Institutional Repository
Technical University of Crete
EN  |  EL



My Space

Χρήση νευρωνικών δικτύων ή/και νευροασαφών συστημάτων για την εκτίμηση αποθεμάτων ποιότητας κοιτασμάτων και εύρεση βέλτιστων ορίων εκσκαφής

Nasai Risalnt

Full record

Year 2023
Type of Item Diploma Work
Bibliographic Citation Ρισάλντ Νάσαϊ, "Χρήση νευρωνικών δικτύων ή/και νευροασαφών συστημάτων για την εκτίμηση αποθεμάτων ποιότητας κοιτασμάτων και εύρεση βέλτιστων ορίων εκσκαφής", Διπλωματική Εργασία, Σχολή Μηχανικών Ορυκτών Πόρων, Πολυτεχνείο Κρήτης, Χανιά, Ελλάς, 2023
Appears in Collections


The grade and reserve estimation of a deposit and determining the ultimate pit limit is an extremely important process, as it is a key pillar of the design and planning of the open pit mine. The bestknown methods are the geostatistical method Kriging, the inverse distance methods, and geometrical methods. Finding the ultimate pit limit is important because overestimation may resultin increased stripping ratio, while underestimation can lead to the rejection of a part of the deposit of economic interest. The best-known methods of finding the ultimate pit limit are based on linear and dynamic programming, the floating cone and graph theory. Recent advances in the field of artificial intelligence offer the possibility of applying them to such nonlinear problems without simplifications and assumptions that conventional methods usually require.In this paper, it was studied the possibility of estimating the grade and resources of a copper deposit by developing the digital deposit model in the shape of blocks using neural networks and then determining the ultimate pit limits including economic elements and applying the “maximum flow” algorithm (Pseudoflow).Three different variants of neural networks were developed which all were feedforward networks with back propagation. In the one all the composite drilling samples were used for training with one hidden layer, the second one with one hidden layer and resampling of the training data to improve its efficiency since the first one had a significant underestimation of the high copper grade and finally one with two hidden layers and resampling of the training data. This neural network achieved satisfactory results and with shorter training time than the second one.The neural network results were compared with those of the Kriging and the inverse distance squared (IDS) methods and that they can identify areas of high copper grade that were not possible with Kriging and IDS methods. Economic data were then included in the results of the digital deposit model produced by the neural network, Kriging and IDS methods and the economic valuesof the blocks were calculated. Finally, the Pseudoflow algorithm was applied to find the ultimate pit limit. The results showed that although the differences in the number of blocks within the limits are not significantly different for the three methods, the differences between the economic value are significant as well as the average economic value per block. This difference in economic valueof the blocks in the ultimate pit limit is directly related to the grade estimation of each method.

Available Files