An industrial solid waste mixing pile carbonization construction control method based on soil layer identification
Patent Information
- Application Number
- CN202611060739.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-16
- Publication Date
- 2026-10-27
Smart Images
Figure CN122980428A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the fields of soft soil foundation treatment technology in geotechnical engineering, resource utilization technology of industrial solid waste, and carbon capture, utilization and storage technology, and in particular to a method for controlling the carbonization construction of industrial solid waste mixing piles based on soil layer identification. Background Technology
[0002] Cement mixing piles are one of the most widely used technologies in soft soil foundation treatment. They involve forcibly mixing cement slurry with soft soil, utilizing the cement hydration and hardening reaction to improve the bearing capacity of the foundation and control settlement. However, traditional cement mixing pile technology faces increasingly severe challenges. On the one hand, cement production is a typical high-energy-consuming and high-emission industry, with a carbon emission intensity of approximately 0.85 tons per ton of cement clinker produced. In mixing pile projects, the cement content is typically as high as 15%-25%, resulting in considerable cement consumption and implicit carbon emissions per pile. Finding low-carbon alternatives to cement has become a critical issue urgently needing breakthroughs in the field of geotechnical engineering. On the other hand, the large amounts of industrial solid waste generated annually, such as converter smelting slag, refining slag, and fly ash, not only occupy land resources but also pose a potential pollution risk to the surrounding water and soil environment. How to efficiently utilize these solid wastes for soft soil foundation reinforcement and achieve a circular economy model of "treating waste with waste" has significant environmental value and engineering significance.
[0003] In recent years, research on the preparation of low-carbon mixing pile solidifying agents using industrial solid waste has made some progress. Existing technologies have attempted to replace part of the cement with converter steel slag, blast furnace slag, fly ash, etc., alone or in combination, utilizing their potential cementitious activity to achieve soil solidification. However, existing solid waste-based mixing pile technologies mainly suffer from the following shortcomings: First, most schemes still use cement as the main cementing material, with solid waste only used as an auxiliary admixture. The cement replacement ratio is limited, and the carbon emission reduction effect is not significant. Second, the early strength development of pure solid waste systems is slow under normal temperature conditions, making it difficult to meet the early bearing capacity requirements of engineering projects. Third, existing schemes mostly rely on single solid waste or simple binary blends, lacking a precise matching design for the phase composition and activity differences of different solid waste components. This results in insufficient synergy in the cementitious reaction, large fluctuations in the solidification effect, and difficulty in ensuring the stability and reliability of engineering quality.
[0004] To achieve truly clinker-free, low-carbon solidification, some studies have attempted to introduce CO2 carbonation reaction to assist in the enhancement of solid waste-based solidifiers. Existing carbonation mixing pile technology injects CO2 into the pile body during mixing, causing a carbonation reaction in the solidifier and calcium and magnesium components in the soil to generate carbonate precipitates, which fill pores and increase strength. However, existing carbonation mixing pile construction technologies still have significant drawbacks: First, CO2 injection parameters often use fixed process values, failing to consider the impact of key differences in soil permeability, moisture content, and mineral composition on carbonation reaction efficiency, leading to significant variations in carbonation effects with geological conditions. Second, there is a lack of effective monitoring methods for the carbonation reaction process during construction, making it impossible to know the amount and uniformity of carbonate formation and distribution within the pile body in real time, hindering closed-loop quality control of the carbonation process. Third, existing methods have coarse control over CO2 injection pressure, resulting in severe gas loss in high-permeability soil layers, while high-pressure injection in low-permeability soil layers easily triggers soil fracturing, making it difficult to balance carbonation efficiency and construction safety.
[0005] In terms of soil layer identification and intelligent construction, existing mixing pile equipment mainly relies on geological survey reports before construction to determine construction parameters, and its responsiveness to changes in soil properties during construction is limited. A few existing technologies attempt to collect single signals such as soil resistivity or penetration resistance by installing sensors on the drill bit for soil layer inversion, but the types of sensors used are limited and the signal dimensions are low, making it difficult to comprehensively characterize the key soil layer properties affecting carbonation. More importantly, existing methods rely on empirical statistical thresholds for soil layer identification, lacking physical understanding of the intrinsic mechanism of carbonation, and their accuracy decreases when encountering unfamiliar strata combinations. Furthermore, the data collected by existing methods cannot be used to optimize subsequent construction; the system lacks self-learning and self-evolution capabilities, making it difficult to continuously improve carbonation control accuracy under heterogeneous strata conditions. Summary of the Invention
[0006] To address the shortcomings of existing mixing pile carbonization technology in terms of engineering applicability without clinker curing agents, multi-dimensional and accurate perception of soil layer characteristics, dynamic matching of CO2 injection and soil layer properties, and physical interpretability and adaptability of control models, this application provides a construction control method for industrial solid waste mixing pile carbonization based on soil layer identification.
[0007] The carbonization construction control method for industrial solid waste mixing piles based on soil layer identification provided in this application adopts the following technical solution: A method for controlling the carbonization construction of industrial solid waste mixing piles based on soil layer identification includes the following steps: A clinker-free curing agent is injected into the soil layer, and the physical and spectral parameters of the target soil layer are collected in real time through a multi-source sensor array integrated at the drill bit to construct a multi-dimensional feature vector of the current soil layer. The multidimensional feature vector is input into a pre-constructed classification-regression hybrid random forest model. The classification branch of the model outputs the soil layer type identifier; the regression branch of the model uses the convection-diffusion partial differential equation of the embedded physical core model as the boundary constraint condition, and outputs the carbonation reaction potential index and the initial maximum carbon sequestration amount of the current soil layer; the regression branch of the model also corrects the initial maximum carbon sequestration amount based on the theoretical carbon sequestration upper limit calculated by the physical core model, and outputs the corrected maximum carbon sequestration prediction value. The central control system dynamically generates differentiated CO2 injection strategies based on soil type identification and carbonation reaction potential index, and controls the constant-pressure variable-pressure gas supply system to execute CO2 injection according to the differentiated CO2 injection strategies. The differentiated CO2 injection strategies include differentiated adjustment of CO2 injection volume, injection rate and injection pressure.
[0008] Furthermore, the multidimensional feature vector includes: the stirring torque obtained by the torque sensor, the drilling speed obtained by the rotation speed sensor, the soil conductivity obtained by the current-voltage four-terminal electrode, the ultrasonic pulse velocity obtained by the ultrasonic testing module, the near-infrared spectral intensity obtained by the near-infrared spectral sensor, the soil moisture and temperature obtained by the soil temperature and humidity sensor, and the soil pressure obtained by the piezoelectric crystal matrix earth pressure sensor.
[0009] Furthermore, the constraint embedding process of the physical core shrinkage model includes: the regression branch outputs the initial maximum carbon sequestration based on the input multidimensional feature vector, and at the same time calculates the key parameters in the physical core shrinkage model based on the real-time data obtained by the multi-source sensor array, and numerically solves the convection-diffusion equation of the physical core shrinkage model to calculate the theoretical upper limit of carbon sequestration that conforms to the law of conservation of mass under the current working conditions. The initial maximum carbon sequestration is corrected through a post-processing interception mechanism: if the initial maximum carbon sequestration does not exceed the theoretical upper limit of carbon sequestration, the initial maximum carbon sequestration is directly output as the predicted value of the maximum carbon sequestration; if the initial maximum carbon sequestration exceeds the theoretical upper limit of carbon sequestration, the excess is reduced by a predetermined ratio and then output as the predicted value of the maximum carbon sequestration.
[0010] Furthermore, the effective diffusion coefficient and chemical reaction rate constant in the physical core shrinkage model are calculated and dynamically updated in real time through a sensor array. The effective diffusion coefficient is determined by the real-time inverted porosity, real-time temperature, and activation energy calibrated by molecular dynamics simulation. The chemical reaction rate constant is determined by the real-time conductivity and the calcium carbonate formation reaction progress inverted by near-infrared spectroscopy.
[0011] Furthermore, the pre-built classification-regression hybrid random forest model is established using the following cross-scale physical model initialization method: Collect macroscopic geological survey data and laboratory autoclave control test data; Molecular dynamics simulations were performed to extract microscopic parameters, including the reaction interface binding energy, key ion diffusion coefficient, and CO2 dissociation carbonization unit reaction rate constant. The microscopic parameters are mapped and transformed into the dynamic boundary conditions and effective diffusion coefficient benchmark values of the macroscopic convection-diffusion equation in the physical core shrinkage model. A high-dimensional matrix database containing the correlation between ultrasonic pulse velocity, near-infrared spectrum, reaction interface binding energy, carbonate content and compressive strength is constructed as the training basis for the model.
[0012] Furthermore, the differentiated CO2 injection strategy specifically includes: When the soil layer type is determined to be a sandy soil layer with high porosity, high permeability and high reaction potential, the gas supply system is controlled to execute a staged foaming high-pressure injection mode, injecting CO2 at a first rate before the initial setting period, and increasing to a second rate after the initial setting period. When the soil layer is determined to be a cohesive soil layer with low porosity, low permeability, and low reaction potential, the gas supply system is controlled to execute a low-pressure slow-release and concentration-compensated injection mode to inject CO2 at a constant low-pressure rate.
[0013] Furthermore, it also includes closed-loop self-learning control steps: During and after the construction of a single mixing pile, the actual carbon fixation amount inverted from the actual near-infrared calcium carbonate formation characteristic peak fed back by the drill bit is compared with the maximum carbon fixation amount predicted by the model to calculate the residual. Representative samples generated during the current construction are extracted as new data to perform periodic incremental training on the model; and the feature importance weights are dynamically updated based on the new data to achieve adaptive evolution under heterogeneous geological conditions.
[0014] Furthermore, the carbonization reaction potential index includes the predicted value of CO2 absorption efficiency, the intensity enhancement potential level, and the sensitivity index of reaction duration.
[0015] Furthermore, the raw materials for preparing the clinker-free curing agent include converter smelting slag, RH furnace refining slag, fly ash, magnesium olivine, and alkali activator.
[0016] This application also provides a soil mixing pile device for dynamically regulating carbon partial pressure in identifiable soil layers to implement the above method, comprising: The drilling system includes an outer drill rod and an inner drill rod that are coaxially mounted. The outer drill rod and the inner drill rod are driven by independent servo motors and rotate in opposite directions. The outer and inner drill rods are respectively provided with stirring blades on their outer peripheries for bidirectional shearing and stirring of the soil and the solidifier without clinker. The data acquisition system includes a multi-source sensor array and a data transmission channel integrated at the drill bit. The multi-source sensor array is used to synchronously acquire multi-source parameters of the soil layer during the mixing and carbonization process. The multi-source parameters of the soil layer are transmitted back to the surface control platform in real time via the data transmission channel. The material injection system includes a material storage module, a material transmission channel, and a material outlet located at the drill bit, for injecting the clinker-free curing agent into the soil layer; The gas injection system, including a constant and variable pressure gas supply system, a gas transmission pipeline, and a gas outlet located at the bottom of the drill bit, is used to inject CO2 into the soil layer. The central control system is used to receive multi-source soil parameters transmitted back by the data acquisition system, identify the current soil type and predict the carbonation reaction potential index through the built-in prediction model based on physical constraint machine learning, dynamically generate a differentiated CO2 injection strategy, and control the constant pressure gas supply system to adjust the amount, rate and pressure of CO2 injected into the soil layer.
[0017] In summary, this application includes at least one of the following beneficial technical effects: 1. This application constructs a complete closed-loop control system of "real-time perception by multi-source sensors - dynamic discrimination by physical constraint machine learning - adaptive execution by differentiated injection strategy". It embeds the convection-diffusion partial differential equation constraints of the classification-regression hybrid random forest model and the physical core shrinkage model into the carbonization construction control of the mixing pile. This overcomes the defect of prediction distortion in complex strata of pure data-driven models, and makes CO2 diffusion prediction strictly conform to the mass conservation and dynamic laws, thereby improving the physical interpretability and engineering reliability of carbonization control.
[0018] 2. By integrating a multi-source sensor array into the drill bit, especially the "current-voltage four-terminal method" conductivity measurement electrode with an insulating hard shell and a near-infrared spectral sensor, high-frequency in-situ synchronous acquisition of soil conductivity, pore water pressure, temperature and humidity, and characteristic peaks of calcium carbonate formation is achieved. This eliminates the contact resistance error caused by the metal mud pack on the drill bit in traditional measurement methods, and provides a data foundation for accurate identification of soil types and real-time tracking of carbonation reaction processes.
[0019] 3. Based on the soil layer identification results and carbonation reaction potential index, this application implements differentiated CO2 injection strategies for different soil types: high-reaction-potential sandy soil layers adopt staged foamed high-pressure injection to maximize carbon sequestration, while low-reaction-potential cohesive soil layers adopt low-pressure slow release and concentration compensation injection to avoid soil splitting failure. This achieves optimal matching between CO2 injection parameters and soil properties, improving carbonation efficiency while ensuring construction safety.
[0020] 4. This application adopts a coaxial outer drill rod and inner drill rod structure with opposite rotation directions. The bidirectional mixing blades perform strong shearing and mixing of soil and curing agent, which significantly improves the dispersion uniformity of clinker-free curing agent in soil layer. At the same time, a multi-source sensor array is integrated into the drill bit to realize an integrated operation mode of "mixing, sensing and carbonizing at the same time", which improves construction efficiency and carbonization reaction uniformity.
[0021] 5. By establishing a cross-scale physical model initialization method from molecular dynamics simulation to macroscopic core shrinkage model, microscopic parameters such as microscopic interface binding energy, ion diffusion coefficient and elementary reaction rate constant are mapped into the dynamic boundary conditions of macroscopic convection-diffusion equations, providing physical support for the carbonization potential index and overcoming the deficiency of traditional empirical models in the lack of generalization ability in unseen strata.
[0022] 6. By introducing a closed-loop self-learning mechanism based on incremental training, the system automatically extracts representative samples after the construction of each pile to perform periodic incremental training on the random forest model, and dynamically updates the feature importance weights according to working conditions such as groundwater abundance, so that the system has the adaptive evolution capability of "the more you drill, the more accurate it becomes", and can continuously improve the carbonization control accuracy and engineering reliability under heterogeneous strata conditions.
[0023] 7. This application utilizes a variety of industrial solid wastes, such as converter smelting slag, RH furnace refining slag, and fly ash, in conjunction with the natural mineral magnesium olivine to prepare a clinker-free curing agent. This eliminates the need for traditional silicate cement, enabling large-scale resource utilization of industrial solid waste and reducing the implicit carbon emissions of mixing pile projects. Under the dual drive of alkali activation and CO2 carbonization, the various solid waste components form a calcium-magnesium-iron cross-linked mineralization structure, which ensures the mechanical properties of the pile while achieving permanent chemical sequestration of CO2. This approach combines the triple benefits of solid waste disposal, low-carbon emission reduction, and soil reinforcement, and has good environmental and economic value. Attached Figure Description
[0024] Figure 1 This is a schematic diagram of the overall structure of a mixing pile device according to Embodiment 2 of this application; Figure 2 This is a schematic diagram of the current-voltage four-terminal electrode structure in Embodiment 2 of this application; Figure 3 This is a flowchart of the mapping method from micro to macro when constructing a high-dimensional matrix database in Embodiment 3 of this application.
[0025] Reference numerals in the attached diagram: 1. Support frame; 2. Pulley block; 3. Data acquisition center; 4. Data transmission channel; 5. Material transmission channel; 6. Material temporary storage module; 7. External drill rod servo motor; 8. External drill rod gearbox; 9. External drill rod drive gear; 10. External drill rod; 11. Internal drill rod; 12. External drill rod stirring blade; 13. Material outlet one; 14. Internal drill rod stirring blade; 15. Multi-source sensor array; 16. Gas outlet; 17. Material outlet two; 18. Internal drill rod drive gear; 19. Internal drill rod gearbox; 20. Internal drill rod servo motor; 21. Gas transmission pipeline; 22. Gas control module. Detailed Implementation
[0026] The following is in conjunction with the appendix Figure 1-3 This application will be described in further detail.
[0027] Example 1 This embodiment provides a clinker-free curing agent, whose raw materials consist of industrial solid waste and natural minerals, and does not contain silicate cement clinker. The industrial solid waste includes converter smelting slag, RH furnace refining slag, and fly ash, and the natural mineral is magnesium olivine.
[0028] The above raw materials are pretreated as follows: Pretreatment of converter smelting slag: The converter smelting slag is crushed and ground using a vibrating mill. The grinding time is controlled at 20-40 minutes to ensure that the converter smelting slag powder completely passes through a 100-mesh sieve, ensuring that its fineness meets the activity requirements of the subsequent mixing reaction.
[0029] Pretreatment of RH furnace refining slag: The same method as that used for converter smelting slag is adopted, which involves crushing and grinding the slag with a vibrating mill until it can completely pass through a 100-mesh sieve.
[0030] Pretreatment of fly ash: Place fly ash in an oven and dry it at 105±5℃ to constant weight to remove free moisture; then grind the dried fly ash in a ball mill for 20-30 minutes to make the particle size uniform; finally, screen it through a 5mm sieve to remove large particle impurities.
[0031] Pretreatment of forsterite: The forsterite ore is coarsely crushed to a particle size of less than 5mm by a jaw crusher, and then ground by a vibrating mill until it completely passes through a 100-mesh sieve.
[0032] Pretreatment of palm leaf fibers (optional): Take palm leaf fibers, cut them into short fiber segments of 5-10 mm in length, immerse them in a saturated Ca(OH)₂ solution for 24 hours, remove them, rinse them with deionized water until neutral, and dry them for later use. Pretreated palm leaf fibers possess a natural porous microtubule structure, enabling them to release water into the matrix based on osmotic pressure difference during the exothermic and water-consuming carbonization stage, maintaining the thickness of the nanoscale liquid film at the interface and driving the continuous carbonization reaction.
[0033] Weigh the pretreated raw materials according to the following mass percentages: converter smelting slag powder 28%-45%, RH furnace refining slag powder 15%-20%, fly ash 17%-30%, and magnesium olivine 20%-33%. As a preferred embodiment, the mass percentages of each raw material are: converter smelting slag powder 35%, RH furnace refining slag powder 18%, fly ash 22%, and magnesium olivine 25%. The sum of the above four raw materials is 100%.
[0034] Weigh out 6%-9% of the total mass of the above raw materials as an alkaline activator. Preferably, the amount of alkaline activator added is 7.5% of the total mass of the raw materials.
[0035] Addition of palm leaf fiber (optional): In some embodiments, pretreated palm leaf fiber at a volume ratio of 1% may also be added.
[0036] Mix the weighed ingredients as follows: First, add the weighed converter slag powder, RH furnace refining slag powder, fly ash, and magnesium olivine to a mixer and dry-mix for 3-5 minutes to ensure the four raw materials are fully and evenly mixed. Then, add water (water-ash ratio controlled at 0.3-0.5) to the mixed powder and add the weighed Na2SiO3·9H2O alkali activator. Continue to stir at low speed for 5-8 minutes to activate polymerization and ensure the alkali activator is evenly distributed in the mixed system.
[0037] After mixing, CO2 gas is introduced into the mixture for preliminary carbonization. The preliminary carbonization process is carried out at 20-25℃, with the CO2 gas flow rate controlled at 0.1-0.3 cubic meters per minute, and the introduction time at 30-60 minutes to ensure sufficient gas penetration and reaction with the raw materials. The mixture after preliminary carbonization is the clinker-free curing agent. At the mixing pile construction site, the above-mentioned clinker-free curing agent is transported to the material outlet at the bottom of the drill rod through a material conveying system and injected into the soil layer to be reinforced.
[0038] The synergistic alkaline activation and carbonization mechanism of industrial solid waste and natural minerals is as follows: High-calcium and high-alumina glassy components of converter smelting slag and RH furnace refining slag undergo network depolymerization in an alkaline environment, generating hydrated calcium aluminosilicate (CASH) gel to provide early strength; excess calcium oxide and magnesium oxide generate micro-nano-scale carbonate precipitates through carbonization, filling the gel voids; siliceous components in fly ash dissolve in an alkaline environment and undergo secondary polymerization, forming a three-dimensional network geopolymer structure; magnesium olivine, utilizing its significantly delayed dissolution kinetics compared to amorphous solid waste phases, releases Mg in the later stages of the reaction. 2+ When combined with CO2, magnesite (magnesium carbonate) crystals with expansion compensation effect are generated in situ in the micropores of the gel, forming a calcium-magnesium-iron interlocking mineralization structure.
[0039] Example 2 This embodiment provides a mixing pile device, such as... Figure 1 As shown, the entire equipment is fixed to the construction platform by the support frame 1, and the vertical lifting of the drill bit is achieved by the pulley block 2. The data acquisition center 3 is connected to the multi-source sensor array 15 at the drill bit via the data transmission channel 4. The material storage module 6 is located at the top of the equipment and is used to store the clinker-free curing agent. It is connected to the material outlet 13 and material outlet 17 at the bottom of the drill rod via the material transmission channel 5. The material outlet 13 and material outlet 17 are symmetrically located near the mixing blade at the bottom of the inner drill rod 11, and are used to uniformly spray the clinker-free curing agent into the soil layer being mixed and cut.
[0040] The constant-pressure variable-pressure gas supply system is connected to the gas outlet 16 located at the bottom of the drill bit via a gas control module 22 and a gas transmission pipeline 21. The system comprises a CO2 reaction generator, a CO2 storage device, a CO2 compression device, a flow meter, and a pressure regulating valve, and is specifically designed for CO2 injection to promote the carbonation reaction. This system can adjust the injection volume and flow rate of CO2 according to different soil characteristics, ensuring a thorough carbonation reaction and improving the strength and durability of the solidified structure. The material transfer system ensures accurate material delivery and proportioning, guaranteeing the consistency of the mixture. The gas outlet 16 is located at the bottom of the drill bit, ensuring that CO2 can permeate upwards from the bottom of the pile body for complete carbonization.
[0041] The drilling tool includes an outer drill rod 10 and an inner drill rod 11 coaxially mounted. The bottom end of the inner drill rod 11 extends out of the outer drill rod 10, and the drill bit is located at the bottom end of the inner drill rod 11. An outer drill rod mixing blade 12 and an inner drill rod mixing blade 14 are respectively installed on the outer periphery of the outer drill rod 10 and the inner drill rod 11. The power mixing system is driven independently by two servo motors. The outer drill rod servo motor 7 drives the outer drill rod 10 to rotate forward via the outer drill rod gearbox 8 and the outer drill rod transmission gear 9; the inner drill rod servo motor 20 drives the inner drill rod 11 to rotate in the opposite direction via the inner drill rod gearbox 19 and the inner drill rod transmission gear 18, thereby causing the outer drill rod mixing blade 12 and the inner drill rod mixing blade 14 to perform bidirectional high-intensity shearing and mixing of the soil and the solidifying agent.
[0042] The multi-source sensor array 15 is integrated into the drill bit, specifically including: (1) a current-voltage four-terminal electrode for measuring the electrical conductivity of the soil layer, such as Figure 2As shown, the electrode is a modified four-terminal electrode with an insulating hard shell, including outer current injection electrodes A and B and inner voltage measurement electrodes C and D; (2) a piezoelectric crystal matrix earth pressure sensor for measuring earth pressure; (3) a soil temperature and humidity sensor for monitoring soil temperature and humidity; (4) an ultrasonic testing module for measuring the ultrasonic pulse velocity of soil layers; and (5) a near-infrared spectral sensor for high-frequency in-situ scanning of mineral composition and evolution in the 1000-2500nm band. The analog signals collected by all sensors are converted into digital signals in real time, transmitted back to the data acquisition center 3 via the data transmission channel 4, and finally summarized and transmitted to the ground central control system.
[0043] In the current-voltage four-terminal measurement system, the shortest current path passes through the surface of the mixing pile. A constant current power supply inputs a stable current I through terminals A and B. Simultaneously, a voltage instrument measures the potential difference V through terminals C and D. Conductivity It can be represented as:
[0044] Among them, L AC L AB L BC L BD These represent the arc lengths between the corresponding ends, in meters (m).
[0045] Example 3 This embodiment provides a construction control method for carbonization of industrial solid waste mixing piles based on soil layer identification. It utilizes the clinker-free curing agent provided in Embodiment 1 and the mixing pile equipment provided in Embodiment 2. The construction control method of this embodiment includes the following steps: Step 1: Inject the clinker-free curing agent into the soil layer. Real-time acquisition of the physical and spectral parameters of the target soil layer is achieved using a multi-source sensor array integrated at the drill bit. This constructs a multi-dimensional feature vector of the current soil layer, specifically including: The equipment is started. The servo motor of the outer drill rod drives the outer drill rod to rotate forward, while the servo motor of the inner drill rod drives the inner drill rod to rotate in the opposite direction. The mixing blades of the outer and inner drill rods pre-cut the soil layer in situ. The material conveying system is started, and the clinker-free curing agent is pumped under pressure through the material conveying channel to the bottom of the drill rod, and then sprayed into the soil layer being mixed and cut from the material outlet. The bidirectional counter-rotating mixing blades strongly shear and mix the soil and curing agent.
[0046] During drilling and mixing, a group of sensors at the drill bit are activated synchronously to collect multi-source soil parameters in real time. Among these, the soil conductivity is measured using a modified "current-voltage four-terminal method" with an insulating hard shell: a constant current power supply injects a stable current into the soil layer through the outer electrode, a voltage measuring instrument measures the potential difference through the inner electrode, and the insulating hard shell of the four electrodes eliminates contact resistance errors caused by the metal mud pack on the drill bit. A piezoelectric crystal matrix senses the pore water pressure in the soil, a temperature and humidity sensor monitors soil temperature and humidity, an ultrasonic testing module measures the soil wave velocity, and a near-infrared spectroscopy sensor performs high-frequency in-situ scanning of the soil layer around the drill bit in the 1000-2500 nm band, tracking the characteristic peaks of kaolinite and calcium carbonate in real time. The absorbance of the calcium carbonate characteristic peak output by the near-infrared spectroscopy sensor in real time is normalized to obtain the normalized reaction progress θ of calcium carbonate. NIR This is used for the quantitative characterization of the carbonization reaction process in subsequent steps.
[0047] All analog signals collected by the sensors are converted from analog to digital and then transmitted at high speed back to the data acquisition center via a data transmission channel. The data acquisition center aggregates the signals from each sensor to construct a multidimensional feature vector of the current soil layer. This multidimensional feature vector specifically includes: stirring torque T, drilling speed v, soil resistivity ρ (calculated from conductivity), and wave velocity V. p Near-infrared spectral intensity λ, soil moisture, soil temperature, and soil pressure.
[0048] When the drill bit is in different soil layers, the multidimensional feature vectors collected by the sensor array exhibit different numerical characteristics. These differences provide a data basis for subsequent soil layer type identification.
[0049] Step 2: Input the multidimensional feature vector into the pre-constructed classification-regression hybrid random forest model. The classification branch of the model outputs the soil layer type identifier; the regression branch of the model uses the convection-diffusion partial differential equation of the embedded physical core model as the boundary constraint condition, and outputs the carbonation reaction potential index and the initial maximum carbon sequestration amount of the current soil layer; the regression branch of the model also corrects the initial maximum carbon sequestration amount based on the theoretical carbon sequestration upper limit calculated by the physical core model, and outputs the corrected maximum carbon sequestration prediction value, specifically including: The central control system has a pre-built classification-regression hybrid random forest model with the following settings: 500 decision trees, a maximum depth of 15 layers, and a split threshold of 0.05.
[0050] Before training the model, the cross-scale physics model needs to be initialized: First, macroscopic geological survey data of the target construction area are collected, including soil layer profile distribution, porosity, water content and other physical and mechanical parameters of each soil layer; at the same time, laboratory autoclave control tests are conducted. Under laboratory conditions, a clinker-free curing agent is added according to the construction ratio, and the amount of carbonate generated and the compressive strength of the samples under different carbonation times are measured to establish a macroscopic basic database.
[0051] Then, numerical simulations were performed using the molecular dynamics simulation software LAMMPS. The force field was selected from the ClayFF force field optimized for mineral crystal and geopolymer systems and superimposed with the SPC / E water model to establish a heterogeneous interface supercell model (the total number of atoms in the system is about 25,000) including a crystalline forsterite layer, an amorphous base-excited geopolymer network, and a saturated CO2 porous aqueous solution. The simulation was carried out under ambient temperature (298.15 K) and ambient pressure (0.1 MPa). First, structural relaxation was performed for 0.5 ns in the NPT ensemble to eliminate initial configurational stress and achieve system density convergence. Then, the simulation was switched to the NVT ensemble for 10 ns of equilibrium production simulation (time step 1.0 fs). The microscopic parameters, including the reaction interface binding energy, the critical ion diffusion coefficient, and the CO2 dissociation carbonization unit reaction rate constant, were quantitatively output through the trajectory file.
[0052] Finally, as Figure 3 As shown, the aforementioned microscopic parameters are mapped and transformed into the dynamic boundary conditions and effective diffusion coefficient benchmark values for the macroscopic convection-diffusion equations in the physical core model. Specifically, based on the ion diffusion coefficient and the soil porosity measured in the field, the benchmark value of the effective diffusion coefficient within the macroscopic porous medium of the mixing pile is calculated; the boundary constant for macroscopic compressive strength growth is corrected based on the reaction interface binding energy. Based on the integrated macroscopic features and initial labels, a high-dimensional matrix database containing the correlation between ultrasonic pulse velocity, near-infrared spectrum, reaction interface binding energy, carbonate content, and compressive strength is constructed as the training basis for the random forest model.
[0053] During the model training phase, samples from a high-dimensional matrix database were used as the training set, and supervised learning was employed to train both the classification and regression branches simultaneously. The classification branch was labeled with the unique thermal encoding of the soil layer type, while the regression branch was labeled with maximum carbon sequestration, CO2 absorption efficiency, strength enhancement potential level, and reaction duration sensitivity index. Specifically, the predicted CO2 absorption efficiency represents the ability of a unit mass of curing agent to absorb and fix CO2 under specific soil layer conditions, expressed as the number of kilograms of CO2 absorbed per ton of curing agent; the strength enhancement potential level indicates the grade of the increase in compressive strength of the pile after carbonization curing relative to the untreated soil; and the reaction duration sensitivity index represents the sensitivity of the carbonization reaction efficiency to changes in the CO2 injection duration, used to determine the marginal benefit of extending the injection time.
[0054] Cross-validation (5-fold cross-validation) and a sample reweighting mechanism are introduced to assign higher sample weights to soil types with fewer samples in the training set, thereby enhancing the model's ability to identify samples from uneven soil layers. At the same time, the feature weight distribution is dynamically updated based on the ranking of feature importance, so that the model pays more attention to key parameters that contribute more to the carbonation reaction (such as near-infrared spectral intensity and resistivity).
[0055] The data acquisition center aggregates and transmits the multi-dimensional feature vectors collected by the multi-source sensor array to the central control system. After receiving the multi-dimensional feature vectors, the central control system inputs them into the trained classification-regression hybrid random forest model for parallel computation.
[0056] The classification branch classifies and distinguishes the input feature vector and outputs the unique heat code of the current soil layer type, such as sandy soil layer, cohesive soil layer, silty clay layer, etc.
[0057] The regression branch uses the convection-diffusion partial differential equations of the embedded physical core shrinkage model (SCM) as boundary constraints. Based on multi-dimensional eigenvector inputs, it preliminarily predicts the initial maximum carbon sequestration capacity and carbonization reaction potential index. The carbonization reaction potential index includes the predicted CO2 absorption efficiency, the strength enhancement potential level, and the reaction duration sensitivity index. These three dimensions each reflect different aspects of the carbonization effect: CO2 absorption efficiency determines the carbon sequestration capacity, strength enhancement potential determines the solidification mechanical properties, and reaction duration sensitivity determines the optimization direction of construction parameters. Together, they determine the generation of differentiated injection strategies.
[0058] The predicted CO2 uptake efficiency (denoted as α) is directly output from the regression branch and transformed into a dimensionless value between 0 and 1 through normalization. The normalization method is as follows: the historical maximum measured uptake efficiency for this soil type is used as the baseline value E. max E max This is derived from historical data statistics in a high-dimensional matrix database. α = predicted absorption efficiency value / E max When the predicted value exceeds E max When α is 1.0, the closer α is to 1, the closer the soil layer is to the optimal carbon sequestration state.
[0059] Strength enhancement potential level (denoted as β): determined by the regression branch based on predicted compressive strength, micro-interfacial binding energy, and θ. NIR Comprehensive Judgment. The regression branch compares the comprehensive evaluation results with preset multi-level thresholds, outputting three levels: "High Potential," "Medium Potential," and "Low Potential." To facilitate the calculation of the potential index, it is mapped to numerical values: 1.0 for high potential, 0.5 for medium potential, and 0.2 for low potential.
[0060] The sensitivity index to reaction duration (denoted as γ) is derived from the regression branch based on the chemical reaction rate constant k.chem With θ NIR The decay curve of the change was calculated. The regression branch outputs three levels: "high sensitivity," "medium sensitivity," and "low sensitivity." High sensitivity indicates that a unit time extension results in a significant increase in carbon sequestration (k). chem The slow rate of descent and low sensitivity indicate that the carbonization reaction is approaching saturation (k). chem (The rate of decline is rapid). To facilitate the calculation of the potential index, it is mapped to numerical values: 1.0 for high sensitivity, 0.5 for medium sensitivity, and 0.2 for low sensitivity.
[0061] The carbonization reaction potential index (denoted as P) is calculated using the weighted comprehensive evaluation method: P = w1·α + w2·β + w3·γ Where w1, w2, and w3 are the weight coefficients of each sub-indicator, satisfying w1 + w2 + w3 = 1. The weight coefficients are dynamically adjusted based on the feature importance ranking results of the random forest model. The feature importance ranking is determined by the contribution of each feature to the prediction results during model training. Generally, CO2 absorption efficiency has the largest weight in terms of its impact on carbonization effect, followed by the potential for intensity enhancement, while the sensitivity of reaction duration as an auxiliary decision indicator for injection duration has the smallest weight.
[0062] To achieve the aforementioned regression branch's "using the physical core shrinkage model as boundary constraints," the physical core shrinkage model specifically employs convection-diffusion partial differential equations to describe the transport and reaction process of CO2 in the soil pore fluid. The equation's form is as follows:
[0063] in, C This refers to the mass concentration of CO2 dissolved in the pore fluid of the soil layer, expressed in kg / m³. 3 ; t The carbonization stirring time is expressed in seconds (s). r These are the axial and radial coordinates of the pile, in meters. The effective diffusion coefficient in the SCM governing equation; u R is the convective velocity generated by the gas supply system, measured in m / s. CO2 This is a consumption item for the carbonization reaction. , It is the chemical reaction rate constant. This represents the area of the unreacted core.
[0064] The boundary conditions for the above partial differential equations are: Initial conditions (t=0): The overall concentration inside the pile is the background value. ; External boundary conditions (r=Rpile, pile edge): ; Inner boundary conditions ( (Drill bit gas outlet): ( For gas output partial pressure, (where is the solubility constant).
[0065] Meanwhile, the central control system calculates key parameters in the physical core shrinkage model based on real-time feedback data from the multi-source sensor array, including the effective diffusion coefficient D. eff With the chemical reaction rate constant k chem .
[0066] Effective diffusion coefficient D eff The results were obtained based on real-time inversion of porosity (converted from earth pressure and moisture sensor data), real-time temperature, and carbonization activation energy pre-calibrated by molecular dynamics simulation:
[0067] in, The free water diffusion coefficient, Real-time porosity obtained by inversion using earth pressure and moisture sensors. The bonding index is 2.0. The real-time temperature (K) is given. The carbonization activation energy is calibrated by MD simulation.
[0068] Chemical reaction rate constant k chem The calculations were based on the calcium carbonate formation reaction progress derived from real-time conductivity and near-infrared spectroscopy:
[0069] in, Based on the reaction rate, The real-time conductivity is measured by an anti-interference composite sensor. The conductivity reference value is used. To normalize the reaction progress by scanning the absorbance of the characteristic peaks of calcium carbonate generated in the 1000-2500 nm band using near-infrared spectroscopy ( ),Should It is obtained by real-time acquisition and conversion by a near-infrared spectral sensor.
[0070] Based on the aforementioned key parameters, the convection-diffusion equations of the physical core shrinkage model are numerically solved using the finite difference method. This allows for the calculation of the theoretical upper limit value Y for carbon fixation under the current operating conditions, which conforms to the law of mass conservation and carbonization reaction kinetics. max .
[0071] The system uses a post-processing interception mechanism to estimate the maximum carbon sequestration capacity Y. raw Make corrections: If Y raw ≤Y maxThen output Y directly. raw As the final predicted carbon sequestration amount; if Y raw >Y max If so, then correct it as follows: Y final = Y max - 0.1×(Y raw -Y max This mechanism uses physical laws to enforce constraints on the abnormal output of purely data-driven models, preventing prediction results from violating basic physical and chemical conservation laws.
[0072] Step 3: The central control system dynamically generates differentiated CO2 injection strategies based on soil type identification and carbonation reaction potential index, and controls the constant-pressure variable-pressure gas supply system to execute CO2 injection according to the differentiated CO2 injection strategies. The differentiated CO2 injection strategies include differentiated adjustments to CO2 injection volume, injection rate, and injection pressure, specifically including: When the soil layer is identified as a sandy soil with high porosity and permeability, and the carbonation reaction potential index is higher than a preset threshold, the system switches to a staged foaming high-pressure injection mode: In the first stage (before the initial setting of the geopolymer gel), the gas supply system is controlled to inject foamed CO2 at a rate of 5-10 tons / hour, utilizing foaming to increase the contact area between the gas and soil particles; in the second stage (during the initial setting period), the injection rate is increased to 12-18 tons / hour. At this time, the gel network is initially formed, and the higher CO2 concentration drives the deep carbonation reaction, maximizing carbon sequestration. The specific method of foaming injection is as follows: an appropriate amount of foaming agent (such as sodium dodecyl sulfate, added at 0.5%-1% of the CO2 mass) is added to the CO2 gas, and micro-foam is formed through a dedicated foaming device in the gas control module before being injected into the soil layer.
[0073] When the soil layer is determined to be a cohesive soil layer with low porosity and low permeability and the carbonation reaction potential index is lower than the preset threshold, the system switches to a low-pressure slow release and concentration compensation injection mode: the gas supply system is controlled to continuously inject CO2 at a constant low pressure rate of 5-9 tons / hour, and the injection pressure is controlled within a safe range below the soil splitting pressure through the pressure regulating valve to avoid high-pressure gas causing soil splitting failure in dense soil layers; at the same time, the concentration of CO2 in the gas is appropriately increased to compensate for the insufficient gas diffusion efficiency in low-permeability soil layers.
[0074] As a further preferred embodiment of this application, this embodiment also includes a closed-loop self-learning control step. During and after the construction of a single mixing pile, a near-infrared spectral sensor continuously tracks the intensity changes of the characteristic peaks of calcium carbonate formation and inversely generates the actual carbon sequestration amount. After construction, the system automatically calculates the residual between the actual carbon sequestration amount and the carbon sequestration amount predicted by the model during the pile construction process. If the absolute value of the residual exceeds a preset tolerance threshold (e.g., ±15% of the predicted value), the system automatically labels the sample as a "high-bias sample" and gives it a higher learning weight in subsequent incremental training. The system extracts representative samples from all the data generated by the current construction, and the number of new samples is controlled to be one-fifth of the existing total training set size. Then, the background incremental training module is started to perform a limited number of retraining rounds based on the original model parameters, rather than retraining from scratch, in order to balance learning efficiency and model stability.
[0075] During incremental training, the system dynamically updates the feature importance weights based on the working conditions in newly added samples. For example, when the system operates in areas rich in groundwater, it adaptively increases the weights of soil moisture and electrical conductivity features in the model's decision-making through feature contribution analysis during incremental training, while correspondingly decreasing the weights of other features. Through this dynamic weighting mechanism, the model's prediction accuracy under different geological conditions continuously improves, enabling the carbonization system to possess an adaptive evolutionary capability that becomes more accurate with each drilling operation.
[0076] After the completion of all the mixing pile construction, the pile body was inspected for quality. The inspection results showed that the pile core samples were continuous and intact, the solidification of each layer was uniform, and the compressive strength and carbon sequestration both met the design requirements. Compared with traditional carbonized mixing piles that do not identify soil layers and use constant pressure and constant rate injection, the method of this application significantly improves the carbon sequestration efficiency in high-permeability soil layers, effectively avoids soil splitting accidents in low-permeability cohesive soil layers, and significantly reduces overall carbon emissions compared with traditional cement mixing piles.
[0077] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.
Claims
1. A method for controlling the carbonization construction of industrial solid waste mixing piles based on soil layer identification, characterized in that: Includes the following steps: A clinker-free curing agent is injected into the soil layer, and the physical and spectral parameters of the target soil layer are collected in real time through a multi-source sensor array integrated at the drill bit to construct a multi-dimensional feature vector of the current soil layer. The multidimensional feature vector is input into a pre-constructed classification-regression hybrid random forest model. The classification branch of the model outputs the soil layer type identifier; the regression branch of the model uses the convection-diffusion partial differential equation of the embedded physical core model as the boundary constraint condition, and outputs the carbonation reaction potential index and the initial maximum carbon sequestration amount of the current soil layer; the regression branch of the model also corrects the initial maximum carbon sequestration amount based on the theoretical carbon sequestration upper limit calculated by the physical core model, and outputs the corrected maximum carbon sequestration prediction value. The central control system dynamically generates differentiated CO2 injection strategies based on soil type identification and carbonation reaction potential index, and controls the constant-pressure variable-pressure gas supply system to execute CO2 injection according to the differentiated CO2 injection strategies. The differentiated CO2 injection strategies include differentiated adjustment of CO2 injection volume, injection rate and injection pressure.
2. The method for controlling the carbonization construction of industrial solid waste mixing piles based on soil layer identification according to claim 1, characterized in that: The multidimensional feature vector includes: stirring torque obtained by torque sensor, drilling speed obtained by rotation speed sensor, soil conductivity obtained by current-voltage four-terminal electrode, ultrasonic pulse velocity obtained by ultrasonic testing module, near-infrared spectral intensity obtained by near-infrared spectral sensor scanning, soil moisture and temperature obtained by soil temperature and humidity sensor, and soil pressure obtained by piezoelectric crystal matrix earth pressure sensor.
3. The method for controlling the carbonization construction of industrial solid waste mixing piles based on soil layer identification according to claim 1, characterized in that: The constraint embedding process of the physical core shrinkage model includes: the regression branch outputs the initial maximum carbon sequestration based on the input multidimensional feature vector, and at the same time calculates the key parameters in the physical core shrinkage model based on the real-time data obtained by the multi-source sensor array, and numerically solves the convection-diffusion equation of the physical core shrinkage model to calculate the theoretical upper limit of carbon sequestration under the current working condition that conforms to the law of conservation of mass. The initial maximum carbon sequestration is corrected through a post-processing interception mechanism: if the initial maximum carbon sequestration does not exceed the theoretical upper limit of carbon sequestration, the initial maximum carbon sequestration is directly output as the predicted value of the maximum carbon sequestration; if the initial maximum carbon sequestration exceeds the theoretical upper limit of carbon sequestration, the excess is reduced by a predetermined ratio and then output as the predicted value of the maximum carbon sequestration.
4. The method for controlling the carbonization construction of industrial solid waste mixing piles based on soil layer identification according to claim 3, characterized in that: The effective diffusion coefficient and chemical reaction rate constant in the physical core shrinkage model are calculated and dynamically updated in real time by a sensor array. The effective diffusion coefficient is determined by the real-time inverted porosity, real-time temperature and activation energy calibrated by molecular dynamics simulation, and the chemical reaction rate constant is determined by the real-time conductivity and the calcium carbonate formation reaction progress inverted by near-infrared spectroscopy.
5. The method for controlling the carbonization construction of industrial solid waste mixing piles based on soil layer identification according to claim 1, characterized in that: The pre-built classification-regression hybrid random forest model is established using the following cross-scale physical model initialization method: Collect macroscopic geological survey data and laboratory autoclave control test data; Molecular dynamics simulations were performed to extract microscopic parameters, including the reaction interface binding energy, key ion diffusion coefficient, and CO2 dissociation carbonization unit reaction rate constant. The microscopic parameters are mapped and transformed into the dynamic boundary conditions and effective diffusion coefficient benchmark values of the macroscopic convection-diffusion equation in the physical core shrinkage model. A high-dimensional matrix database containing the correlation between ultrasonic pulse velocity, near-infrared spectrum, reaction interface binding energy, carbonate content and compressive strength is constructed as the training basis for the model.
6. The method for controlling the carbonization construction of industrial solid waste mixing piles based on soil layer identification according to claim 1, characterized in that: The differentiated CO2 injection strategy is as follows: When the soil layer type is determined to be a sandy soil layer with high porosity, high permeability and high reaction potential, the gas supply system is controlled to execute a staged foaming high-pressure injection mode, injecting CO2 at a first rate before the initial setting period, and increasing to a second rate after the initial setting period. When the soil layer is determined to be a cohesive soil layer with low porosity, low permeability, and low reaction potential, the gas supply system is controlled to execute a low-pressure slow-release and concentration-compensated injection mode to inject CO2 at a constant low-pressure rate.
7. The method for controlling the carbonization construction of industrial solid waste mixing piles based on soil layer identification according to claim 1, characterized in that: It also includes closed-loop self-learning control steps: During and after the construction of a single mixing pile, the actual carbon fixation amount inverted from the actual near-infrared calcium carbonate formation characteristic peak fed back by the drill bit is compared with the maximum carbon fixation amount predicted by the model to calculate the residual. Representative samples generated during the current construction are extracted as new data to perform periodic incremental training on the model; and the feature importance weights are dynamically updated based on the new data to achieve adaptive evolution under heterogeneous geological conditions.
8. The method for controlling the carbonization construction of industrial solid waste mixing piles based on soil layer identification according to claim 1, characterized in that: The carbonization reaction potential index includes the predicted value of CO2 absorption efficiency, the intensity enhancement potential level, and the sensitivity index of reaction duration.
9. The method for controlling the carbonization construction of industrial solid waste mixing piles based on soil layer identification according to claim 1, characterized in that: The raw materials for preparing the clinker-free curing agent include converter smelting slag, RH furnace refining slag, fly ash, magnesium olivine, and alkali activator.
10. A soil mixing pile device for implementing the method described in any one of claims 1-9, characterized in that: include: The drilling system includes an outer drill rod and an inner drill rod that are coaxially mounted. The outer drill rod and the inner drill rod are driven by independent servo motors and rotate in opposite directions. The outer and inner drill rods are respectively provided with stirring blades on their outer peripheries for bidirectional shearing and stirring of the soil and the solidifier without clinker. The data acquisition system includes a multi-source sensor array and a data transmission channel integrated at the drill bit. The multi-source sensor array is used to synchronously acquire multi-source parameters of the soil layer during the mixing and carbonization process. The multi-source parameters of the soil layer are transmitted back to the surface control platform in real time via the data transmission channel. The material injection system includes a material storage module, a material transmission channel, and a material outlet located at the drill bit, for injecting the clinker-free curing agent into the soil layer; The gas injection system, including a constant and variable pressure gas supply system, a gas transmission pipeline, and a gas outlet located at the bottom of the drill bit, is used to inject CO2 into the soil layer. The central control system is used to receive multi-source soil parameters transmitted back by the data acquisition system, identify the current soil type and predict the carbonation reaction potential index through the built-in prediction model based on physical constraint machine learning, dynamically generate a differentiated CO2 injection strategy, and control the constant pressure gas supply system to adjust the amount, rate and pressure of CO2 injected into the soil layer.