In mining, the mixing of extracted minerals with waste materials or materials with a grade lower than the limit grade is called dilution. It is promoted to reduce the grade of the mineral and increase the extraction costs. Title is known as a disturbing factor and hence reducing it is one of the ways to increase product quality and thus reduce costs and increase profitability. Due to the variety of parameters affecting the dilution as well as the variety of underground mining methods, it is impossible to provide a comprehensive solution for estimating the dilution that applies to all mining methods. But in recent years, due to the importance of dilution in underground mining, extensive research has been done in the field of prediction of dilution, and methods that are often developed considering limited parameters for use in one or more specific methods. In this thesis, by collecting the data of the open stope underground mines (such as the method of shrinkage stoping, Sublevel stoping and cut and fill) from reliable sources, using the given methods, models have been developed to estimate dilution in these stope. The main goal of this thesis is to develop new models to estimate the dilution in open stopes in such a model that the result can be presented in the form of a relationship or diagram that can easily predict the dilution rate. Another goal of this thesis is to identify the importance of input parameters in the progress event. For this purpose, have been used a database containing 100 data sets as training data and 20 data sets as test data. This database consists of numerical modified stability number (N′), hydraulic radius (HR), average borehole deviation (BD), powder factor (PF) and equivalent linear over break slough (ELOS). Using regression models, four models (multiple linear regression (MLR), gene expression programming (GEP), multivariate adaptive regression spline (MARS) and classification and regression tree (CART)) in order to estimate the amount of ELOS and also using classification models, five models (logistic model tree (LMT), logistic regression (LR), J48 tree, MARS and CART) have been developed to predict ELOS (determine ELOS class label). In order to eva‎luate the developed models and determine the most appropriate model in each of the regression and classification models, both the given the test data and the training data are used. It was determined that among the four advanced regression models, "CART" model with a coefficient of determination of 0.85 and also based on the "GEP" model test has a better performance and the input centers N' and BD are at the highest and higher levels. Development of these models. Also, among the five developed classification models, the "LMT" training software model has a good performance with an accuracy of 0.92, as well as based on the test of the "CART" model with an accuracy of 0.75. In this model, by using this research and using features such as the geometry of the stopes and the quality of stones around the stope, it is possible to estimate the dilution.

ابراهيم قاسمي

امين ازهري، احمدرضا مختاري