Predicting geological and geomechanical rock properties using data analytics and multi-sensor core logging data

地质和岩土岩体特性的详细建模对于具有成本效益和安全的采矿作业至关重要。尽管矿山规划和设计的几乎每一个技术方面都有了改进，但缺乏足够和一致的测量数据来拟订详细的地质和地质技术模型仍然是一个挑战。为了模拟岩体性质的空间变化，降低矿山规划设计中的不确定性，需要更高质量和数量的地质岩土数据。岩心测井是获取地质岩土资料的一种基本方法。然而，人工岩心测井方法具有主观、耗时和不一致性等特点，大大降低了地质和岩土模型的可靠性。为了克服这些限制，需要新的岩石测量和分析技术来自动收集和分析数据。该研究项目旨在利用多传感器岩心测井系统的数据推进岩石地质和地质力学性质的预测。这将通过从非破坏性测试中收集多种岩石物理、机械、矿物学和结构特性，以及岩心样品的高分辨率图像来实现。然后，这些数据将用于开发机器学习模型，以预测地质和地质力学岩石特性。岩土工程钻孔的岩心样本将使用手动和多传感器岩心测井系统进行记录。将岩心样品的高质量图像与多变量岩心测井数据进行处理，形成多模态数据集。该数据集将用于地质和地质力学岩石特性的预测模型。监督机器学习技术将用于训练使用多模态混合数据的预测模型。该模型将用于岩性单元及其力学性质的自动预测和分类；利用图像分析检测和表征自然不连续面几何属性；利用数字岩心测井数据预测不连续面剪切行为；基于多参数数字岩心测井资料的岩体地质力学性质分类并探讨数据质量和数量对矿山设计岩土模型的影响。将数字岩心测井系统的预测地质力学模型与基于人工岩心测井数据建立的模型进行比较。研究结果有望改善岩石表征和分类；允许在矿山规划和优化中做出明智的决策；减少与不可预测的地面条件相关的时间和成本；并降低矿山开挖失败的风险及其相关费用。

A detailed modelling of geological and geotechnical rock mass properties is essential for a cost effective and safe mining operation. Despite improvements in almost every technological aspect of mine planning and design, lack of sufficient and consistently measured data for the development of a detailed geological and geotechnical model remains a challenge. A higher quality and quantity of geological and geotechnical data is required to model the spatial variation of rock mass properties and to lower uncertainties in mine planning and design. Core logging is a fundamental method in obtaining geological and geotechnical data. However, the manual core logging methods are subjective, time consuming and inconsistent, which can significantly reduce the reliability of the resulting geological and geotechnical models. To overcome these limitations, new rock measurement and analysis techniques that automate the collection and analysis of data are required. This research program aims to advance prediction of rock geological and geomechanical properties using data from a multi--sensor core logging system. This will be achieved through the collection of multi-variate rock physical, mechanical, mineralogical and structural properties from non--destructive tests, together with high resolution images of the core samples. The data will then be used to develop machine learning models to predict geological and geomechanical rock properties. Core samples from geotechnical boreholes will be logged using both the manual and the multi--sensor core logging system. The high- quality images of the core samples along with the multi--variate core logging data will be processed to develop a multi--modal dataset. The dataset will be used for predictive models of geological and geomechanical rock properties. Supervised machine learning techniques will be used to train predictive models using the multi--modal mixed data. The models will be employed to automatically: predict and classify lithological rock units and their mechanical properties; detect and characterize geometrical attributes of natural discontinuities using image analysis; predict shear behaviour of discontinuities using digital core logging data; classify geomechanical rock mass properties based on multi--parameter digital core logging data; and to investigate the influence of data quality and quantity on geotechnical models used for mine design. The predictive geomechanical models from the digital core logging system will be compared to the models developed based on manual core logging data. The results of the research are expected to improve rock characterization and classification; allow informed decision making in mine planning and optimization; reduce the time and cost associated with unpredicted ground conditions; and reduce the risk of mine excavation failure and its associated costs.