Abstract
Time-series surface resistivity (SR) measurements with up to 56 days of hydration on concretes containing potential supplementary cementitious materials (SCMs) cured at an ambient temperature of 23°C and an elevated temperature of 38°C have the potential to indirectly determine the reactivity of many of these SCMs more accurately than conventional indirect testing metrics, such as the strength activity index in ASTM C618, Standard Specification for Coal Fly Ash and Raw or Calcined Natural Pozzolan for Use in Concrete. SR time-series measurements can distinguish between microstructural densification caused by inert materials and densification caused by pozzolanic materials. However, it may be possible to assess pozzolanicity earlier in the test series, particularly when additional physical and chemical data regarding the material are known. The aim of this paper is to develop predictive models using machine learning on a broad range of both ASTM C618 conforming and nonconforming materials’ time-series SR curves based on the materials’ physical and chemical characteristics and early SR measurement data. Gaussian process regression models were used to predict the SR values of concretes cured at two different temperatures. These models can rapidly screen materials based solely on their chemical and physical characteristics to predict SR curves, which can then be used to determine a material’s suitability for beneficial use in concrete. Models using early age SR measurements were produced to predict the 56-day SR value for concretes cured at both elevated and nonelevated temperatures. This data-driven approach allows for the design of novel reactive materials, the inclusion of existing reactive materials in construction, and a reduction in testing durations.