Determination of cement quality via optical spectroscopy and machine learning-based image recognition

Abstract

Recent events in Colombia like the collapse of the Space building in Medellin, or the Chirajara motorway bridge have undermined public opinion generating a bad perception regarding the construction industry. Some of these problems were related to structural design flaws but also poor quality materials. To partially tackle this problem a novel instrument was designed to assess the quality of the cement by capturing inelastic scattering vibrational modes of the C2S, C3S, and CSH. By combining the spectral information with a machine learning model able to recognize the intensity levels of the spectral bands associated with a high quality cement it was possible to detect low quality samples. After the training process of the neural network, the tool successfully recognized 95% of the samples provided.