Freeze-Thaw Damage Evolution of PVA-Fly Ash-Slag Composite Concrete and Low-Dimensional Mapping of Weibull...

To investigate freeze-thaw damage evolution and mix proportion effects in PVA fiber-fly ash-slag powder composite concrete (PVA-FA-SPC), nine composite concrete mixtures were designed using an L₉(3³) orthogonal array at a water to binder ratio of 0.45, with plain concrete serving as the reference, and subjected to 200 rapid freeze-thaw cycles. Freeze-thaw resistance was evaluated using surface deterioration, the mass loss rate, and the relative dynamic elastic modulus. Within the investigated factor levels, PVA fiber volume content exhibited the strongest main effect trend, followed by the total mineral admixture replacement rate and the fly ash to slag powder mass ratio. T20R1:2P0.3, containing 20% total mineral admixture replacement, a fly ash to slag powder mass ratio of 1:2, and 0.3% PVA fiber, showed the best measured performance, retaining a relative dynamic elastic modulus of 79.18% after 200 cycles. A two-parameter Weibull function was used as a phenomenological description of the damage evolution; compared with the classical exponential model, the average RMSE decreased from 0.032 to 0.019 and the average MAPE decreased from 5.07% to 2.98%. Because only nine independent orthogonal mixtures were available for parameter mapping, a common shape parameter of α₀ = 2.1403 was adopted, and a parsimonious equation for the scale parameter β was selected using the small-sample-corrected Akaike Information Criterion (AICc) together with leave-one-mixture-out cross-validation (LOMO-CV). LOMO-CV yielded R² = 0.886, RMSE = 0.034, MAE = 0.025, and WMAPE = 19.98%; however, T20R1:2P0.3 exhibited a node WMAPE of 72.75%, indicating a local limitation of the reduced order mapping. Three non-orthogonal mixtures within the same material system yielded R² = 0.943, RMSE = 0.026, and WMAPE = 15.36%. The proposed model is therefore intended for local trend analysis and preliminary mix screening within the calibrated material system and parameter range rather than for universal service life prediction.