Electrochemical phase transition composite membrane and ai4ai distributed agent modular industrial carbon capture device

Abstract

The application discloses an electrochemical phase change composite membrane and an AI4AI distributed agent modular carbon capture device, and belongs to the cross field of CCUS membrane separation and distributed intelligent control. The application adopts a plasma modified polyether sulfone phase change membrane, and low-voltage electric field is used to realize CO2 separation and dust self-cleaning; a decentralized AI4AI sensing base and a one-to-one distributed agent are matched, zero sample is used to realize single-membrane differential regulation, and digital twin is used to predict the service life of the membrane and replace consumables in batches. The device operates at a low pressure of 0.3-0.5 MPa, the adaptation period of the newly added module is shortened by 80%, the service life of the membrane is prolonged to 6.7 years, and the industrial application short board of traditional membrane method, such as unified regulation, high consumable loss and insufficient anti-pollution ability, is effectively solved.

Description

Technical Field

[0001] This invention belongs to the cross-technical fields of CCUS (Carbon Capture, Utilization and Storage) membrane separation carbon capture, electrochemically responsive membrane materials, and distributed multi-agent decentralized unsupervised regulation. It is suitable for distributed modular carbon capture of high-dust, low-concentration flue gas in cement and electrolytic aluminum production. It innovates a polyethersulfone-based electrochemical phase change composite membrane + multi-parallel spiral wound membrane monomer differential regulation + AI4AI (Artificial Intelligence for Agent Interaction) decentralized four-node peer-to-peer sensing base and four-level distributed agent (artificial intelligence agent) array architecture. It is a pure membrane technology route, without involving wet ionic liquids or solid-state electrochemical pumps, and has no technical overlap with other CCUS technology routes. Background Technology

[0002] CCUS (Carbon Capture, Utilization and Storage) is the core technology path for achieving carbon neutrality in high-energy-consuming and heavily polluting industries such as cement and electrolytic aluminum. Membrane separation carbon capture has the advantages of short process flow, small footprint and high degree of equipment modularity, and has become the mainstream development direction for decarbonization of low-concentration industrial flue gas.

[0003] Existing industrialized membrane materials are mainly hybrid matrix membranes and metal-organic framework membranes. The material preparation process is complicated and the mass production cost is high. During the operation of conventional membrane elements, there is a trade-off effect between CO2 / N2 permeation selectivity and permeation flux. Moreover, industrial flue gas dust and acidic impurities can easily cause irreversible blockage of membrane pores, which significantly shortens the service life of membrane elements.

[0004] Meanwhile, existing control methods are divided into two categories: manual fixed parameter control and single-point control based on machine learning, and there are multiple technical bottlenecks overall: ① A constant voltage is applied to all membrane modules, making it impossible to adjust the electric field parameters according to the differences in the blockage and aging status of individual membranes, leading to premature failure of some membrane elements; ② The dust cleaning cycle on the membrane surface is set by human experience. Excessive reverse cleaning will damage the membrane substrate, while insufficient cleaning will cause the flux to continue to decline. ③ Traditional machine learning models rely on a large number of labeled working condition samples to complete training, and the debugging cycle of newly built membrane collection devices can be as long as 25 days or more; ④ The lack of a digital twin estimation mechanism for the remaining lifespan of membrane elements means that only batch replacement of membrane cores is possible, resulting in high consumable maintenance costs.

[0005] It is evident that the existing membrane carbon capture materials, hardware equipment, and intelligent control systems are fragmented, making it impossible to achieve independent and differentiated regulation of a single membrane, adaptive self-cleaning, and integrated operation with full life-cycle prediction. This restricts the large-scale implementation of membrane separation CCUS technology in high-dust industrial scenarios. Summary of the Invention

[0006] 3.1 Purpose of the Invention Original polyethersulfone-based ionic liquid-doped electrochemical phase change composite membrane, with reversibly controllable pores at low voltage of 0.5~1.0V, and self-cleaning of dust on charged surfaces; A four-level distributed agent (artificial intelligence intelligent agent) array is constructed, with each spiral film equipped with an independent single intelligent agent, to achieve differentiated control of multiple parallel modules; It is equipped with a self-developed AI4AI (intelligent agent mutual collaboration framework) decentralized four-node peer-to-peer unsupervised perception base, zero-label working condition start-up, and digital twin real-time simulation of the remaining service life of each membrane. Adaptive dynamic adjustment of reverse self-cleaning cycle distinguishes between reversible dust clogging and irreversible aging of membrane substrate, resulting in a 32% reduction in consumable costs; Modular expansion and reuse of historical operating experience shortens the adaptation cycle for new membrane modules by 80%. The entire unit operates at a low pressure of 0.3~0.4MPa, with carbon capture cost controlled at 175~220 yuan / ton CO2 (carbon dioxide), suitable for small and medium-sized distributed cement and electrolytic aluminum production plants.

[0007] 3.2 Complete Integrated Technical Solution Module 1: Electrochemically Response Phase Change Composite Membrane Materials Polyethersulfone-based membranes doped with ionic liquid functional phases, plasma-modified surfaces, roll-to-roll continuous mass production, 1m width; membrane thickness 50~100μm (micrometers, unit of length), porosity 30%~40%; CO2 / N2 separation selectivity ≥1200, permeation flux ≥1500GPU (gas permeation flux standard unit); pore wettability controlled by voltage 0.5~1.0V (volts, unit of voltage), reverse voltage desorption dust self-cleaning, theoretical service life ≥6.7 years.

[0008] Module 2: Multi-parallel modular spiral wound membrane hardware unit N independent spiral wound membrane modules are connected in parallel, each with independent low-voltage power supply, pressure, and flux sensing modules; front-end dust pretreatment unit and back-end CO2 compression and purification unit; each membrane can be started and stopped independently and the applied voltage can be adjusted independently; the overall system operating pressure is 0.3~0.5MPa.

[0009] Module 3: Four-level distributed agent array + AI4AI decentralized perception base Agent hierarchy: top-level global coordination agent, middle-level group agent, bottom-level single membrane dedicated agent, fault diagnosis auxiliary agent, each membrane is assigned an independent bottom-level intelligent agent; AI4AI's decentralized foundation kernel: a four-node neighborhood peer-to-peer interactive time-series clustering unsupervised framework with no central server; every 7 days, the entire domain data is clustered and divided into three operating conditions: congestion, aging, and normal; for any membrane component failure, surrounding agents automatically fill in the monitoring, with no monitoring blind spots; Closed-loop operation process: Sensor data from each membrane unit is uploaded to a dedicated Agent → Group aggregation → Global issuance of differentiated voltage and cleaning cycle instructions → After execution, data is sent back to the base to iterate the membrane lifetime digital twin model.

[0010] 3.3 Core Innovation Protection Points Continuous roll-to-roll preparation process for plasma-modified polyethersulfone-ionic liquid electrochemical phase change composite membrane; A multi-parallel spiral-wound membrane four-level distributed one-to-one agent array differentiated control architecture; The self-developed AI4AI decentralized four-node peer-to-peer temporal clustering unsupervised sensing base is adapted for zero-sample startup of distributed membrane devices. An unsupervised automatic classification algorithm for membrane flux decline distinguishes between reversible dust blockage and membrane substrate aging. An adaptive reverse voltage self-cleaning cycle dynamic adjustment mechanism balances separation efficiency and membrane lifetime. Digital twin model for predicting the remaining service life of a single membrane element; Instantaneous high dust and flue gas pre-circuit buffer protection control logic; The new capacity expansion membrane module reuses and adapts to historical operating experience. A multi-module overall energy consumption balancing scheduling strategy is adopted to avoid premature aging of local modules due to overload. The factory reduces production and operates in a low-load, rotating, sleep-in energy-saving mode. A fault-finding and neighborhood-supplementation monitoring mechanism for membrane modules eliminates monitoring blind spots. A customizable switching and control module for both carbon capture cost and product purity; An adaptive fine-tuning algorithm for membrane voltage under all four seasons' temperature and humidity conditions ensures stable and selective separation. A micro-online replenishment model for intramembrane ionic liquids can delay flux decay. A remote iterative architecture that synchronizes industry-leading control strategies in the cloud.

[0011] 3.4 Beneficial Technical Effects Consumables: Membrane lifespan extended from 5 years to 6.7 years, and membrane replacement consumable costs decreased by 33%; Separation performance: CO2 / N2 selectivity is stable at 1230~1280, with an annual flux decline of ≤3.8%; Capacity expansion: The adaptation period for new modules has been reduced from 30 days to 5 days; Economics: The total operating cost is 175-220 yuan / ton CO2, which is 24% lower than that of traditional fixed-parameter membrane systems; Operation and maintenance: Manual operation and maintenance workload is reduced by 70%, and it can operate unattended for a long time; Operating conditions: Perfectly adapted to the harsh fumes and high dust levels in cement and electrolytic aluminum production, with significantly improved pollution resistance. Attached Figure Description

[0012] Figure 1. Overall layout diagram of the four-level distributed agent multi-membrane component; Figure 2. Schematic diagram of the cross-sectional structure of the electrochemical phase change composite membrane; Figure 3. AI4AI Decentralized Four-Node Peer-to-Peer Interaction Logic Block Diagram; Figure 4. Digital twin projection curve of membrane flux decay; Figure 5. Timing comparison curves of adaptive self-cleaning. Detailed Implementation

[0013] Test platform: Standardized 1500Nm³ / h cement kiln tail modular membrane pilot plant, with 12 membrane elements of the same specification connected in parallel, dust dynamic fluctuation of 5~45mg / m³; test cycle of 120 days, except for the intelligent control architecture, membrane material, flue gas and pressure conditions are completely uniform.

[0014] Comparative Example 1: Artificially Unified Constant Parameters All membrane elements are uniformly charged with 0.7V, and reverse cleaning is performed for 30 minutes every 7 days. Actual test results showed that after 120 days of operation, the maximum difference in flux among the membranes was 31%, and the lowest selectivity was only 980, requiring a complete batch replacement of all membrane cores.

[0015] Comparative Example 2: Global Single Backpropagation (BP) Neural Network The entire set of equipment uses a single set of control parameters, making it impossible to independently adjust the voltage and cleaning cycle of a single membrane. Actual test results showed that localized membrane blockage due to dust impact resulted in a high frequency of consumable replacements.

[0016] Example 1: The decentralized AI4AI distributed complete device of the present invention The model converged and operated stably for the first 20 days: the flux difference between membranes was ≤4%, and the annual flux decay was 3.8%; failed membrane cores were replaced in batches, reducing consumable costs by 31%; bypass protection was automatically activated under high dust conditions, and selectivity was stable at ≥1230.

[0017] Example 2: Low-load rotation hibernation verification When the flue gas load drops to 60%, the system automatically switches half of the modules to sleep mode, reducing the overall membrane aging rate by 18%.

[0018] Example 3: Module Expansion and Adaptation Verification After 90 days of operation, 6 new membrane modules were added, and the system completed adaptive operation in 4 days, with no decrease in separation efficiency.

Claims

1. An electrochemical phase change composite membrane and an AI4AI distributed agent modular industrial carbon capture device, characterized in that, include: Polyethersulfone-based electrochemical phase change composite membrane element, multi-parallel independent spiral wound membrane hardware module, single membrane one-to-one four-level distributed agent array, AI4AI decentralized four-node peer-to-peer unsupervised sensing base; The composite membrane is a plasma-modified polyethersulfone-doped ionic liquid phase change membrane, achieving reversible CO2 permeation and reverse dust self-cleaning under low pressure of 0.5~1.0V. The four-level agent system includes a top-level global agent, a middle-level group agent, a bottom-level single-membrane-specific agent, and a fault diagnosis agent. Each membrane independently and differentially adjusts its voltage and self-cleaning cycle. The AI4AI decentralized base has no central server; time-series clustering and unsupervised iterative construction of a digital twin of membrane lifespan are used. When a module fails, a neighboring agent automatically fills the gap for monitoring. The entire device operates under low pressure of 0.3~0.4MPa, suitable for distributed carbon capture of high-dust cement and electrolytic aluminum flue gas.

2. The apparatus according to claim 1, characterized in that, The composite membrane is prepared using a roll-to-roll continuous process, with a CO2 / N2 selectivity ≥1200, a permeation flux ≥1500 GPU, and a theoretical service life ≥6.7 years.

3. The apparatus according to claim 1, characterized in that, AI4AI's decentralized base can reuse existing operational data, and the adaptation cycle for new membrane modules is shortened by 80%.

4. The apparatus according to claim 1, characterized in that, The system is configured to rotate the sleep mode of the reduced production mode module to delay the aging and degradation of membrane elements.

5. The apparatus according to claim 1, characterized in that, Built-in unsupervised classification algorithm for flux decay automatically distinguishes between reversible dust clogging and irreversible aging of the membrane substrate, and formulates differentiated cleaning strategies.

6. A distributed membrane carbon capture intelligent control method based on the device according to any one of claims 1 to 5, characterized by the following steps: S1 Each independent membrane module sensor collects flux, pressure, and voltage data and uploads it to a dedicated underlying agent; The S2 mid-level group agent aggregates data from its own group, while the top-level global agent issues differentiated control instructions. S3 single membrane performs different applied voltages and reverse self-cleaning sequences; S4 runtime data is fed back to AI4AI decentralized base, time-series clustering iterative membrane lifetime digital twin; The S5 features automatic switching and control logic for high dust, low load, and module expansion scenarios, ensuring continuous closed-loop operation.

Images