MACHINE LEARNING BASED DESIGNED AND OPTIMIZATION OF SALINE FACTIONALIZED ACTIVATED CARBON DERIVED FROM COCONUT SHELL FOR THE ADSORPTION OF HEAVY METALS

CHAPTER ONE

1.0 INTRODUCTION

The contamination of water resources by heavy metals represents one of the most significant environmental challenges of the 21st century. Industrial activities, including mining, metallurgy, battery manufacturing, and chemical production, lead to the discharge of toxic metals such as lead (Pb²⁺), cadmium (Cd²⁺), chromium (Cr⁶⁺), copper (Cu²⁺), and arsenic (As) into aquatic ecosystems (Briffa, Sinagra & Blundell, 2020). Unlike organic pollutants, heavy metals are non-biodegradable, prone to bioaccumulation in living organisms, and can cause severe health detriments even at low concentrations, including neurological damage, kidney dysfunction, and cancer (Tchounwou, Yedjou, Patlolla & Sutton, 2012).

1.1 Background of the Study

To mitigate these risks, regulatory bodies like the World Health Organization (WHO) have established strict permissible limits for heavy metals in drinking water (World Health Organization, 2021). This has spurred the development of various water treatment technologies. Traditional methods for heavy metal removal, such as chemical precipitation, ion exchange, and membrane filtration, are often hampered by limitations like high energy consumption, insufficient removal efficiency, and the generation of toxic sludge (Crini, Lichtfouse, Wilson & Morin-Crini, 2019). In contrast, adsorption is widely regarded as a promising and cost-effective technique due to its simplicity, high efficiency, and potential use of low-cost materials (De Gisi, Lofrano, Grassi & Notarnicola, 2016).

Among various adsorbents, activated carbon (AC) is the most commonly used material for water purification (Bhatnagar, Sillanpää & Witek-Krowiak, 2013). However, the production of commercial AC from non-renewable sources can be expensive and unsustainable (Yahya, Al-Qodah & Ngah, 2015). In line with green chemistry and circular bioeconomy principles, there is a growing push to develop renewable and environmentally benign adsorbents from agricultural waste (Ioannidou & Zabaniotou, 2007). Coconut shells, an abundant byproduct of the coconut industry, are an excellent precursor for AC production due to their high carbon content, natural microporous structure, and mechanical strength (Guo & Lua, 2003).

The performance of activated carbon in heavy metal removal can be significantly enhanced through surface functionalization. Introducing nitrogen-containing functional groups via saline functionalization with agents like ammonium salts (e.g., NH₄Cl, (NH₄)₂SO₄) has shown particular promise (Park & Jang, 2003). This process increases surface basicity, improves electron-donor characteristics, and provides specific complexation sites for metal ions, thereby boosting both adsorption capacity and selectivity (Li et al., 2013).

The emergence of machine learning (ML) in materials science offers transformative potential for accelerating adsorbent development (Butler, Davies, Cartwright, Isayev & Walsh, 2018). ML algorithms can efficiently navigate complex parameter spaces, identify non-linear relationships, and predict optimal synthesis conditions that would be difficult to discover through traditional experimental approaches alone (Ramprasad, Batra, Pilania, Mannodi-Kanakkithodi & Kim, 2017).

1.2 Problem Statement

This passage critiques the traditional, empirical "one-variable-at-a-time" (OVAT) approach to developing and optimizing adsorbents, highlighting its critical inefficiencies and shortcomings:

1.       Inefficiency in Complex Systems: OVAT is fundamentally unequipped to handle the high dimensionality and complex, non-linear interactions between the many synthesis parameters (pyrolysis, activation, functionalization), leading to a "combinatorial explosion" of possibilities that is impossible to test exhaustively.

2.       Resource and Time Intensive: The approach requires an impractical number of experiments, making it slow, costly, and resource-heavy, often forcing researchers to settle for suboptimal results.

3.       Suboptimal Outcomes: Traditional methods frequently get stuck in local optima instead of finding the global best solution, resulting in inferior adsorbent performance.

4.       Limited Insight: Empirical methods provide little fundamental understanding or predictive capability of the underlying relationships between how a material is made, its properties, and its final performance.

5.       Poor at Balancing Trade-offs: They struggle with multi-objective optimization (e.g., balancing performance with cost), which is essential for practical applications.

1.3 Research Objectives

This research aims to develop and implement a comprehensive machine learning (ML) framework for the design and optimization of saline-functionalized coconut shell activated carbon for heavy metal adsorption. The specific objectives are:

Primary Objective: To establish an integrated machine learning pipeline that systematically optimizes the synthesis parameters of saline-functionalized CS-AC for maximum heavy metal adsorption capacity.

Secondary Objectives:

This project aims to develop a machine learning-driven framework to optimize the synthesis of saline-functionalized chitosan-activated carbon (CS-AC) for heavy metal adsorption. The core methodology involves:

1.      Data Compilation: Creating a comprehensive dataset from literature on synthesis parameters, material characteristics, and adsorption performance.

2.      Model Development & Comparison: Building and training multiple ensemble machine learning models to predict adsorption capacity.

3.      Model Interpretation: Using techniques like SHAP analysis to identify the most influential features and understand their complex relationships with performance.

4.      Parameter Optimization: Employing genetic algorithms to find the global optimum synthesis conditions for maximizing adsorption capacity.

5.      Validation & Recommendation: Validating the model against traditional methods and providing specific, actionable synthesis guidelines for experimental testing.

6.      Continuous Improvement: Establishing a closed-loop active learning system to iteratively improve the model with new experimental data.

1.4 Scope and Limitations

Scope of the Study

1.      Material: The study is limited to activated carbon made exclusively from coconut shells and functionalized with nitrogen-based salts (e.g., ammonium chloride/sulfate).

2.      Targets: The primary goal is to optimize adsorption for lead (Pb²⁺), with cadmium and copper as secondary targets.

3.      Parameters: The investigation will cover a defined range of key synthesis variables, including pyrolysis temperature and time, chemical activation agents and ratios, and saline functionalization conditions (concentration, pH, time).

4.      ML Approach: The project will use established supervised machine learning methods for regression, incorporating feature engineering, hyperparameter tuning, and model interpretation to ensure robust predictions.

Limitations and Delineations

his research acknowledges several key limitations that set boundaries on the study's immediate applicability:

1.       Data-Dependent Foundation: The initial models rely on data from published literature, which can be inconsistent in methodology and reporting, potentially introducing uncertainty.

2.       Laboratory-Scale Focus: The optimization is for lab-scale batch processes, excluding industrial-scale considerations like economics, continuous flow systems, and real-world engineering factors.

3.       Simplified Systems: The study focuses on single-metal adsorption in pure water, leaving the complexities of multi-metal competition and real wastewater chemistry for future work.

4.       Computational and Time Constraints: Model complexity may be limited by available computing power, and full experimental validation of all predictions may extend beyond the current project's timeline.

5.       Incomplete Material Data: The reliance on published data may mean that some material characterization features are missing or inconsistent, limiting the model's input features.

This study offers significant and multi-faceted contributions by introducing a novel machine learning framework to optimize sustainable adsorbents. Its key impacts are:

i.                    A New R&D Paradigm: It establishes a faster, more efficient data-driven methodology for materials discovery, drastically reducing experimental trials (by an estimated 60-70%) and development time.

ii.                  Promoting Sustainability: It advances the circular economy by valorizing coconut shell waste into high-performance adsorbents, supporting clean water and sustainable consumption goals.

iii.                Scientific Insight: It uses interpretable ML (like SHAP analysis) to uncover fundamental "structure-property" relationships, moving beyond simple correlations to a mechanistic understanding.

iv.                Practical Water Treatment: It directly contributes to more effective and economical heavy metal removal, with a projected 25-35% improvement in adsorption capacity.

v.                  Educational Value: It serves as a comprehensive, cross-disciplinary resource for integrating data science with environmental engineering and materials science.