Automated Mineral Identification of Pozzolanic Materials Using XRD Patterns

Abstract

Use of pozzolanic materials in concrete has become increasingly popular due to their ability to improve long-term strength and durability. To better understand their reactivity and reaction mechanism, proper characterization of their chemical properties is required. X-ray diffraction (XRD) is a widely used nondestructive testing method that can identify the different constituent phases of pozzolanic materials. However, the conventional qualitative approach for XRD analysis can be challenging due to the significant peak overlap and an amorphous hump in the baseline of XRD patterns. To address this issue, this study proposes a deep learning–based automated method for accurately identifying the minerals in pozzolanic materials using XRD patterns. The proposed method involves the deep learning–based peak picking algorithm, establishment of mineral candidates, and automated mineral identification using an optimization process. The performance of the model was validated using eight pozzolanic materials belonging to four classes (slag, fly ash, ground bottom ash, and silica fume).