Contribution to the evaluation of influencing factors on the intensity of blast induced ground vibrations with the application of artificial neural networks

[J. J. Radisavljević]

153 str.

Rudarstvo / Mining

Doktorska disertacija predstavlja prilog vrednovanju pojedinačnih uticaja faktora seizmičkih potresa na brzinu oscilovanja tla izazvanih miniranjima, upotrebom pouzdanog sistema za prognozu intenziteta seizmičkih potresa. Pouzdan sistem, u ovom slučaju, bila je obučena i verifikovana veštačka neuronska mreža. Polazne osnove za razvoj modela bile su zakon oscilovanja tla i skup ulazno-izlaznih podataka miniranih serija na površinskom kopu Veliki Krivelj. Skup ulazno-izlaznih podataka sadrži vrednosti faktora seizmičkih potresa pojedinačnih minskih serija, rastojanje minskih serija do mernih mesta i izmerene brzine oscilovanja tla prilikom miniranja tih serija. Prvo je izvršena obuka i verifikacija veštačke neuronske mreže, čime je dobijen pouzdan sistem za predviđanje intenziteta seizmičkih potresa izazvanih miniranjima. Obučena i verifikovana mreža je korišćena za postizanje pomenutog cilja. Formirani su dijagrami zavisnosti između pojedinačnih faktora seizmičkih potresa i brzine oscilovanja tla. Zavisnosti su predstavljene jednačinama regresionih linija uz odgovarajuću vrednost koeficijenta determinacije R2. Zatim je kroz rad utvrđen i vrednovan uticaj svakog faktora seizmičkih potresa na brzinu oscilovanja tla, određivanjem vrednosti koeficijenta uticaja. Upoređenjem koeficijenta uticaja pojedinih faktora seizmičkih potresa dobijen je i njihov redosled uticaja na brzinu oscilovanja tla. Za uslove na razmatranoj lokaciji najveći uticaj ima rastojanje minske serije do mernog mesta (R), zatim maksimalne mase eksploziva inicirane u jednom nominalnom vremenu iniciranja (Qi) i maksimalna masa eksploziva inicirana u jednom vremenskom intervalu od 8 milisekundi (Q8ms). Slede srednje mase eksploziva u minskim bušotinama (Qbsr), koeficijent punjenja minskih bušotina (Kp), ukupne mase eksploziva u minskoj seriji (Qbuk) itd. Rezultat predstavljen ovom disertacijom je model dobijen kombinacijom terenskih istraživanja i sposobnosti neuronskih mreža da reše složenije problema koji uključuju brojne faktore od kojih zavisi ishod ili rezultat. Model je kasnije poslužio za formiranje algoritma primenljivog na povšinskim kopovima, koji u tehnološkom lancu imaju fazu bušenja i miniranja, uz neophodnost korišćenja izvornih podatka ulazno-izlaznog skupa.

The Ph.D. dissertation represents a contribution to the evaluation of individual effects of ground vibration factors on the peak particle velocity caused by blasting, using a reliable system for predicting the intensity of ground vibrations. The reliable system in this case was a trained and verified ANN (Artificial Neural Network). The starting bases for the development of the model were the propagation law, and the dataset of blasted patterns at the "Veliki Krivelj" open pit. The dataset contains the values of ground vibration parameters of individual blast patterns, the distance of blast patterns from the measuring points, and the measured peak particle velocities during the blasting of those patterns. First, training and verification of the artificial neural network were performed, which resulted in a reliable system for predicting the intensity of seismic vibrations caused by blasting. The trained and verified network was used to achieve the aforementioned primary goals. Diagrams showing the dependency between individual factors of ground vibrations and peak particle velocity were formed. Dependencies are represented by regression line equations with the corresponding value of the coefficient of determination R2. Then, the influence of each ground vibration factor on the peak particle velocity was determined and evaluated through the paper, by determining the value of the influence coefficient. By comparing the influence coefficient of individual ground vibration factors, their order of influence on the peak particle velocity was obtained. For the conditions at the considered location, the greatest influence is the distance of the blast pattern from the measuring point (R), then the maximum charge weight initiated within a nominal delay (Qi) and the maximum charge weight initiated within a delay of 8 milliseconds (Q8ms). After that, the average charge weight in blastholes (Qbsr), the drill holes charging coefficient (Kp), the total charge weight in a blast pattern (Qbuk), etc. The result of research through the dissertation is a model obtained by a combination of field research and the ability of neural networks needed to solve more complex problems that include numerous factors on which the outcome or result depends. The model was later used for the formation of the algorithm, which can be applied to open pits that have a drilling and blasting phase in the technological chain, with the necessity of using the original data of the dataset.

Brzina oscilovanja tla, ANN, koeficijent determinacije R2

Peak particle velocity, ANN, coefficient of determination R2

622.235:550.34.038(043.3)

Tekst