School of Civil and Environmental Engineering 

Ph.D. Thesis Defense Announcement 

Accelerated assessment of candidate supplementary cementitious materials using statistics and machine learning

By Renee Rios

Advisor: 

Dr. Kimberly Kurtis

Committee Members: Dr. Emily Grubert (CEE), Dr. Iris Tien (CEE), Dr. Yao Xie (ISyE), Dr. Newell Washburn (Chem - CMU)

Date and Time:  Friday, March 11, 2 pm

Location:  Mason Building Room 5134 / Online

Complete announcement, with abstract, is attached. 

The supply of the most traditional supplementary cementitious materials (SCMs) used in concrete is not expected to keep up with demand in the coming years. To combat this problem, a wider range of both naturally occurring and artificially made SCMs are beginning to be used in cement blends, but each potential SCM comes with its own set of challenges and advantages. The overarching theme of this work is to accelerate the quality assessment of these potential SCMs for use in concrete using data analysis techniques, in particular statistics and machine learning. Four novel applications are presented in this work whose outcomes provide performance properties related to concrete durability more rapidly and robustly than the existing paradigms provided by ASTM C618 and Departments of Transportation. Time-series surface resistivity (SR) measurements are proposed as potential test methods for use as an indicator of reactivity in concretes containing materials of both known and unknown reactivities cured at standard and accelerated temperatures. Two statistical techniques, change-point detection and slope analysis, are applied for the first time to the time-series data to allow a comprehensive analysis of a material's reactivity. Accurate predictive models using machine learning to predict a given mix's SR values are produced based on potential SCMs' physical and chemical characteristics and early-age SR measurement data. Machine learning was further used to create a model to predict another performance property, ASTM C1567 14-day normalized expansion results, as an indicator of a material's susceptibility to the alkali-silica reaction.
Lastly, a traditional mainly prescriptive concrete design was compared with three new concrete designs which do not comply with current local specifications. A framework is proposed to statistically analyze these designs' compressive strength data to pre-qualify mix designs, which can be broadly applied to reduce time consuming iterative testing and to help meet sustainable development goals.
Recommendations are provided to accelerate widespread adoption of machine learning and statistics for use in the concrete industry.