.png){kind=icon}
.png){kind=icon}
Return
Lithology Identification and Lithological Interface Location from MWD Data in the Open-Pit Mine Blast Area Using Machine Learning
.png){kind=icon}
PRE
AI
Z
Q
X
X
Z
Y
Z
DOI：10.1007/s00603-026-05823-2.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Acquisition of lithological information in blasting zones relies on engineer experience and geological exploration. This ambiguous information is identified as the primary cause of suboptimal fragmentation, secondary blasting, and associated additional costs. To address this limitation, a real-time lithological identification model, termed the MWD-LI framework, was established. Utilizing existing drilling parameters, the MWD-LI Framework identifies lithology around boreholes and locates lithological interfaces. High-resolution lithological information at the blast-hole level is provided, enabling precision blasting designs, such as rational deck charging. The MWD-LI framework includes three main stages: data preprocessing, lithology identification, and lithological interface location. Data preprocessing is primarily executed through SMOTE and the Isolation Forest algorithm. The lithology identification module integrates BOHB, CatBoost, and an adversarial training network. To ensure robustness and generalization performance, leave-one-out cross-validation is implemented during the training process, thereby establishing a stable foundation for the subsequent precise location of lithological interfaces. Field-acquired MWD data covering four lithological types, diverse drilling conditions, and multiple bit wear levels were collected for model training and testing. Field drilling operations were conducted to demonstrate the efficacy of the MWD-LI Framework. Experimental results demonstrate that Recall values exceeding 0.820 were achieved for each lithological type, while a MMCC of 0.829 and Macro-average Recall of 0.869 were recorded. Errors in lithological interface location were found to satisfy the precision requirements necessitated by explosive charging operations. Field validation confirmed the effectiveness of the MWD-LI Framework in addressing the aforementioned challenges, offering novel perspectives for ML applications in MWD technology and precision blasting research.
Keywords:
Measurement while drilling
Machine learning
Drilling
Blasting
Lithology
Lithological interface
Journal
IF:
6.6
Papers:
6.0K
Citations:
3.0W
Organization
Cited Papers