Maturity-based prediction of UHPC early-age strength: systematic assessment and a new piecewise fitting approach

Ultra-high-performance concrete (UHPC) offers exceptional early-age strength, making it ideal for accelerated bridge construction. However, accurately predicting UHPC strength at early ages remains challenging, particularly when using the ASTM C1074 maturity method, which was developed for conventional concrete and has not been systematically validated for UHPC. This study systematically evaluates the applicability of ASTM C1074 for very early-age ( \(\le\) 1 day) and early-age (1–3 day) strength predictions of both proprietary and non-proprietary UHPC mixtures, using eight different mixtures and 620 specimens (cubes and cylinders). Temperature histories were collected using embedded wired and wireless sensors, and strength–maturity relationships were established using both the Nurse–Saul and Arrhenius functions. Results show that ASTM C1074 can yield significant prediction errors, often exceeding 100%, at very early ages ( \(\le\) 1 day), though it showed excellent performance for early ages (1–3 day). To address this, an Iteration Search Method (ISM) was developed to derive improved recommendations within the ASTM C1074 framework for UHPC, including practical calibration age configurations and maturity constants that reduce early-age prediction errors. In addition, a novel Piecewise Linear Method (PWLM) based on segmented maturity index zones was proposed and validated. PWLM improved performance under a strict independent evaluation protocol; for example, the EQ cube median overprediction across the two early-age windows ( \(\le\) 1 day and 1–3 days) decreased from about 139% to about 16%, and the NS cube median absolute error across the same windows decreased from about 110% to about 25%. These findings indicate that PWLM is a promising sensor-compatible framework for real-time UHPC strength estimation in time-sensitive applications, provided that mixture-specific validation is performed before field implementation.