Even though it has a great progress in rock mechanics research, we don 't have enough knowledge in solving and processing complex rock engineering problems because of comprehensive effects of geological, water, and construction, etc. Therefore, It is much attractive way to automatically extract knowledge from engineering cases and experimental data. The paper reviewed several effective methods proposed recently, such as data mining, learning based on neural network, determining unknown parameter values using genetic algorithm, search of unknown structure of mathematical model in global optimum space using genetic programming. Applications of these methods to assess risks of rock burst at great depth mining recognize slope stability model, recognize nonlinear constitutive model, and back analyze rock mass mechanical parameters for permanent shiplock in Three Gorges Project are also given.


The stability of rock engineering is affected by many factors such as geological conditions, goo-stress, raining, underground water, excavation scheme, earthquake, and blasting, etc. We don't have enough knowledge to analyze and control the stability of engineering. Therefore, it is an attractive way to learn knowledge from case histories. However, how to obtain an optimum solutions in global space is key question. In many cases, it is very difficulty to know accurately structure of models in advance. Therefore, auto-recognition of the structure of the model is another key problem. Some effective methods have been proposed to automatically extract knowledge. Up to representation schemes, the auto-extraction methods are divided into learning of uncertain inferring knowledge, learning of mathematical or mechanical model, and learning based on neural networks. This paper summarized these learning methods and gave some application examples to rock engineering.


A new data mining method is initially proposed to discover associated rules used for uncertain referring in expert systems for Rock engineering problems. An associate rule is written as: The rule X Y exists in the trade set D, and S% trade in D contains xnY, C% trade containing X also contains Y. Discovering associate rules is to automatically extract a depending relationship satisfying threshold of support degree and believing degree. The algorithm for discovering associate rules is described as: Step 1: Divide data of each attribute in case data base into several intervals. Step 2: Determine a frequent set whose length equals I from case database. With this algorithm, a group of associate rules was discovered from all selected frequent set for stability and failure case histories of underground opening and slope, and rock bursts. The discovered knowledge was used to build the expert systems for engineering applications. They were/are applied to safety analysis of the field. As an example, some rules have been discovered for assessing rock burst risks at great depth mining. The discovered knowledge was used to assess risk of rockbursts Induced by mining at depth of more than 1000 m at several gold mines. The accuracy is higher than 94%.

Learning nonlinear relationship among many factors.
This content is only available via PDF.
You can access this article if you purchase or spend a download.