Established Data Points, cGAN Augmented Data and related Code: International Conference on Sustainable Materials and Infrastructure IIT Jammu, INDIA AI - Optimised Basalt - Graphene Cementitious Composite: Sustainable High - Performance Concrete for

This entry contains the Established Data Points as well as the Conditinal GAN (cGAN) augmented Data Points on which Computational Optimization was performed to arrive at the conclusions of "Dutta Roy, A., & Makanadar, A. (2025, December 17). International Conference on Sustainable Materials and Infrastructure IIT Jammu, INDIA AI - Optimised Basalt - Graphene Cementitious Composite: Sustainable High - Performance Concrete for Seismic and Blast Resilience with Integrated Structural Monitoring. International Conference on Sustainable Materials and Infrastructure 2025 (ICSMI 2025), Indian Institute of Technology, Jammu. https://doi.org/10.5281/zenodo.18223083" The References of the Established Data Points can be obtained by checking the References of the Full Conference Paper available at "Dutta Roy, A., & Makanadar, A. (2025, December 17). International Conference on Sustainable Materials and Infrastructure IIT Jammu, INDIA AI - Optimised Basalt - Graphene Cementitious Composite: Sustainable High - Performance Concrete for Seismic and Blast Resilience with Integrated Structural Monitoring. International Conference on Sustainable Materials and Infrastructure 2025 (ICSMI 2025), Indian Institute of Technology, Jammu. https://doi.org/10.5281/zenodo.18223083" The creation of this dataset is referenced in the Original Paper available at https://doi.org/10.5281/zenodo.18223083 Core dataset references (Data/ Data.zip): UHPC dataset (Kashem et al.)Kashem, A., Karim, R., Malo, S. C., & Das, P. (2023). Ultra-High-Performance Concrete (UHPC) [Data set]. Mendeley Data, V1. https://doi.org/10.17632/85r7bh4zsz.1 Global SCM-blended concrete dataset – articleLiu, W., Mo, H., & Lee, C. K. (2025). Comprehensive dataset for performance evaluation of SCM-blended concrete. Data in Brief, 60, 111616. https://doi.org/10.1016/j.dib.2025.111616 Global SCM-blended concrete dataset – repository record Liu, W., Mo, H., & Lee, C. K. (2025). A global dataset of SCM-blended concrete and its properties [Data set]. Mendeley Data, V2. https://doi.org/10.17632/npwgdyy4np.2 “ASTM International. Environmental Product Declarations for concrete mixes from Altaview Concrete and Argos USA (multiple EPDs; see dataset columns ‘EPD Program Operator’ and ‘EPD Source Link’ for specific documents).” “NREL. U.S. Life Cycle Inventory (USLCI) database, and US‑EI 2016/2020 process datasets (gravel, water, waste, etc.).” “Ecoinvent 3.4 database (electricity and chemical processes) and ELCD PP fibre dataset (EU‑27, 2005).” “Portland Cement Association industry average cement EPDs (2016, 2021); Slag Cement Association industry average slag cement EPD (2014); EFCA admixture EPDs (2015).” __________________ The Features of the dataset Merged_Concrete_Reference_Data.csv are as follows: The dataset was created by calibrating multivariate linear models to existing data that is referenced in the Core dataset [Core dataset references (Data/ Data.zip)]. Each row in the dataset contains data on cement content, fly ash replacement ratio, water to cement ratio, basalt fiber content, graphene oxide content, and superplasticizer content, as well as 28 - day compressive strength, permeability index, and CO₂ footprint per cubic meter of concrete. The limitations of this dataset are: The small number of samples (n = 10), The use of regression models instead of direct experimental data, Contradictory trends seen for some CO₂ composition correlations, and the extrapolation of basalt fiber and graphene oxide properties outside of the explicitly documented ranges in the source literature and EPDs. __________________ The Features of the dataset Final_PreGAN_Dataset.csv are as follows: Starting from the 10 point merged and generated reference [Merged_Concrete_Reference_Data.csv], we map the UHPC and SCM datasets into a common feature space and arrive at 2,200+ records, while keeping strength, permeability, and CO₂ values within the broad ranges seen in the underlying sources and seeding a small number of fibre and nano added mixes as anchors for later augmentation work. Limitations: The current CO₂ model shows inverted trends with binder composition (CO₂ decreasing with higher OPC_pct and increasing with higher SCM contents), the binder fraction (OPC_pct + Flyash_pct + GGBS_pct) falls noticeably below full mass balance in certain rows, BF and GO are present in approximately 1 out of 100 records, and much of the permeability information for SCM ‑ type mixes is inferred via Computational means. __________________ Features of Conditional_GAN.ipynb notebook, Final_PreGAN_Dataset.csv, and cGAN_Augmented_GFBC_Data.csv: Final_PreGAN_Dataset.csv serves as the training backbone, merging UHPC_Data.csv, SCM_Concrete_Data.csv, and EPD ‑ derived carbon/ durability information via Merged_Concrete_Reference_Data.csv into a single, general schema. Each record in the pre ‑ GAN dataset includes OPC content, fly ash and GGBS proportions, water to cement ratio, superplasticizer dosage, optional basalt fibre and graphene oxide contents, 28 ‑ day compressive strength, a permeability indicator, and an estimated CO₂ imprint per cubic metre. The Conditional_GAN.ipynb notebook implements a fully connected conditional GAN, trained on a MinMax scaled subset of these variables (binder percentages, fibre/ nano additive dosages, strength, CO₂, and permeability). Conditioning is applied through two binary flags that indicate the presence or absence of basalt fibre and graphene oxide in each mix. After training, the generator is used to sample 1,000 additional concrete mix designs, which are concatenated with the original 2,200+ rows from Final_PreGAN_Dataset.csv to create cGAN_Augmented_GFBC_Data.csv. The pre ‑ GAN dataset is broadly consistent with the magnitudes and basic physical behaviour of the original UHPC, SCM, and EPD‑derived sources, but shows non ideal CO₂ binder trends and imperfect binder mass balance (OPC + fly ash + GGBS). The augmented dataset should be treated as a V1.0 with future Versions coming to increase the efficiency: it preserves key empirical correlations and improves coverage of fibre and nano added mixes, but some synthetically generated rows contain physically inconsistent or out of range values and a few missing design entries __________________ The Novel Contributions and Limitations of SurrogateModellingOptimalMix.ipynb (with reference to the data cGAN_Augmented_GFBC_Data it is operating on): Novel Contributions: Proposes a multi - objective surrogate framework that jointly targets 28 ‑ day strength, CO₂ footprint, durability/permeability, and proxy blast and seismic indices for basalt–graphene green concrete and then proceeds to Construct a unified mix ‑ design feature space (OPC, fly ash, GGBS, basalt fibre, graphene oxide, water–cement ratio, superplasticizer) so that heterogeneous UHPC/ SCM/ EPD sources can be analysed and optimised within one schema. Employs a conditional GAN trained on this feature space to augment the sparse literature data and densify the design region, especially for mixes containing basalt fibres and graphene oxide, yielding a large cGAN_Augmented_GFBC_Data dataset for learning. Trains task specific surrogate models, a two layer MLP for strength, random forest for CO₂, gradient boosting for durability, and MLPs for blast and seismic proxies so that each objective can exploit an appropriate regression architecture. Implements a numerical optimisation loop over continuous mix variables to generate candidate “optimal” mixes for each objective (e.g., high OPC + GO for strength; low ‑ OPC, SCM ‑ rich, fibre/ nano reinforced mixes for low CO₂ and toughness related indices). Explicitly includes blast and seismic performance proxy indices in concrete mix optimisation, extending typical strength durability CO₂ trade off studies toward extreme‑event resilience. Demonstrates how conditional GAN augmentation plus surrogate optimisation can serve as a transferable template for other multiproperty materials design tasks. Limitations: The cGAN_Augmented_GFBC_Data dataset contains some physically inconsistent rows (e.g., negative strengths, negative or >100% binder fractions, extreme or negative permeability values), so surrogates may partially fit GAN artefacts rather than true material behaviour. The current CO₂ model exhibits inverted trends with binder composition (CO₂ decreasing with higher OPC and increasing with higher SCM content), which undermines the physical reliability of the “low‑CO₂” objective and its predicted optima. Durability, blast, and seismic targets are surrogate indices built from limited or inferred data; in optimisation runs they can yield non‑intuitive solutions (e.g., durability optimum near 1% OPC with a negative durability score), signalling extrapolation beyond the well‑constrained region of the data. The high R² values reported for the surrogates (0.80 - 0.97) are not consistently reproduced in the latest notebook execution, where several models show negative R² on held‑out data, indicating sensitivity to data preprocessing, outlier handling, and train–test splitting. The framework has not yet been validated against an independent set of experimentally cast mixes, so all “optimal” designs should currently be interpreted as computational hypotheses requiring laboratory confirmation. The dataset provided here is used to work upon the paper "Dutta Roy, A., & Makanadar, A. (2025, December 17). International Conference on Sustainable Materials and Infrastructure IIT Jammu, INDIA AI - Optimised Basalt - Graphene Cementitious Composite: Sustainable High - Performance Concrete for Seismic and Blast Resilience with Integrated Structural Monitoring. International Conference on Sustainable Materials and Infrastructure 2025 (ICSMI 2025), Indian Institute of Technology, Jammu. https://doi.org/10.5281/zenodo.18223083)". The paper was originally pr