Decoupling the etching-collapse competition in biochar activation: A domain-robust machine learning framework for...

Area of Science:

• Materials Science
• Chemical Engineering
• Data Science

Background:

• Potassium hydroxide (KOH) activation is crucial for enhancing biomass-derived biochar porosity.
• Optimizing KOH activation is challenging due to precursor texture dependency and literature variability.
• Predicting biochar porosity requires robust models that account for these complexities.

Purpose of the Study:

• To develop a domain-aware and interpretable machine-learning framework for biochar porosity engineering.
• To quantify activation-induced specific surface area gain relative to the precursor state using a normalized target (Δlog(S_BET)).
• To provide guidance for matching activation severity to precursor texture, reducing empirical trial-and-error.

Main Methods:

• Compiled 123 literature-derived records from 29 distinct sources.
• Developed tree-based ensemble machine-learning models for porosity prediction.
• Utilized Shapley additive explanations (SHAP) and permutation-based analyses for model interpretability.
• Performed source-grouped holdout validation and applicability-domain analysis.

Main Results:

• Machine learning models showed stable performance for predicting specific surface area gain (median R² ≈ 0.8).
• Precursor baseline texture was identified as the dominant factor influencing model response.
• KOH/Char ratio served as a controllable signal for activation severity, consistent with observed trade-offs.
• Micropore-volume prediction demonstrated weaker transferability across different literature sources.

Conclusions:

• The developed framework offers domain-qualified guidance for biochar porosity engineering.
• It enables more rational optimization of KOH activation by linking precursor properties to outcomes.
• The approach facilitates efficient porous carbon production from biomass, minimizing experimental efforts.