Predicting Relaxed Density, Durability, and Compressive Strength of Blended Agricultural Waste Biomass Briquettes

Description

This dataset contains 63 treatment-level observations from a full factorial experiment on blended agricultural biomass briquettes, comprising 7 blend compositions × 3 particle sizes × 3 compaction pressures. Eight agricultural residues were used across the blends: banana peel, banana bunch, maize cob, maize stalk, millet bran, coffee husk, groundnut shell, and bean husk. For each treatment combination, the dataset reports mean values and standard deviations for three mechanical quality indicators: compressive strength (MPa), impact resistance index (%), and relaxed density (g/cm³), along with process parameters (particle size, compaction pressure, moisture content) and blend composition (percentage by mass of each feedstock). This dataset supports the study "Interpretable Machine Learning for Predicting Relaxed Density, Durability, and Compressive Strength of Blended Agricultural Biomass Briquettes."

Steps to reproduce

Collect eight agricultural residues from Mbarara City and Isingiro District, south-western Uganda: Banana Peel (BP), Banana Bunch (BB), Maize Cob (MC), Maize Stalk (MS), Millet Bran (MB), Coffee Husk (CH), Groundnut Shell (GS), and Bean Husk (BH). Manually sort each to remove inorganic contaminants (stones, soil, plastic). Sun-dry each feedstock to a stable moisture content, then grind using a multipurpose laboratory blender. Sieve into three particle size classes: fine (<1.00 mm), medium (<2.75 mm), and coarse (<4.75 mm). Formulate seven blend compositions (B1–B7) using simplex-centroid mixture design principles across three feedstock groups (banana, cereal, coffee/legume residues), with the fuel fraction fixed at 70% of briquette mass, as specified in Table 1 of the manuscript. Prepare the binder at a fixed 30% of total briquette mass: cow dung (20%), cassava starch (5%), and molasses (5%). Mix binder thoroughly into each feedstock blend. Compact each blend using a laboratory hydraulic press with a 60 mm internal-diameter cylindrical steel die, at one of three compaction pressures (8, 10, or 12.5 MPa) for a 30-second dwell time. This produces the full factorial of 7 blends × 3 particle sizes × 3 pressures = 63 experimental conditions, with 25–30 briquettes produced per condition. Condition briquettes at ambient temperature and humidity for 48 hours after ejection, before any mechanical testing. Measure relaxed density: record diameter (D) and height (H) using a digital Vernier caliper, and mass (m) using a digital analytical balance. Compute density as ρ = m / (π(D/2)²H). Measure 3–20 briquettes per condition. Measure durability (Impact Resistance Index): drop each briquette 5 times from 2 m onto a concrete floor, weigh the retained mass after the fifth drop, and calculate IRI (%) = (retained mass / original mass) × 100. Measure 2–7 briquettes per condition. Measure compressive strength: load each briquette axially to failure using a Universal Testing Machine (protocol adapted from ASTM D2166), and calculate σc = Fmax / A, where A = π(D²)/4. Compile the resulting dataset (particle size, compaction pressure, moisture content, seven blend fractions, and the three measured quality outcomes) as provided in Supplementary Material S1, then run the machine learning pipeline (Python 3.7; scikit-learn, XGBoost, CatBoost, LightGBM, SHAP) with an 80/20 stratified train-test split, RandomizedSearchCV hyperparameter tuning under repeated stratified 5-fold CV, and SHAP TreeExplainer analysis on the best-performing model per target.

Institutions

• Mbarara University of Science and Technology

  Western Region, Mbarara
• KU Leuven

  Flanders, Leuven
• University of Antwerp

  Flanders, Antwerp
• Maseno University

  Kisumu County, Kisumu

Categories

Funders

• University as a Facilitator Community Based Sustainable Solutions to Demographic Challenges in South Western Uganda(UCoBS)project of Mbarara University of Science and Technology (MUST) funded by VLIR-UOS Institutional University Cooperation (IUC) program

Licence