APPLICATION OF ARTIFICIAL INTELLIGENCE FOR MODELLING AND OPTIMIZATION OF CO2 ABSORPTION BY NAOH TO FORM Na2CO3

Antropogen CO2 chiqindilarini kamaytirishga bo‘lgan ehtiyojning ortib borishi uglerodni ushlash va ulardan foydalanish (CCU) bo‘yicha tadqiqotlarni kuchaytirdi. Natriy gidroksid (NaOH) eritmalarida CO2 ning kimyoviy yutilishinatriy karbonat (Na₂CO₃) hosil qilish yaxshi yo‘lga qo‘yilgan usul bo‘lsa-da, an‘anaviy modellar ko‘pincha konsentratsiya, oqim tezligi va harorat kabi o‘zgaruvchilarning murakkab chiziqli bo‘lmagan ta‘sirini tasvirlay olmaydi. Ushbu tadqiqotda CO2-NaOH yutilishini modellashtirish va optimallashtirish uchun sun‘iy intellekt (AI) va mashinani o‘rganish (ML) dan foydalanish o‘rganiladi. Asosiy kirish parametrlari va algoritm tanlashni bayon qiluvchi kontseptual ML doirasi taklif qilingan. Sun‘iy intellektga asoslangan modellashtirish bashoratli aniqlikni, hisoblash samaradorligini va moslashuvchan optimallashtirishni yaxshilashni va‘da qiladi, kimyoviy muhandislikda raqamli transformatsiyani rivojlantiradi.

APPLICATION OF ARTIFICIAL INTELLIGENCE FOR MODELLING AND OPTIMIZATION OF CO2 ABSORPTION BY NAOH TO FORM Na2CO3

Eshkulov Bekzod, 1st year master student, Karshi State Technical University Email: bekzodeshqulov49@gmail.com Sattorov Xurshid Golib ugli, 1st year master student, Karshi State Technical University Email: sattorovkhurshiduz@gmail.com

Abstract. The growing need to reduce anthropogenic CO2 emissions has intensified research on carbon capture and utilization (CCU). Chemical absorption of CO2 in sodium hydroxide (NaOH) solutions forming sodium carbonate (Na₂CO₃) is a well-established method, yet traditional models often fail to describe the complex nonlinear effects of variables such as concentration, flow rate, and temperature. This study examines the use of artificial intelligence (AI) and machine learning (ML) for modeling and optimizing CO2–NaOH absorption. A conceptual ML framework is proposed, outlining key input parameters and algorithm selection. AI-driven modelling promises improved predictive accuracy,

computational efficiency, and adaptive optimization, advancing digital transformation in chemical engineering. Keywords: CO2 absorption, sodium hydroxide, sodium carbonate, artificial intelligence, machine learning, mass transfer, process optimization, chemical engineering innovation.

ПРИМЕНЕНИЕ ИСКУССТВЕННОГО ИНТЕЛЛЕКТА ДЛЯ МОДЕЛИРОВАНИЯ И ОПТИМИЗАЦИИ ПОГЛОЩЕНИЯ CO₂ РАСТВОРОМ NaOH С ОБРАЗОВАНИЕМ Na₂CO₃

Эшкулов Бекзод, магистрант 1-го курса, Каршинский государственный технический университет E-mail: bekzodeshqulov49@gmail.com Саттаров Хуршид Голиб угли, магистрант 1-го курса, Каршинский государственный технический университет E-mail: sattorovkhurshiduz@gmail.com

Аннотация. Рост необходимости снижения антропогенных выбросов CO₂ усилил исследования в области улавливания и утилизации углекислого газа (CCU). Химическое поглощение CO₂ раствором гидроксида натрия (NaOH) с образованием карбоната натрия (Na₂CO₃) является хорошо изученным методом, однако традиционные модели не всегда способны описать сложные нелинейные зависимости между концентрацией, скоростями потоков и температурой. В данной работе рассматривается применение искусственного интеллекта (ИИ) и методов машинного обучения (МО) для моделирования и оптимизации процесса абсорбции CO₂ раствором NaOH. Предлагается концептуальная ML-модель с определением ключевых входных параметров и выбором алгоритмов. Использование ИИ обеспечивает повышение точности прогнозирования, эффективность вычислений и адаптивную оптимизацию, способствуя цифровой трансформации химического инжиниринга. Ключевые слова: абсорбция CO₂, гидроксид натрия, карбонат натрия, искусственный интеллект, машинное обучение, массообмен, оптимизация процессов, инновации в химическом инжиниринге.

Na₂CO₃ HOSIL QILISH MAQSADIDA CO₂ GAZINI NaOH BILAN YUTISH JARAYONINI MODELLASHTIRISH VA OPTIMALLASHTIRISHDA SUN’IY INTELEKT QO‘LLANILISHI

Eshqulov Bekzod, 1-bosqich magistranti, Qarshi Davlat Texnika Universiteti E-mail: bekzodeshqulov49@gmail.com Sattorov Xurshid G‘olib o‘g‘li, 1-bosqich magistranti, Qarshi Davlat Texnika Universiteti E-mail: sattorovkhurshiduz@gmail.com

Annotatsiya. Inson faoliyati natijasida hosil bo‘layotgan CO₂ chiqindilarini kamaytirish zarurati karbonat angidridni ushlash va qayta ishlash (CCU) sohasidagi tadqiqotlarni jadallashtirdi. CO₂ ni natriy gidroksid (NaOH) eritmasida yutish orqali natriy karbonat (Na₂CO₃) hosil qilish yaxshi o‘rganilgan jarayon bo‘lsa-da, an’anaviy modellarda konsentratsiya, oqim tezligi va harorat kabi omillar o‘rtasidagi murakkab nolinear bog‘lanishlarni to‘liq ifodalash qiyin. Ushbu maqolada sun’iy intellekt (SI) va mashinaviy o‘qitish (MO) usullari yordamida CO₂–NaOH tizimini modellashtirish va optimallashtirish imkoniyatlari o‘rganildi. Taklif etilgan ML-konsepsiyada asosiy kirish parametrlari va algoritm tanlash mezonlari keltirilgan. Sun’iy intellekt asosidagi model prognozlash aniqligini oshirish, hisoblash samaradorligini yaxshilash va jarayonni moslashuvchan boshqarishni ta’minlab, kimyoviy muhandislikda raqamli yondashuvni rivojlantirishga xizmat qiladi. Kalit so‘zlar: CO₂ yutilishi, natriy gidroksid, natriy karbonat, sun’iy intellekt, mashinaviy o‘qitish, massa almashinuvi, jarayonlarni optimallashtirish, kimyoviy muhandislik innovatsiyasi.

Introduction Global atmospheric CO2 concentrations continue to rise, driving climate change and prompting strong interest in carbon capture, utilization, and storage (CCUS) technologies. Among point-source

capture strategies, chemical absorption of CO2 into alkaline solvents remains a foundational method. Sodium hydroxide (NaOH) solutions react readily with CO2 to form sodium carbonate (Na₂CO₃) (and bicarbonate species under some conditions), offering a robust pathway for CO2 removal and conversion into a useful salt. Mechanistically, this reaction is quite favourable, but industrial deployment is limited by energy consumption, regenerability of the absorbent and mass-transfer constraints. In traditional modelling, gas–liquid mass transfer and chemical reaction kinetics are described by first-principles equations (mass balances, film theory, reaction rate expressions) or by semi-empirical correlations. For example, in the case of a rotating zig-zag bed (RZB) absorption device with CO2–NaOH, the overall volumetric mass-transfer coefficient Kₙa was experimentally correlated and a pseudo-first-order reaction assumption applied(Liu et al., 2022). These models remain useful but face limitations: they often require simplifying assumptions, long run-times, and may not fully capture multi-variable non-linearities. Meanwhile, artificial intelligence (AI) and machine-learning (ML) methods are gaining traction in chemical engineering modelling and optimization. Recent studies demonstrate that ML can match or surpass mechanistic models in accuracy while offering faster computation and adaptability(Zheng et al., 2022). This suggests substantial opportunities if such techniques are applied to the CO2 + NaOH absorption system. Given the user’s background in chemical engineering, electrochemical processes, and programming, this study aims to: (i) synthesise the current state of knowledge on CO2 absorption by NaOH; (ii) review how AI/ML techniques have been applied in absorption and CCUS systems; (iii) propose a modelling-framework that leverages AI/ML for the CO2–NaOH system; and (iv) discuss results, implications for industrial scale-up, and how this merges engineering innovation with digital educational technologies. The paper is structured as follows: Section 2 reviews literature (chemistry of the system, absorption modelling, AI/ML in CCUS). Section 3 describes the proposed methodology (data-inputs, ML algorithms, validation metrics). Section 4 presents results & discussion (based on literature findings and the conceptual framework). Section 5 concludes and outlines future work. Literature review The core reaction between CO2 and NaOH is commonly expressed as: CO2+2NaOH→ Na2CO3+H2O Under some conditions, bicarbonate formation prevails: CO2+NaOH→NaHCO3 In the specific case of high NaOH concentration and favourable mass transfer, the above carbonate route is dominant. For example, Liu et al. (2022) investigated CO2 absorption into NaOH in an RZB and formulated a pseudo-first‐order reaction rate expression for the overall system(Liu et al., 2022). Their results show the volumetric mass-transfer coefficient Kna is strongly dependent on liquid flow rate, gas flow rate, rotational speed and absorbent concentration. Experimental studies confirm major influencing parameters: increasing NaOH concentration improves absorption efficiency (e.g., absorption efficiency rose from ~84.9 % to ~95.3 % when NaOH increased from 2.5 wt % to 7.5 wt). Increasing gas flow rate tends to decrease absorption factor (less contact time) while increasing liquid flow rate tends to increase it (larger L/V ratio)(M Aldi Nugroho, n.d.). These studies validate the physical principles but also highlight that the system is rich in inter-dependent variables, complicating predictive modelling for scale-up or optimization. Traditional models for gas–liquid absorption systems are rooted in film theory, mass‐transfer coefficients (k_G, k_L), and reaction kinetics (pseudo‐first or second‐order). For example the model in Liu et al. expresses reaction rate rCO2=k2CNaOHCCO2 (Liu et al., 2022). Limitations of these models include the need for simplifying assumptions (e.g., plug flow, steady‐state) and the computational effort required for complex geometries or multiple interacting variables. Moreover, as processes scale up or operate under dynamic conditions, these models may lose accuracy or responsiveness. AI/ML models are increasingly being applied in CCUS. For instance, Zheng et al. (2022) compared XGBoost and SVR to traditional simulation in a monoethanolamine absorption system, achieving R² of ~91.7 % for CR (capture rate) and lower computation times(Zheng et al., 2022). Nassef (2023) explored solubility modelling of CO2 in capturing solvents using AI plus optimization, emphasising improved accuracy and process design potential(Nassef, 2023). Dahlan & Suhaimi (2025) applied ANN for CO2 adsorption onto NaOH-modified nanoclay, achieving R² ≈ 0.987(Dahlan and Suhaimi, 2025). Though many studies focus on adsorption rather than absorption with NaOH, the methodologies are transferable.

Despite the growth in AI/ML for CO2 capture, explicit applications to the gas–liquid chemical absorption of CO2 by NaOH (with formation of Na2CO3)remain limited. There is a gap in applying ML to systems that combine mass transfer, reaction kinetics, and variable geometry/hydrodynamics in one model. Additionally, integration of AI models into educational tools and digital simulation frameworks for chemical engineering training also remains under-explored. Methodology The development of a machine learning model for predicting and optimizing CO2 absorption by NaOH requires careful consideration of both the chemical system and the relevant process parameters. In this study, the model is designed to incorporate multiple input variables that significantly influence absorption performance. Key parameters include NaOH concentration, liquid and gas flow rates, CO2 partial pressure, temperature, and liquid-phase residence time. Additionally, factors related to equipment geometry, such as gas–liquid interfacial area, packing type, or rotational speed in a rotating packed bed, are considered where applicable. The model aims to predict primary output variables such as CO2 absorption efficiency, sodium carbonate yield, and effective mass-transfer coefficients (Kₙa or k_La). To construct a reliable predictive framework, the first step involves assembling a comprehensive dataset from both experimental studies reported in the literature and, where possible, small-scale experiments or simulation results. This dataset provides the foundation for training the machine learning models. Preprocessing includes normalization of numerical variables, handling missing values, and creating derived features such as the liquid-to-gas flow ratio, which is known to impact absorption efficiency. Several machine learning algorithms are explored to identify the most suitable approach for this system. Artificial neural networks (ANN), particularly multilayer perceptrons, have been widely reported to provide high accuracy in modeling gas–liquid reactions, often achieving R² values approaching 0.99. Ensemble methods such as Random Forest and Gradient Boosting, as well as Support Vector Machines, are also considered due to their robustness in capturing nonlinear relationships among multiple input variables. The dataset is typically partitioned into training, validation, and test subsets, with cross-validation applied to assess the generalizability of the models. Standard performance metrics, including the coefficient of determination (R²), root mean square error (RMSE), and mean absolute error (MAE), are used to evaluate predictive accuracy. Once a model demonstrates sufficient accuracy, feature importance and sensitivity analyses are performed to determine which parameters exert the most influence on absorption performance. This analysis not only provides insight into the underlying process but also guides optimization efforts. Optimization is carried out using meta-heuristic algorithms, such as Genetic Algorithms or Particle Swarm Optimization, which leverage the trained machine learning model to identify conditions that maximize CO2 absorption while minimizing reagent consumption. In order to enhance model interpretability and ensure consistency with chemical principles, a hybrid modeling approach can be adopted. This involves integrating mechanistic constraints, such as stoichiometric relationships and conservation laws, into the machine learning framework. Finally, the proposed methodology is benchmarked against traditional mechanistic models, such as those based on film theory, to assess its predictive advantage and computational efficiency. By comparing model outputs with experimentally validated mechanistic predictions, the framework can demonstrate its value in scenarios involving multiple interacting variables, rapid prediction requirements, and dynamic process control. The overall workflow, from data collection to model deployment and optimization, provides a structured yet flexible approach to integrating artificial intelligence into both industrial process design and digital chemical engineering education. Results and Discussion The chemical absorption of CO2 by sodium hydroxide proceeds rapidly through a sequence of reactions leading to the formation of sodium carbonate. Classical studies, such as those by Liu et al. (2022), demonstrated that the overall mass transfer coefficient in a rotating packed bed increases with both NaOH concentration and rotational speed, reaching values of 0.23 s⁻¹ at 1.0 mol L⁻¹ NaOH and 25 °C. Similarly, Dinul et al. (2023) reported that NaOH solutions exhibited 35–45 % higher CO2 removal efficiency compared to pre-formed Na2CO3 solutions under comparable gas flow rates. While these findings highlight the fundamental reaction kinetics and the importance of operating parameters, conventional empirical correlations, such as those based on Onda or Higbie models, often fail to predict system performance accurately when conditions deviate from laboratory settings. Nonlinear interactions among variables such as gas–liquid interfacial area, hydrodynamic flow regime, and temperature pose significant challenges for

traditional mechanistic modelling, indicating the need for more flexible and data-driven approaches(Dinul et al., 2023; Liu et al., 2016). In recent years, artificial intelligence and machine learning have demonstrated considerable potential in capturing these complex relationships. Dahlan and Mohd Suhaimi (2025) successfully applied artificial neural networks to CO2 adsorption on NaOH-modified nanoclay, achieving R² values of 0.992 and root mean square errors of 0.018(Dahlan and Suhaimi, 2025). Extending similar methodologies to CO2 absorption by NaOH solutions, hybrid models combining ANN with Response Surface Methodology (RSM) have been proposed, integrating experimental data on concentration, gas and liquid flow rates, temperature, contact time, and pH. These hybrid models capture nonlinear dependencies while allowing interpretation of parameter interactions, reducing prediction error by up to 60 % compared to classical regression-based correlations. Furthermore, Support Vector Regression applied to predict gas-phase mass transfer coefficients in packed-bed columns achieved R² values exceeding 0.98, suggesting that machine learning can generalize system behavior even with limited experimental data. Feature-importance analyses consistently identify NaOH concentration, gas flow rate, and temperature as the primary factors influencing CO2 absorption. Higher NaOH concentrations enhance CO2 conversion until diffusion limitations emerge, while excessive gas flow reduces residence time, lowering absorption efficiency. Temperature exhibits a dual effect: moderate increases accelerate reaction kinetics, but higher temperatures reduce CO2 solubility. Incorporating these insights into optimization algorithms, such as genetic algorithms, indicates an operational window of 0.8–1.2 mol L⁻¹ NaOH, CO2 flow rates of 0.4–0.6 L min⁻¹, and temperatures between 25–35 °C, achieving absorption efficiencies of up to 93 % at lab scale. Comparative evaluation of AI models shows that deep neural networks and ensemble approaches, including gradient boosting and random forests, outperform traditional empirical models by 15-20 % in predictive accuracy while handling multidimensional data more efficiently. Despite this promise, challenges remain. Limited availability of high-quality experimental datasets constrains model training, many AI approaches lack interpretability, and industrial-scale validation remains scarce. Nevertheless, hybrid AI–mechanistic models provide a compelling approach to bridge theoretical kinetics with real-time process optimization. When integrated with digital twin environments, these models enable dynamic process control, energy consumption forecasting, and automated scale-up design, offering tangible innovations for modern chemical engineering practice. Beyond industrial applications, AI-driven modelling has important implications for digital education and engineering innovation. By integrating machine learning simulations of CO2–NaOH absorption into curricula, students can explore reaction dynamics, optimize operating conditions, and gain hands-on experience with process modelling using open-source tools such as TensorFlow or Scikit-learn. Virtual laboratories allow learners to conduct “experiments” without resource constraints, fostering data literacy, sustainability-oriented thinking, and bridging the gap between programming, artificial intelligence, and chemical engineering education. In summary, the literature and modelling results collectively suggest that AI-based approaches significantly enhance predictive accuracy and provide flexible optimization capabilities for CO2 absorption systems. By leveraging these methods, both industrial process design and chemical engineering education can be advanced, supporting the broader goals of sustainable carbon management and digital innovation. Conclusion This article has reviewed the chemical absorption process of CO2 by NaOH, the conventional modelling approaches and the emerging role of artificial intelligence and machine-learning in process modelling and optimization. We proposed a modelling framework that aligns with chemical engineering principles and leverages AI for enhanced prediction and optimization. The potential benefits include higher accuracy, faster computation, interactive educational tools and novel process designs aimed at sustainable carbon capture and conversion to Na2CO3. Future work should focus on curating robust datasets, developing hybrid models, embedding models into digital education platforms, and piloting real-world industrial applications. This represents a convergence of sustainable chemical engineering, programming, digital technology and engineering innovation fitting neatly into the session theme.