Date of Award

8-17-2023

Document Type

Thesis

Abstract

Ground hazards are a major safety concern in underground mines, causing deaths, injuries, and lost work time to miners. Although ground hazards such as roof fall accidents have steadily and significantly declined over the last decade, the safety hazards associated with ground hazards still account for a significant portion of the total number of accidents in the underground mining industry. One of the main reasons behind ground hazards in underground mines is abrupt changes in geological conditions that may decrease the competence of the rocks surrounding an excavation, resulting in ground failure. The changes in geological can be detected in a mined-out area, and a possible roof fall accident can be avoided if prompt countermeasures are taken, but changes cannot be easily foreseen in unexcavated areas. An obvious solution to this issue would be to predict strata geological conditions in advance of mining, followed by risk assessment and design of a safe and cost-effective ground support system. The prediction of geological conditions plays a vital role in mine planning and design. It is significantly important for the mine workers' safety and could be economically attractive in terms of increased productivity and reduced production loss. However, predicting changes in geological conditions in unexcavated areas is a complex task, and it requires proper mine strata characterization and mine design. This research aims to address ground control issues in the mining industry by utilizing Artificial intelligence (AI) techniques to predict lithology and RQD based on the location input data ( x, y, and z coordinates). The accurate prediction of lithology and RQD is crucial for effective ground control management in mining operations. The methodology developed in this research involves the creation of a predictive model using various machine learning algorithms. The input dataset is comprised of drill hole x, y, and z coordinates, and the outputs are lithology and the RQD. The model was trained, optimized, and validated using a large set of drill hole data from an underground metal mine. The results show that the random forest (RF) model performed better than other models, with a prediction accuracy of 98%. The trained model was used to build an interactive-user interface. By inputing just any location x,y, and z coordinates within the mine, into the interface, authorized users can get output of lithology and RQD for that location. This research may help improve the accuracy and the efficiency of ground control management practices in the mining industry, as well as economically benefit the industry.

Handle

http://hdl.handle.net/11122/14567

Download

Share

COinS