How AI-Powered Concrete Mix Design Accelerates Project Delivery
Developing a printable concrete mix is an iterative process. A researcher or mix designer typically starts with a reference formulation, adjusts binder ratios or admixture dosages, casts or prints specimens, waits for test results, and repeats. For 3DCP specifically, the search space is large: binder type, supplementary cementitious material loading, water-to-binder ratio, fiber content, and admixture combinations all interact in ways that are not fully captured by existing design rules. Each physical trial batch takes time and materials.
Computational screening does not eliminate that process. But it can reduce the number of iterations needed before physical testing begins.
What CEMFORGE Does
CEMFORGE is a web-based mix design prediction platform for 3D-printed concrete. The input is a mix proportion — binder content, SCM type and loading, aggregate gradation, water-to-binder ratio, and admixture dosages. The output is a set of predicted property values for that formulation.
The platform predicts properties across fresh-state behavior, hardened mechanical performance, and durability — covering the range of characteristics that determine whether a mix will pump, extrude, build, and meet structural requirements. A researcher can enter multiple formulation variants and compare predicted outputs before selecting candidates for physical trial.
Model Accuracy
The models are trained exclusively on 3D-printed specimen data drawn from open-access 3DCP research, including the RILEM interlaboratory study on 3D concrete printing mechanical properties. The training dataset is continuously updated as new research is published and validated. The models are not adapted from general concrete databases.
Cross-validated performance varies by property, but the models consistently explain the majority of variance across held-out test records for their trained targets. That said, prediction reliability depends on how close a given input formulation is to the training distribution. Formulations at the edges of or outside the covered design space will carry higher uncertainty.
A Realistic Workflow
Consider a researcher developing a Portland cement–fly ash printable mortar. They have a target compressive strength range and a print speed that constrains their open time window. Rather than running a full factorial trial matrix across multiple water-to-binder ratios and SCM loadings, they enter each candidate formulation into CEMFORGE and review the predicted fresh and hardened property values.
Formulations that fall outside a workable fresh-state range are set aside. The trial matrix narrows significantly. Physical testing still happens — but it is focused on the candidates most likely to perform, rather than the full combination space.
This is the use case the platform is built for: reducing the number of physical iterations, not replacing them.
Tiered Access
Current plans and pricing are listed on the subscriptions page. A free tier is available as a functional starting point — not a demo. A researcher can screen compressive strength across formulation variants without a subscription. Paid tiers unlock additional predicted properties, full dataset access, and advanced features.
What CEMFORGE Cannot Do
Physical testing cannot be skipped. CEMFORGE predictions are model outputs based on training data — they carry uncertainty, and that uncertainty increases for formulations that sit at the edges of or outside the training distribution. Geopolymer binders are not well represented in the current training set and are not modeled.
The platform does not account for printer geometry, nozzle configuration, delivery line length, or environmental conditions during print. It predicts material properties from composition, not print outcomes from equipment parameters. Setting time predictions are for the mix in isolation, not for a specific print environment.
Predictions are a starting point for trial design. Physical characterization — fresh-state testing, print trials, specimen casting, and mechanical testing — remains the standard for validating a mix design before production use.
About the Platform
CEMFORGE is developed by Sunnyday Technologies, based in Wisconsin, USA. The training dataset is built from open-access 3DCP research and updated as new data becomes available. The platform is designed for engineers and researchers working with printable cementitious materials.
Mix design screening for 3DCP starts at cemforge.ai.