Prediction of concealed faults in front of a coalface using feature learning

The existence of concealed faults not only decreases the production efficiency of a coal mine but also wastes resources and increases the risk of mine disasters. In this study, a method was developed to predict concealed faults in front of a coalface. The spatial distribution law of faults developed in the study area was characterized using the locations and attributes of fault zones, which can be determined by learning the strikes and locations of the faults with the K-means algorithm. Then, the concealed faults in front of coalfaces can be predicted by extending the fault zones along their strikes to unmined areas within the study area. Three attributes of fault zones, including extending index, buffer radius, and average throw, were defined and calculated to provide a quantitative evaluation of prediction results. The extending index represented the existence probability of the predicted fault. The buffer radius denoted the possible offset of the actual exposure point relative to the predicted location. The average throw gave the throw of the predicted fault. The method could also provide dynamic prediction as mining works were going on. Finally, the method was applied in mining region 302 of the Yanzishan Coal Mine in north China, and it was illustrated to be effective. In the test, the faults successfully predicted accounted for 82%, 89% of which was located within the range of buffer radius and also 89% had throw errors less than 50%.