Bagheri et al. (2019) — The Use of Machine Learning in Boron-Based Geopolymers: Function Approximation of Compressive Strength by ANN and GP

Citation

Bagheri, A., Nazari, A., & Sanjayan, J. (2019). The use of machine learning in boron-based geopolymers: Function approximation of compressive strength by ANN and GP. Measurement, 141, 241–249.

• DOI: 10.1016/j.measurement.2019.03.001
• Atlas layer: extension
• Related Victor Li book chapter: Chapter 9: Green ECC (Alkali-Activated Geopolymer Binders) & Chapter 11: Computational Materials Design
• Source PDF: bagheri-2019-the-use-of-machine-learning-in.pdf
• Extracted text: full_text/bagheri-2019-the-use-of-machine-learning-in_full_text.md
• Source note: source_notes/bagheri-2019-the-use-of-machine-learning-in_source_note.md

Why this paper matters

Applies supervised machine learning—specifically Artificial Neural Networks (ANN) and Genetic Programming (GP)—to establish explicit mathematical functions predicting the compressive strength of boroaluminosilicate geopolymers (BASG) based on chemical precursor ratios and activator ion proportions.

Main contribution

• Develops an AI-driven predictive modeling framework for boroaluminosilicate geopolymers (BASG) synthesized from fly ash, slag, borax, sodium hydroxide, and sodium silicate.
• Curates an experimental database of 114 mixture formulations mapping 5 input variables: % fly ash, % slag, and molar ratios of B, Si, and Na ions in the alkaline activator solution.
• Derives an explicit, closed-form symbolic regression formula via Genetic Programming to estimate 7-day ambient-cured compressive strength ($R^2 = 0.95$, $RMSE = 0.07$).
• Identifies that Si and B ion concentrations exert the strongest positive influence on geopolymer network cross-linking and compressive strength development.

Evidence summary

• Material Matrix: Boroaluminosilicate Geopolymer (BASG) using Australian fly ash and blast furnace slag activated by $\text{NaOH}$ (3, 5, 8 M), $\text{Na}_2\text{SiO}_3$, and Borax ($\text{Na}_2\text{B}_4\text{O}_7\cdot 10\text{H}_2\text{O}$).
• Database & Model Parameters: 114 experimental points; inputs: %F, %S, B/AA, Si/AA, Na/AA; output: 7-day ambient compressive strength (5–70 MPa).
• Model Performance:
• ANN: Optimized topology with hidden layer neurons yielding training/testing $R^2 > 0.96$.
• Genetic Programming (GP): Explicit non-linear function achieving $R^2 = 0.95$ and $RMSE = 0.07$ on testing dataset.
• Sensitivity Findings: Increasing the Si/AA ratio enhances aluminosilicate gel polymerization, while boron substitution reinforces the network structure when paired with calcium from slag.

Linked Atlas nodes

• 04_material_systems/engineered_geopolymer_composites.md
• 06_sustainability/industrial_byproduct_binders.md
• 02_concepts/machine_learning_models.md

Relationship to Victor Li book

• Extends Victor Li (2019) Chapter 9 (Green ECC) by exploring boron-modified geopolymer chemistry (BASG) to broaden the scope of low-carbon alkali-activated binders.
• Connects with emerging data-driven and computational design paradigms for predicting binder properties without exhaustive trial-and-error laboratory batches.

Claim-evidence rows to add

Verification status

• PDF preserved: yes (bagheri-2019-the-use-of-machine-learning-in.pdf)
• Text extracted: yes (full_text/bagheri-2019-the-use-of-machine-learning-in_full_text.md)
• DOI verified: yes (10.1016/j.measurement.2019.03.001)
• Metadata verified: yes (Measurement, Vol. 141, pp. 241–249, 2019)
• Claim-evidence matrix ready: yes

Cautions

• The paper focuses on binder matrix compressive strength prediction using machine learning and does not investigate fiber reinforcement or uniaxial tensile strain-hardening (ECC).
• Models predict 7-day ambient cured compressive strength; long-term strength and durability kinetics require separate verification.