Our Technology
How the CEMFORGE 3D Concrete Printing Mix Design Engine Works
A multi-stage formulation pipeline combining physics-based particle packing, competitive multi-model machine learning, and real additive construction specimen data.
Every supported CEMFORGE 3D concrete printing mix design output is generated through the currently validated data and model pipeline. Additional optimization modules are being matured through early-access validation before being represented as production-complete.
The six stages below describe the full sequence from aggregate evaluation through to export-ready documentation. Together, they form a 3D concrete printing mix design system built for engineering rigour, not marketing claims.
// 01 Modified Andreasen & Andersen Particle Packing
CEMFORGE is designed to evaluate aggregate gradation against particle-packing targets. Current outputs should be treated as decision support and paired with physical validation before construction use.
This step prevents the engine from producing a formulation that cannot physically close volumetrically. If the aggregate package does not pass packing validation, the formulation does not proceed. RILEM packing standards inform the acceptance thresholds used at this stage.
// 02 Competitive Model Selection
The 3D concrete printing mix design prediction engine trains and evaluates property models where sufficient validated data exists for each target property.
Model selection and exported property coverage depend on available validated samples and holdout performance. Supported outputs should be reviewed with their model-performance context.
This is a fundamental departure from black-box AI tools that apply one model to everything. See the About page for more on our data quality standards.
// 03 Printability Optimization
Printability checks consider accelerator, superplasticizer, viscosity modifier, set-time, and paste-volume factors where the current data and model coverage support them.
The 3D concrete printing mix design output accounts for open time targets against your specific job site conditions — not just a generic slump value that means nothing to a pumping system. Superplasticizer dosing is matched to printability windows established in the training data from real printed specimens.
Map the formulation that meets your open time target for tomorrow’s weather, not just your strength target.
// 04 Additive Construction Training Data
CEMFORGE trains exclusively on validated additive construction concrete data — real formulations from printed specimens, not conventional concrete adapted for prediction. The database spans OPC-based, LC3, UHPC, and fiber-reinforced systems.
Every record in the training corpus includes process metadata beyond mix proportions: extrusion parameters, layer geometry, nozzle configuration, fresh-state rheology measurements, and curing conditions. The training schema includes dozens of features — the most complete digital twin representation of the 3DCP process in any available formulation dataset.
Data quality is evaluated against experimental standards established by RILEM and ACI before any formulation enters the training pipeline. Theoretical or unvalidated mix designs are excluded entirely.
More controlled data. More measured variables. More accurate predictions.
// 05 Carbon Footprint Modeling
Carbon analysis is under active development and may be provided as an estimated or experimental output where the supporting material data is available. Carbon outputs should be treated as planning estimates until independently verified for the specific material supply chain and project.
This stage runs against the same validated additive construction material library used for formulation predictions. Emission factors are sourced from published EPD data for materials in the active 3DCP material set. The output is a decision-support number — not a certified environmental product declaration.
// 06 Complete Output Traceability
Supported CEMFORGE outputs include the input fingerprint, material selection rationale, model-performance metrics, and predicted properties where available – giving engineering teams an exportable record for validation planning.
Outputs are formatted for direct use on the print floor and as supporting documentation for research reporting. Pro-tier outputs include all exported properties — the set of available predictions grows as the validated training database expands and new properties meet export thresholds.
Capabilities at a Glance
AI-Driven Formulations
Custom 3D concrete printing mix designs fine-tuned for your specific materials, equipment, and performance targets.
Extrusion-Optimized Mixes
Formulations designed for extrusion-based deposition — optimized for layer adhesion, buildability, and open time.
Continuously Retrained
Models are retrained as the validated database grows. More data produces more accurate predictions across a wider range of formulations.
Property-Targeted Design
Define your target properties — strength, workability, set time, density — and the engine works within those physical constraints.
Physics-Informed Prediction
Combines Modified Andreasen–Andersen particle packing with ML predictions trained on real experimental data — not black-box correlation alone.
Production-Ready Output
Every formulation includes volumetric validation, admixture dosing, model performance metrics, and export-ready documentation for the print floor.
Have questions about how CEMFORGE fits your formulation workflow?
Our team is ready to discuss your specific 3D concrete printing mix design requirements.