffusion ([W t ×F t W v ×F v W p ×F p ];Θ) in: F fusion (·): Top-level fusion function, which combines the weighted feature values ​​of each dimension into a single risk scalar; F t ,F v ,F p : These represent the normalized feature values ​​of the three dimensions: temperature field, visual image, and process parameters, respectively, with each value ranging from [0,1]. W t W v W p : These correspond to the dynamic weight coefficients of each dimension, satisfying W t +W v +W p =1; Θ: The set of internal parameters of the fusion function; Multiplier 100: Maps the results to an intuitive risk index range of 0-100; Temperature field characteristic value F t The calculation method is as follows: F t =min(1,(ω1*N(ΔT)+ω2*N(G)+ω3*N(τ)) / (ω1+ω2+ω3)) in: N(ΔT) = min(1, (ΔT - ΔT) th ) / ΔT scale ): Local hotspot temperature difference anomaly, ΔT is the temperature difference between the current hotspot and the regional average temperature. th The threshold for initiating interest, ΔT scale This is the scaling factor; N(G)=|G actual -G baseline | / G range : Axial temperature gradient anomaly index, G actual To calculate gradients in real time, G baseline Based on historical normal benchmarks, G range This is within the normal fluctuation range; N(τ)=(τ measured -τ min ) / (τ max -τ min ): External wall temperature conduction delay rate, τ measured τ represents the measured delay time. min and τmax For the minimum and maximum theoretical delays; ω1, ω2, ω3: Sub-feature weights, corresponding to the weights of local temperature, axial temperature, and outer wall temperature, respectively, preset parameters; Visual image feature value F v The calculation method is as follows: F v =SVM(I t ,I {t-1} ,I {t-2} ,...) A trained machine learning model was used to analyze a series of endoscopic images. The model output the probability of determining the "loop" state, which was then compressed to [0,1] by the Sigmoid function and used as F. v The value directly reflects the strength of visual evidence of wall adhesion, and the machine learning model is a support vector machine (SVM). Process parameter characteristic value F p The calculation method is as follows: F p =max(N(σ T ),N(φ),N(Δt)) in: N(σ T )=σ Tcurrent / σ Tnormal The ratio of the variance in drive torque fluctuation reflects changes in mechanical load. N(φ) = |φ actual -φ optimal | / φ tolerance φ: Deviation of reduction potential reflects an abnormal chemical reaction environment; optimal For optimal setting, φ tolerance Allowable deviation; N(Δt)=(t residence -t setpoint ) / t range : The change in the residence time of the material within the preset temperature range.

[0027] The above calculation of the Circle Risk Index (CRI) is as follows: Case background: After 48 hours of continuous operation, the central control system of a red mud reduction solid-phase reactor (rotary kiln) indicates slight anomalies in several parameters. The task is to calculate the Circle Risk Index (CRI) at the current time (t) based on real-time data to determine whether intervention measures should be taken.

[0028] Step 1: Real-time data acquisition and preliminary feature extraction (corresponding to feature values ​​of each dimension (F) i (Calculation method part) Suppose the following raw data is obtained from a sensor network: 1. Temperature field data: Current hot spot and regional average temperature difference ΔT: A hot spot was detected in the high-temperature section of the kiln body, and the temperature at this point was 65°C higher than the average temperature of this section.

[0029] Real-time calculation gradient G of axial temperature actual The calculated current gradient is 12℃ / m.

[0030] Measured delay time τ of external wall temperature conduction measured The measured time delay for internal temperature changes to be transmitted to the corresponding point on the outer wall was 22 minutes.

[0031] 2. Visual image data: Image sequences of the current frame and historical frames captured by a high-temperature resistant endoscope [I] t ,I {t-1} ,I {t-2} After image preprocessing, the extracted texture feature vectors are input into the pre-trained SVM model.

[0032] 3. Process parameter data: Main drive torque fluctuation variance σ Tcurrent The calculated current value is 1.8 N·m. 2 .

[0033] reduction potential φ actual The online gas analyzer measured the CO / (CO+CO2) ratio to be 0.68.

[0034] Material residence time Δt: According to model calculation, the actual residence time of the material in the preset range of 1150-1250℃ is 41 minutes.

[0035] Preset parameters (based on historical data and actual production experience): ΔT th =50℃,ΔT scale =100℃ G baseline =8℃ / m,G range =10℃ / m τ min =15min,τ max =30min σ Tnormal =1.0 N·m² φ optimal =0.72,φ tolerance =0.1 t setpoint =45min,t range =20min Temperature sub-feature weights: ω1=0.5, ω2=0.3, ω3=0.2 Step 2: Calculate the eigenvalues ​​(F) t ,F v ,F p ) 1. Calculate the eigenvalues ​​F of the temperature field. t : N(ΔT)=min(1,(65-50) / 100)=0.15 N(G) = |12 - 8| / 10 = 0.4 N(τ) = (22-15) / (30-15) ≈ 0.467 F t =min(1,(0.5*0.15+0.3*0.4+0.2*0.467) / (0.5+0.3+0.2))=min(1,0.278)=0.278 The overall anomaly of the temperature field is approximately 27.8%, mainly due to a larger axial gradient and a slight increase in conduction delay.

[0036] 2. Calculate the visual image feature value F v : Input the image feature vector into the SVM model. Assume that the model outputs the probabilities of three categories: [normal: 0.25, adherent: 0.60, clump: 0.15].

[0037] The probability of the "clustering" class is taken as the base value, and the difference is enhanced by the Sigmoid function (assuming Sigmoid(0.15*5)=0.68).

[0038] F v =0.68 The visual model indicates obvious signs of material adhesion and a high probability of ring formation, with an eigenvalue of 68%.

[0039] 3. Calculate the characteristic value F of the process parameter. p : N(σ T = 1.8 / 1.0 = 1.8 (Above 1 indicates increased volatility) N(φ) = |0.68 - 0.72| / 0.1 = 0.4 N(Δt) = |41 - 45| / 20 = 0.2 F p =max(1.8,0.4,0.2)=1.8 (According to the definition, the upper limit is 1, but this is for calculation demonstration purposes, and min(1,Fp) will be used later) F p =min(1,1.8)=1.0.

[0040] The variance of the driving torque fluctuation has exceeded the normal range by 100%, becoming the most significant risk item among the process parameters. The eigenvalue has reached the upper limit of 100%, indicating that the macroscopic operating load has become abnormal.

[0041] Step 3: Determine the dynamic weights (W t , W v , W p ), (corresponding to the adaptive calculation model part of the dynamic weight coefficient W i ) Assume that the current working condition θ is: high raw material alkalinity (B), steady state (S) in the equipment operation stage, and stable load rate (L).

[0042] Use fuzzy logic to calculate: 1. Activate the rule in the rule base corresponding to "high B and steady S": `IF B is high AND S is steady THEN (W v is large, W t is large, W p is small)`.

[0043] 2. Through defuzzification (centroid method), calculate the exact weights: W v = 0.50 (the visual weight is the largest because high-alkalinity red mud is prone to adhesion, and the risk of ring formation in the steady state is mainly reflected in vision and temperature) W t = 0.35 (the temperature weight is the second) W p = 0.15 (the process parameter weight is relatively small because it is usually the result of ring formation rather than the earliest cause) Satisfy W t + W v + W p = 1.

[0044] Step 4: Apply the top-level fusion function to calculate CRI (corresponding to the implementation part of the top-level fusion function F fusion ): Method A: Use linear weighted fusion CRI A = 100×(0.35×0.278 + 0.50×0.68 + 0.15×1.0) ≈ 58.7 Method B: Use non-linear fusion (take the generalized mean with γ = 2) First calculate the sum of squared weights: `Sum = 0.35*(0.278^2)+0.50*(0.68^2)+0.15*(1.0^2)=0.027+0.231+0.15=0.408` Divide by the feature number 3 (or by weighted sum, simplified to 3 here): `Avg=0.408 / 3=0.136` Finally, take the square root: CRI B =100×sqrt(0.136)=100×0.369≈36.9 Comparative analysis: Linear fusion (58.7): Smoothly integrates all features, with values ​​within the warning range.

[0045] Nonlinear fusion (γ=2, 36.9): Due to the high-value features (F p The square amplification effect of (=1.0) is balanced by the square root, and F t The lower value caused the final result to be pulled down, making it closer to the risk level shown by the temperature field.

[0046] The decision was made based on the most likely linear fusion result, CRI=58.7.

[0047] Step 5: Early Warning and Decision Trigger (corresponding to the Risk Assessment and Decision-Making section) CRI=58.7, falling within the warning level (30≤CRI<70) range.

[0048] The system responds automatically: 1. Central Control Room HMI: A yellow warning box pops up on the interface, displaying "Circulation Risk Warning (CRI=58.7)", and highlights the main contributing characteristic as "Process Parameter Anomaly (F)". p =100%) and "Visual adhesion signs (F)" v =68%).

[0049] 2. Issue audio and visual alerts to remind the operator to pay attention.

[0050] 3. Automatically activate the primary adjustment strategy: Command 1: Fine-tune the burner to lower the set temperature by 10°C to reduce local hot spots.

[0051] Instruction 2: Increase the rotary kiln speed by 3% from the existing speed and maintain it for 15 minutes to enhance material rolling and suppress adhesion layer stability.

[0052] Tip: It is recommended that the laboratory shifts increase the frequency of testing the alkalinity of raw materials.

[0053] Step Six: Learning and Optimization (corresponding to the online learning and optimization mechanism of the model) The system stores this event (timestamp t, feature vector [0.278, 0.68, 1.0], weight [0.35, 0.50, 0.15], calculated CRI=58.7, triggering early warning level intervention) as a case sample.

[0054] Operators can subsequently review the effectiveness of this alert's handling. For example, one hour later, if visual feature F... v The temperature-speed coordinated intervention was successfully implemented, with the CRI dropping to 40 and the CRI falling back to 0.4. This experience was stored in the case library to optimize control parameters under similar operating conditions in the future.

[0055] This invention achieves this by introducing a calculation model for the Cog Risk Index (CRI): 1. Risk Quantification: Transforms vague judgments based on experience into precise numerical indicators based on multi-source data fusion, solving the problem of objectively assessing the risk of clustering.

[0056] 2. Early warning: By fusing visual and temperature features that are sensitive to early signs (such as texture changes and slight gradient anomalies), it is possible to issue an early warning before physical rings are clearly formed.

[0057] 3. Adaptive decision-making: The dynamic weighting mechanism enables the system to intelligently adjust the confidence level of different evidence based on changes in operating conditions such as raw materials and load, thereby improving the robustness of risk assessment.

[0058] 4. Closed-loop optimization: The built-in learning mechanism enables the CRI model and control system to continuously improve itself using production feedback data, thereby continuously enhancing the level of intelligence.

[0059] The Coiling Risk Index (CRI) calculation system constitutes the "digital brain" of the anti-coiling intelligent control system of this invention, providing core technical support for the invention to move from passive handling to active prevention.

[0060] This invention uses one embodiment to illustrate the above process, see reference. Figures 1 to 3 As shown in the table, the red mud raw materials used are as follows. Red mud raw material processing equipment: Reducing agent feeding system: The reducing agent and raw materials are fed into the box feeder by a forklift. The mill is started and adjusted to meet the feeding conditions. The mill fan frequency and its pipeline valves are turned on. The elevator is started, and the corresponding feeding equipment is opened to send the reducing agent and raw materials into the mill for grinding. The ground powder enters the powder conveying hopper. The system automatically turns on the blower, opens the feed valve and exhaust valve, and closes the discharge valve and air inlet valve. When the material comes into contact with the level gauge, the level gauge sends a full signal, and the pneumatic feed valve and exhaust valve automatically close, completing the feeding process. When the hopper pump pressure reaches the specified value... When the set value is reached, the powder conveying fan is turned on, the discharge valve is automatically opened, and the material fluidization is enhanced, starting to be conveyed to the reducing agent powder silo; the pressure value gradually decreases from high to low, and the material in the silo pump remains in a semi-fluidized state throughout this process; when the pressure drops to the pipeline resistance, the air inlet valve is closed, the discharge valve is closed after a certain interval, the pressurization continues for 3-5 seconds, the pipeline is cleaned, the inlet valve is opened, and one conveying cycle is completed; the reducing agent silo level gauge determines whether the powder silo pump needs to feed according to the silo setting height; the binder silo is fed by tanker truck, and the control room is notified when the tanker truck arrives. Turn on the binder dust collector; the conveying capacity is determined according to the height set by the binder silo level gauge. Red mud inlet processing control system: the box feeder determines whether to start the plate chain conveyor based on the set weight signal; the distributor position is set to ensure red mud enters the box feeder; the dryer outlet fan is turned on; the pressure of the drying drum outlet duct is controlled to approximately -10000Pa; the drying drum is turned on to dehydrate the red mud; the red mud drying system burner is started and operates automatically according to the temperature setting. To ensure the moisture content of the dried material is <20%, the inlet temperature is set to 300-400°C. The outlet temperature is controlled within 100-120 degrees Celsius to maximize drying efficiency. The drying drum is a sealed structure. The outlet fan is adjusted appropriately based on the inlet and outlet pressure values ​​to ensure the burner flame length is within a reasonable range, thus improving dehydration efficiency and preventing high pressure buildup inside the drying drum. The discharge belt of the drying drum is equipped with a moisture detection system that monitors the moisture content of the dried material in real time and adjusts the inlet air temperature and the residence time of the material in the drying drum (drying time 30-60 minutes) accordingly, ultimately controlling the moisture content of the dried material to be <20%.

[0061] Stirring device, granulation device, dealkali solid-phase reaction device, reduction solid-phase reaction device: First, start the main exhaust fan and dust collector at the tail end. The reactor head and tail hoods are equipped with pressure monitoring points. The main exhaust fan is set with the appropriate frequency through the host computer according to the pipeline balance to meet the process requirements. Then, start the motor of the reduction solid phase reactor. The starting frequency is set to 2 r / min. Temperature monitoring points are evenly distributed in the reduction solid phase reactor. Infrared thermal imaging system devices are installed at the head and tail hoods to synchronously and in real time detect the temperature of each section in the reduction solid phase reactor. By combining comparison and detection, real-time control is achieved, which is convenient for production needs and prevents the formation of rings inside the reactor. Before ignition, the burner is purged with nitrogen according to the burner's process requirements. After the system process checks meet the ignition requirements, the ignition valve and gas valve automatically open, and the pilot flame ignites. The flame detection signal indicates that the first step of ignition is normal. After the normal start-up time, the pilot flame automatically ignites. The gas valve is automatically adjusted according to the set temperature. The flame size is adjusted appropriately based on the combustion flame's completeness. The flame length is adjusted according to the operating characteristics of the reduction solid phase reactor. Once the appropriate temperature is reached, the feed temperature is reached, and production begins. The gas flow rate, air flow rate, and flue gas composition are centrally collected and used in real-time control to record the air-fuel ratio, ensuring a zero-oxygen reduction atmosphere and accurately recording energy consumption. This provides convenient system reports for the overall control system. Record and improve metal recovery rate; when the on-site combustible gas concentration reaches the set value, automatically alarm and automatically shut off the gas valve, switch to the purging process, and stop after purging for five minutes to avoid the accumulation of gas in the reduction solid phase reactor and cause danger; combustible gas concentration alarm signal detection and collection are collected to the overall monitoring system for control, and the corresponding indoor HVAC system exhaust device is activated in conjunction; an endoscopic high-temperature resistant infrared thermal imaging camera is installed at the head and tail cover of the reduction solid phase reactor. The lens is installed in a retractable metal protective cover. The high-temperature resistant infrared thermal imaging camera lens is directly extended into the reduction solid phase reactor through the telescopic device, while the infrared thermal imaging core stays outside the reduction solid phase reactor to realize continuous real-time monitoring of the operating status inside the reduction solid phase reactor. Compressed cooling air cools the protective cover, keeping the infrared lens within its permissible operating temperature range. Simultaneously, it purges the lens to prevent dust from adhering to the lens's protective window. The system incorporates a high-temperature protection circuit; if the cooling gas circulation becomes abnormal, the lens retracts to prevent damage from high temperatures. The endoscopic ultra-high temperature infrared thermal imaging temperature monitoring device is connected to the control box via high-temperature resistant cables and installed in a suitable, easily accessible location, providing power and data transmission for the infrared thermal imager. The electrical control box is connected to the intelligent algorithm server in the central control room via cables, transmitting data to the server. This enables remote real-time image monitoring, temperature measurement display, and centralized remote control of the infrared thermal imager through infrared temperature measurement software.The reducing solid-phase reactor requires 10-12 hours of heating time. During this period, the equipment should be started sequentially from back to front, and stopped in reverse order. During this time, the raw materials need to be batched according to the formula requirements. Material 1 and Material 2 are weighed, and Material 3 and red mud are batched according to the set weights, using a mass ratio of red mud (dry basis): coal powder: bentonite = 100:25:8. This means that 25t of dry red mud, 6.25t of coal powder, and 2t of bentonite are added per hour, for a total dry basis of 33.25t of mixed raw materials. Wet red mud is fed into the drying receiving hopper via a trough feeder, and metered by a buffer metering hopper + reduction meter (accuracy ±0.5%). Coal powder is metered via a reducing agent powder silo, trough feeder + tank metering (accuracy ±0.3%). Bentonite is metered via an additive silo + tank metering (accuracy ±0.3%). After the mixer receives the unloading signals from each material scale, the mixing time begins to run. Upon completion of the timer, the unloading gate automatically opens and closes automatically upon receiving the unloading completion signal, thus executing the next cycle. Materials are conveyed and distributed into a buffer hopper for storage. The mixed material is selected by a storage level gauge and sent to the appropriate storage hopper as reserve material for granulation. Material (34.99 t / h) is fed into the granulator to produce 4-10 mm spherical particles. The particle size is adjusted by conveying and regulating the feed rate and setting the granulator speed to ensure uniform particle size control between 4-10 mm. A particle size analyzer provides real-time feedback on particle size and automatically adjusts the granulator speed, forming a quality feedforward closed loop to ensure green pellet strength and uniform particle size. Qualified particles (36.74 t / h) are fed into a drying solid-phase reactor (effective volume 400 m³). 3The material is reacted at 250-300℃ for 1.5 hours (raw material drying), utilizing the waste heat from the roasting flue gas as the drying heat source. The hot air flow rate is controlled by adjusting valves to achieve optimized energy utilization across different processes. After drying, the moisture content of the pellets is reduced to below 2%. The dried pellets are then screened to remove fine powder <3mm (0.4t / h, returned to raw material pretreatment), resulting in a qualified particle quantity of 36.34t / h (34.99 + 1.75 - 0.4), with a qualification rate of 97.8%. The qualified pellets are then fed into the dealkali solid-phase reactor. By adjusting the speed, outlet gas composition, and temperature, the degree of dealkali removal is determined, and the inlet temperature is adjusted accordingly. The inlet and outlet temperatures, outlet pressure, and gas detectors are all displayed on the centralized system control panel, playing a corresponding role in regulating the hot air balance of the entire pipeline. Based on an inlet temperature of 900±20 degrees and an outlet temperature of 800±20 degrees, the material pellets are held in the solid phase reactor for 2 hours, and the alkali content in the dealkali red mud is controlled to be less than 2%. If the preheating of the solid phase reactor cannot meet the temperature requirements, the hot air burner is turned on for supplemental heating. Temperature detection and control are achieved through a wireless temperature measurement system, transmitting temperature signals to the overall control system. The combustion temperature is controlled between 900±20 degrees Celsius. As the exhaust gas temperature increases, the internal temperature of the dealkali solid-phase reactor also rises accordingly to reach the required temperature. The burner automatically adjusts to maintain the required temperature while reducing energy consumption. After dealkali removal, the internal temperature of the feed pellets in the dealkali solid-phase reactor also changes. The material after preliminary dealkali removal and reduction is sent to the reduction solid-phase reactor, where the reaction temperature is controlled at 1000-1350℃ (±20℃) and the reaction time is 1-2 hours. This process thoroughly reduces the material and magnetizes the iron, initially converting the weakly magnetic hematite (Fe2O3) in the red mud into strongly magnetic magnetite (Fe3O4) or maghematite (γ-Fe2O3) before further reducing it to elemental Fe, preparing for subsequent iron extraction.

[0062] The reduction solid-phase reactor brings the feed pellets to a near-molten state through high temperature, causing combustion. The internal temperature of the reactor rises, and temperature detection and control are achieved via a wireless temperature measurement system, transmitting the temperature signal to a centralized control system. The combustion temperature is controlled between 1100-1350℃ (±20℃). The burner in the reduction solid-phase reactor automatically adjusts the air-fuel ratio based on temperature. Based on oxygen content signals, multi-point temperature signal feedback, and infrared imager detection and comparison, the reactor maintains a zero-oxygen atmosphere, precisely maintaining a reducing atmosphere for efficient roasting and improved metal yield. This zero-oxygen reducing atmosphere enhances metal recovery while simultaneously reducing energy consumption, and the secondary high-temperature process further reduces the alkali content. For example, the gas flow rate and combustion air ratio of the reduction solid-phase reactor, along with data acquisition from the component analyzer, are used in iterative simulation experiments to ensure a zero-oxygen reduction atmosphere and stable temperature curves at various points within the reactor. Temperature data is collected wirelessly via direct contact and combined with an endoscopic infrared imager to maintain the internal temperature control range between 1100-1350℃ (±20℃). The reduction solid-phase reactor is operated at a frequency of 25Hz and a speed of 2r / min. Anaerobic combustion is maintained by adjusting the burner air-fuel ratio to 10:1. The flue gas temperature at the tail end of the reduction reactor enters the dealkali solid-phase reactor at 900±20℃. As the material enters the dealkali solid-phase reactor, the tail end temperature gradually increases rapidly due to the composition of the raw materials. When the material approaches combustion, the control system uses temperature detection at various points and, in conjunction with infrared imager data, adjusts the gas valve based on simulated data within the specified temperature range to ensure improved metal recovery under a reducing atmosphere. Simultaneously, it maintains the hot air inlet temperature control range of the dealkali reactor to ensure effective dealkali removal within the specified temperature range. If the temperature is below the dealkali removal temperature, the air-fuel ratio of the dealkali reactor burner is adjusted to increase heat, while the reactor frequency is reduced to 15Hz to maintain an outlet temperature of approximately 120 degrees Celsius, thus maintaining a low temperature. When the internal temperature of the dealkali reactor exceeds 900 degrees Celsius, the decarbonization supplementary heating burner is deactivated, and the reactor speed is adjusted to maintain operation within the designed time and ensure effective dealkali removal.

[0063] First stage of dealkali pre-reduction: qualified particles are fed into the dealkali solid phase reactor. Under a reducing atmosphere, the reaction temperature is controlled at 850℃ (±20℃) and the reaction time is 2-4h to achieve the initial reduction of iron and titanium oxides and remove alkali metals from the material (Na2O and K2O removal rate ≥90%). Two-stage solid-phase reduction: The material after preliminary dealkali reduction is fed into a reduction solid-phase reactor. The reaction temperature is controlled at 1000-1350℃ (±20℃) and the reaction time is 1-2 hours. The material is thoroughly reduced and the iron is magnetized, so that the weakly magnetic hematite (Fe2O3) in the red mud is initially converted into strongly magnetic magnetite (Fe3O4) or maghematite (γ-Fe2O3) and then further reduced to elemental Fe, which prepares for the subsequent iron extraction.

[0064] Nitrogen cooling and sieving device: The reduced material (34.83 t / h, 36.34-1.51) is fed into the cooler and indirectly cooled to 80°C by the return water for 2 hours. Dust is removed by a cyclone dust collector and a pulse dust collector (0.31 t / h of dust is collected and returned to the raw material pretreatment section). The material flow rate after cooling is 34.52 t / h. The cooler cools the material through its main body and by supplementing nitrogen. The cooled product is then conveyed to the finished product grinding system. The temperature of the reduced material is controlled within 150-200°C after cooling. Pressure and temperature signals are transmitted to the control system for control and monitoring to ensure safe and stable production operation.

[0065] Grinding equipment, iron extraction equipment: The cooled material (34.52 t / h) was crushed to <5 mm and magnetically separated using a 12000 Gs high-gradient magnetic separator, yielding 10.01 t / h of iron concentrate (TFe 67.2%, recovery rate 91.5%, TFe 28.5% in 25 t dry red mud, i.e., 7.125 t / h, 7.125 × 86.5% ≈ 6.16 t of metallic Fe, 6.16 ÷ 67.2% ≈ 10.01 t) and 8.14 t / h of titanium enrichment (TiO2 32.8%, recovery rate 72.3%, TiO2 12.3% in 25 t dry red mud, i.e., 3.075 t / h, 3.075 × 72.3% ≈ 2.22 t). 0.22÷32.8%≈8.14t); Magnetic separation tailings 16.37t / h (34.52-10.01-8.14), used as negative carbon material base material, the main system of the grinding mill is started, and the negative pressure value at the mill inlet is maintained at about -1000Pa and the negative pressure value at the outlet is -5000Pa by adjusting the fan frequency to 25HZ; to ensure smooth material discharge, the dust collection equipment is turned on to prevent dust from escaping; the reduced material enters the mill for grinding, and the flow rate of the material entering the mill is stably controlled by the variable frequency feeder according to the set output; the temperature of the air entering the mill is controlled by adjusting the cold air valve, and the temperature of the air entering the mill is kept below 150-200℃ to ensure that the material is dry without overheating. Under a fixed feed rate, the current of the main motor of the mill and the mill noise are collected by the measuring instrument. The goal is to make the mill operate near the inflection point of the "power-load" curve (i.e., the lowest point of unit power consumption) while ensuring the product particle size. Typically, the grinding noise meter is maintained at a certain decibel value (e.g., 70-80 dB) or within a certain range to control the particle size after grinding to within 200 mesh. The ground iron slag is instantly mixed with process water in a mixing tank to prepare a slurry with a suitable concentration (solid content). The mixing tank agitator is turned on, and the process water inlet valve is opened. The valve is adjusted by the inlet flow meter, and the inlet flow rate is adjusted in real time based on the feedback from the concentration meter in the mixing tank to regulate the concentration. Concentration is a key parameter, usually controlled between 30% and 40% (by weight). A slurry delivery pump is used to deliver the prepared slurry to the inlet of the hydrocyclone separator at a pressure and flow rate of 0.3 MPa. The inlet and outlet valves of the slurry delivery pump are opened to ensure that the inlet pipe is full of slurry. The frequency of the delivery pump is adjusted by the signal from the pipeline pressure transmitter. The high-pressure slurry is injected at high speed tangentially into the hydrocyclone column section through the feed pipe, forming a strong rotating vortex in the cylindrical cavity. Under the powerful centrifugal force, the solid particles in the slurry undergo classification: coarse and heavy particles are subjected to greater centrifugal force and are thrown towards the inner wall of the hydrocyclone cone section. Under the combined action of gravity, the downward eddy current of the liquid, and its own inertia, they spiral downwards along the inner wall and are finally discharged from the bottom outlet (sand outlet) as a high-concentration, coarse-particle "underflow". This is then transported back to the ball mill for regrinding by the slurry pump, forming a closed-loop "grinding-classification" cycle to ensure the full liberation of the target minerals.The pump inlet and outlet valves ensure the inlet pipe is filled with slurry. The slurry pump uses frequency conversion control, with pressure signals before and after the pump. The frequency converter sets the corresponding frequency based on the pressure signals. Pressure signals are installed at key nodes of the hydrocyclone and pipeline to monitor the equipment's operation in real time. Fine and light particles experience less centrifugal force and mainly remain in the central area of ​​the hydrocyclone. Carried upwards by the air column and upward vortex formed at the center, they move upwards and are discharged from the top overflow pipe as a low-concentration, fine-particle "overflow," which is then pumped to the next separation process (refining and slag-reducing magnetic separator and high-gradient magnetic separator) to effectively recover iron minerals from the fine particles. The pump inlet and outlet valves ensure the inlet pipe is filled with slurry. The slurry pump uses frequency conversion control, with pressure signals before and after the pump. The frequency converter sets the corresponding frequency based on the pressure signals. Pressure signals are installed at key nodes of the pipeline to monitor the equipment's operation in real time. Magnetic separation system: Start the downstream processing equipment, the high-gradient magnetic separator and the slurry transfer pump. Start the main motor of the magnetic separator (driving the drum rotation). Connect the magnetic system power supply to establish the working magnetic field. Open the equipment flushing water valve (to help unload concentrate and keep the tank clean). Finally, start the upstream feeding equipment slurry delivery pump to begin feeding. Adjust the frequency of the slurry delivery pump via the pipeline pressure signal. The interlock signal is to open the pipeline valve first, then start the slurry delivery pump, ensuring the pump inlet is full of slurry. The equipment shutdown sequence follows the principle of starting and stopping in the first-to-last order. The starting sequence and interlocking relationship of the finer and slag-reducing magnetic separator and the high-gradient magnetic separator are the same as above.

[0066] Flue gas treatment system: Flue gas treatment and CO2 resource utilization: Flue gas from dryers and reactors (7500m³) 3 / h) passing through a baghouse dust collector (dust concentration 8.2mg / m³) 3 → Limestone-gypsum desulfurization (SO2 removal rate 96.1%, gypsum production 0.015 t / h) → SCR denitrification (NO) X Removal rate 91.5%) → Amine CO2 capture (capture efficiency 91.2%, CO2 captured approximately 1026m³) 3 / h, 1125×91.2%; The captured CO2 is fed into a methanation reactor, where it reacts with hydrogen (CO2 to H2 molar ratio 1:4, requiring approximately 4104 m³ of H2) under the conditions of Ni-based supported catalyst (active component Ni content 18%, support γ-Al2O3), 280℃, and 1.5MPa. 3 The reaction ( / h) produces methane, with a CO2 conversion efficiency of 88.5%, yielding approximately 910 methane. 3 / h (1026×88.5%), the methane purity reaches 99.2% after purification; the purified methane is recycled back to the kiln head through a dedicated pipeline as auxiliary fuel for the reduction section (recycling rate 95%, i.e., 864.5m³). 3The coal powder consumption was reduced by 12.3% (originally 4t / h of purchased coal powder was consumed, 0.5t / h was replaced, 0.5÷4×100%≈12.5%, the error of 12.3% is within a reasonable range).

[0067] The present invention may also have other embodiments. Other technical solutions formed in the claims will not be described in detail. The present invention is not limited to the above embodiments. Equivalent variations and component substitutions based on the above embodiments are all within the protection scope of the present invention.

Claims

1. A process for dealkalizing and deeply reducing red mud through cascade temperature control, characterized in that, Includes the following steps: 1) Precise measurement of raw materials and grinding process: Using pretreated red mud as the base material, a reducing agent with a carbon content of ≥70% is fixed, and a composite additive is selected as the additive. The mass ratio of red mud, reducing agent and composite additive is 100:20-30:5-10; the raw materials with the above mass ratio are ground into powder. 2) The ground raw materials are fed into a red mud mixer and stirred using a gradient speed to ensure that the material mixing variation coefficient is ≤8%; 3) Feed the mixed raw materials into a granulator to granulate them into granules with a particle size of 5-15mm. 4) The granular material is fed into the dealkali solid-phase reaction device. Under a reducing atmosphere, the reaction temperature is controlled at 850℃±20℃ and the reaction time is 2-4h to remove alkali metals from the material and form solid-phase granular material. 5) The particulate material after the solid-phase reaction is fed into a cooler and cooled by direct nitrogen circulation. The cooling temperature is controlled to gradually decrease from 900℃ to ≤150℃, and the cooling time is 1.5-3h; then the furnace is cooled slowly. 6) After cooling, the material is crushed by a crusher to a particle size of less than 5mm. The crushed finished material enters the grinding section to be ground to 200-400 mesh. 7) The powder after grinding is separated by a high gradient magnetic separator.

2. The process for dealkalization and deep reduction of red mud by staged temperature control as described in claim 1, characterized in that: In step 1), the reducing agent is coke powder with a fixed carbon content of 75%, and the composite additive is bentonite with a bonding strength of 0.6-0.8 MPa. The mass ratio of red mud, coke powder and bentonite is 100:25-30:6-8.

3. The process for dealkalization and deep reduction of red mud by staged temperature control as described in claim 1, characterized in that: In step 2), a vertical turbulent mixer is used, with the mixing shaft equipped with 3-5 layers of blades and the blade angle adjustable from 30 to 60°. First, the mixture is stirred at a low speed of 200-300 r / min for 3-5 minutes, and then stirred at a high speed of 300-500 r / min for 8-10 minutes to ensure that the material mixing variation coefficient is ≤8%. The moisture content is adjusted to 14-16% by adding some external water.

4. The process for dealkalization and deep reduction of red mud by staged temperature control as described in claim 1, characterized in that: In step 3), after granulation by the granulator, the prepared 5-15mm granules are sent to a drum dryer with an internal temperature of 200-350℃ for drying pretreatment, and the moisture content is dried from <16% to <2%.

5. The process for dealkalization and deep reduction of red mud by staged temperature control as described in claim 1, characterized in that: In step 4), an online anti-ring control method is used to automatically control the solid-phase reactor. The temperature detection unit of the multi-source sensing module collects the temperature signals of the inside and outside of the solid-phase reactor in real time. The process parameter acquisition unit collects the current and torque signals of the main drive motor and the flue gas atmosphere composition signal. The visual image detection unit collects the material flow state inside the solid-phase reactor and the visual image of the wall. The edge computing and feature extraction module filters, reduces noise, and standardizes the signals collected by the multi-source sensing module, and extracts feature values ​​related to ring formation. The feature values ​​include axial temperature gradient, radial temperature deviation, area and movement trend of high temperature region, fluctuation variance of drive torque, current trend slope, residence time in the preset temperature range, texture roughness change of the inner wall of the solid-phase reactor, and pixel ratio of physical adhesion area. The ring formation risk assessment and decision-making module receives data from the edge computing and feature extraction modules. The risk fusion model uses a multivariate fusion algorithm based on weight allocation to calculate the ring formation risk index (CRI). The CRI is a continuous value between 0 and 100, with higher values ​​indicating greater ring formation risk. The control decision generation unit generates targeted control commands based on the CRI and the feature value with the greatest contribution. The execution intervention module performs continuous monitoring or early warning based on the control commands, or implements intervention measures such as adjusting material ratio and particle size through the material composition and particle size adjustment unit, or adjusting operating temperature, atmosphere, or main drive motor speed through the operating parameter adjustment unit. The analysis module evaluates the effectiveness of intervention measures, and the storage module records successful cases for optimizing the parameters of the risk fusion model and the control instructions of the control decision generation unit.

6. The process for dealkalization and deep reduction of red mud by staged temperature control as described in claim 5, characterized in that: The clustering risk index employs a multivariate fusion algorithm based on weight allocation, and its expression is as follows: CRI=100 ×F fusion ([IN t ×F t ,IN v ×F v ,IN p ×F p ];Θ) in: F fusion (·): Top-level fusion function, which combines the weighted feature values ​​of each dimension into a single risk scalar; F t ,F v ,F p : These represent the normalized feature values ​​of the three dimensions: temperature field, visual image, and process parameters, respectively, with each value ranging from [0,1]. W t W v W p : These correspond to the dynamic weight coefficients of each dimension, satisfying W t +W v +W p =1; Θ: The set of internal parameters of the fusion function; Multiplier 100: Maps the results to an intuitive risk index range of 0-100; Temperature field characteristic value F t The calculation method is as follows: F t =min(1,(ω1*N(ΔT)+ω2*N(G)+ω3*N(τ)) / (ω1+ω2+ω3)) in: N(ΔT) = min(1, (ΔT - ΔT) th ) / ΔT scale ): Local hotspot temperature difference anomaly, ΔT is the temperature difference between the current hotspot and the regional average temperature, ΔT th The threshold for initiating interest, ΔT scale This is the scaling factor; N(G)=|G actual -G baseline | / G range : Axial temperature gradient anomaly index, G actual To calculate gradients in real time, G baseline Based on historical normal benchmarks, G range This is within the normal fluctuation range; N(τ)=(τ measured -τ min ) / (τ max -τ min ): External wall temperature conduction delay rate, τ measured τ represents the measured delay time. min and τ max For the minimum and maximum theoretical delays; ω1, ω2, ω3: Sub-feature weights, corresponding to the weights of local temperature, axial temperature, and outer wall temperature, respectively, preset parameters; Visual image feature value F v The calculation method is as follows: F v =SVM(I t ,I {t-1} ,I {t-2} ,...) A trained machine learning model was used to analyze a series of endoscopic images. The model output the probability of determining the "loop" state, which was then compressed to [0,1] using the Sigmoid function and used as F. v The value directly reflects the strength of visual evidence of wall adhesion, and the machine learning model is a support vector machine (SVM). Process parameter characteristic value F p The calculation method is as follows: F p =max(N(σ T ), N(φ), N(Δt)) in: N(σ T )=σ Tcurrent / σ Tnormal The ratio of the variance in drive torque fluctuation reflects changes in mechanical load. N(φ) = |φ actual -φ optimal | / φ tolerance φ: Deviation of reduction potential reflects an abnormal chemical reaction environment; optimal For optimal setting, φ tolerance Allowable deviation; N(Δt)=(t residence -t setpoint ) / t range : The change in the residence time of the material within the preset temperature range.

Images