In cement manufacturing, ensuring simultaneous compliance with compressive (CCS) and flexural strength (FS) requirements is challenging due to their divergent responses to shared inputs. Industrial variability (e.g., fluctuations in C 3 S content, grinding heterogeneity, and curing conditions) exacerbates this challenge, leading to high batch rejection (15–20%) and excessive clinker overdesign that contributes 7% of global C O 2 emissions. Traditional single-property models fail by treating CCS and FS in isolation, producing deterministic predictions that ignore uncertainty, and conflating material-driven (aleatoric) with model-induced (epistemic) uncertainty—precluding risk-aware decisions. To overcome these limitations, this study introduces the Multi-Property Cement Strength Estimator (MPCSE)—the first joint probabilistic framework to model CCS and FS as full distributions via multi-head Gaussian Mixture Models with shared latent representations. MPCSE delivers four innovations: (1) joint probabilistic modeling that captures cross-property dependencies (7.8% lower MAE for CCS and 9.3% for FS versus single-property baselines); (2) explicit uncertainty decomposition enabling targeted process control (e.g., tighter grinding for 36–52 μm particles reduces FS variability by 15%) and strategic data collection (e.g., low- C 3 A regimes, 95 % confidence-interval coverage on industrial data and enables 12–15% clinker reduction with approximately 25% fewer batch rejections through risk-aware rerouting of borderline batches, advancing sustainable cement production. • Joint probabilistic modeling of cement strength (CCS and FS) • Dual-strength compliance metric quantifies multi-property risk • Uncertainty decomposition guides process control to cut clinker by up to 15% • Reduces batch rejection by 25% via risk-aware quality control • Achieves >95% confidence interval coverage on industrial strength data
Islam et al. (Sun,) studied this question.