Application of machine learning techniques to predict the unconfined compressive strength of sustainable cementitious materials used in the mining industry

  • Author / Creator
    Balasooriya Arachchilage, Chathuranga S J
  • Various forms of cementitious materials, including shotcrete, grouts, and cemented paste backfill (CPB), are made with ordinary Portland cement (OPC). They are widely used for both underground and surface mining applications. However, due to the high carbon footprint of OPC production, the mining industry has shown a significant interest in using sustainable cementitious materials. Alkali-activated slag (AAS)-based CPB and calcium sulfoaluminate cement (CSA)-based mixtures have emerged as promising sustainable cementitious materials for different mining applications, owing to their reduced carbon footprint and other advantages (i.e., high early age strength, improved durability, and reduced energy requirements) over OPC-based mixtures. For mining applications, it is required to ensure that the unconfined compressive strength (UCS) of these cementitious materials meets the strength requirements. The UCS of both AAS-based CPB and CSA cement-based mixtures are influenced by multiple features related to cement composition, material proportioning, curing conditions, and admixtures. Currently, those mixture designs depend heavily on the experimental approach, which is usually limited to a limited number of influential features at a given time. This has resulted in an insufficient understanding of the non-linear relationships between multiple input features and UCS of the aforementioned sustainable cementitious materials. The insufficient understanding poses challenges in designing mixtures that can meet the specific strength requirement for mining applications. Compared with the experimental approach, machine learning (ML) methods can consider multiple input features simultaneously to build prediction models. ML is a promising alternative approach that can assist the mixture designs of cementitious materials efficiently by improving the understanding of complex non-linear relationships and by providing accurate and rapid UCS predictions. Despite its significance, to the best of the author’s knowledge, no studies in the current literature have reported the use of ML techniques for predicting the UCS of AAS-based CPB and CSA cement-based mixtures.
    Therefore, this study aimed to utilize ML methods to build complex relationships between multiple input features and the UCS, to predict the UCS of AAS-based CPB and CSA cement-based mixtures, and to help spread the applications of low-carbon cementitious materials in our mining industry. For these purposes, accurate strength prediction models were developed based on meaningful datasets collected from experimental literature. All ML models were evaluated for their performances on testing data using commonly used performance evaluation metrics. Results showed that the extreme gradient boosting regression (XGBR) model constructed on the optimal dataset suggested by the least absolute shrinkage and selection operator (LASSO) method produces the best results with a prediction accuracy of 95% for CSA cement-based mixtures. In addition, the feature importance ranking results revealed curing time, water-to-cement ratio (w/c), belite content, and ye’elimite content as the most influential features on the UCS of CSA cement-based mixtures. Furthermore, the Shapely-Additive exPlanations (SHAP) method could describe the non-linear relationships between input features and UCS, both qualitatively and quantitatively. For AAS-based CPB, gradient boosting regression (GBR) outperformed the other ML models with a prediction accuracy of 96.7 %. The curing time, w/c ratio, and aggregate-to-cement ratio were ranked as the most important input features for the AAS-based CPB mixture designs. Overall, this study will improve the understanding of complex non-linear relationships between input features and the UCS of AAS-based CPB and CSA cement-based mixtures and will guide future mixture designs through rapid UCS predictions. Ultimately, this work will help spread the application of sustainable cementitious materials in the mining industry and contribute to the goal of net-zero mining by facilitating more efficient mixture design processes.

  • Subjects / Keywords
  • Graduation date
    Fall 2023
  • Type of Item
  • Degree
    Master of Science
  • DOI
  • License
    This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.