Improving rock hazard prediction by detecting defects using infrared thermography and machine learning: A novel risk mitigation approach

Abstract

This study explores the use of infrared (IR) emission monitoring combined with machine learning and deep learning techniques to assess the behavior of rock masses under both loaded and unloaded conditions. The research presents a significant step forward in the application of non-invasive and remote techniques for rock mass assessment. The ability to monitor the thermal IR emissions of rock masses in real-time provides a unique advantage in both laboratory and field environments. Unlike the traditional methods, which require physical contact with the rock mass and can be disruptive, the IR monitoring approach is non-invasive and can be performed continuously without interrupting the normal operation of mining or construction processes. In order to make this technique more robust, various machine learning models were incorporated which helped to identify the pattern of cracks and discontinuities. A decision tree algorithm was employed to classify the rock mass based on the responses of the IR data. The study also introduces a novel approach for the digital quantification of the Geological Strength Index (GSI) using IR data, enhancing the ability to assess rock mass strength and behavior without direct physical contact. Additionally, deep learning techniques integrated with the IR emission data in this study facilitated more accurate modeling of complex patterns of anomalies appear in the IR images during loading condition. These models, particularly convolutional neural networks (CNNs), were used to identify complex patterns in the IR images and timeseries data. CNNs are particularly well-suited for image-based data, and their ability to automatically learn spatial features from raw data allowed for the extraction of valuable insights regarding the distribution of thermal emissions during deformation. For assessing rocks at loaded and operational condition, different loading configuration, including uniaxial loading and bending loading, were selected to evaluate the rock mass response. The dissipated energy during these loading conditions was quantified through the analysis of IR characteristics, providing insights into the mechanical behavior of the rock mass. The results demonstrated the potential of integrating IR emission monitoring with advanced machine learning models for non-invasive, real-time monitoring of rock mass properties, contributing to improved engineering applications in geomechanics and mining. This makes it a highly promising tool for monitoring the behavior of rock masses in situ, where access is limited or the environment is hazardous.