Machine learning and multi-objective optimisation of sustainable concrete incorporating seashell powder and construction and demolition waste

Cement manufacture and river-sand extraction are among the largest environmental burdens of concrete production, while seashell and demolition wastes accumulate unused. This study evaluates M20-grade concrete in which ordinary Portland cement is replaced by seashell powder at 5, 10, 15, 20, 25 and 50% by weight, with 10% of the fine aggregate replaced by processed demolition waste. Compressive strength, split tensile strength and rebound number were measured at 7 and 28 days on three specimens per age. Across the series, the 28-day strength varies by only 2.72 MPa; every mix exceeds the target mean strength for the grade, and cradle-to-gate global warming potential falls by 38.6% between the lowest and highest replacement level; strength is therefore not the property limiting how much cement may be replaced. Ten learning algorithms were trained on a design-space corpus calibrated against the measured strengths, the twelve laboratory records being held out for validation; a stacked ensemble reached a coefficient of determination of 0.953, and interpretation ranked replacement level only third among the predictors of strength, behind water-to-binder ratio and curing age. The binding constraint at high replacement is the effective water-to-cement ratio, which almost doubles across the series, and durability rather than strength.