BEMIS in depth

Data-driven and self-calibrating mine model

BEMIS uses data driven model calibration, continuously and automatically inferring multiple key acoustic properties of the rock-mass, including velocity. The BEMIS model is built from observations in seismic data rather than being based on calibration blasting, manual inputs and artificial constraints, such as modeled cave shapes or lithologies.

As data comes in BEMIS assigns higher weights to data with low uncertainty and lower weights to data with high uncertainty, as it consumes the data for model calibration. The model stays up to date with changing rock mass conditions, and keeps improving, with time and more data.

Areas of a mine where there are high seismic activity levels, i.e., areas of key interest, the model will consequently be at its best. This is in contrast to the method of using calibration blasts, which will provide the best calibration where it is practical to blast.

Example of a high quality velocity model. Image number 1 (before), hypocenter location uncertainty is large and donut shaped given the sensor array layout and the use of a basic homogeneous velocity model. Image number 2 (after), uncertainty for the same event has been greatly reduced based on BEMIS model self-calibration.

Precision and Accuracy

Precision and Accuracy of event location are two key aspects of overall seismic system performance. For precision, BEMIS has a number of distinct benefits.

  • The ability to intuitively assess precision by visualizing the location uncertainty. With improved model calibration, precision will improve and uncertainties will be reduced. This is immediately visible for all events in the BEMIS seismic catalogue. Engineers will easily determine where precision is high, and where it can improve.
  • Actual event location precision. The BEMIS approach to seismic processing has been shown to locate events with precision at an entirely new level, as never seen before.

The precision benefit of BEMIS is shown in the figure below. To the left, a legacy processing solution with the corresponding velocity model, resulting in scattering of event locations and a blurred, fuzzy view. To the right, BEMIS processing and a self-calibrated velocity model, resulting in much more precise event locations, with distinct and visible clustering in a clear view, pivotal for its use in decision making.

For obvious reasons, good location accuracy is among the most sought after features of seismic processing. If processing does not locate events to where they actually happened, it will throw engineers off and disturb decision making. By benchmarking with known blasts, BEMIS has been shown to be four times more accurate than the leading legacy solution.

BEMIS also provides intuitive location accuracy assessment. By processing and including blast events in the seismic catalogue (clearly labeled) engineers can assess location accuracy easily by comparing the known locations of blasts to BEMIS location results. An example of this is shown below, where blast events are colored by origin time with older blasts in blue and more recent ones in green. This shows clearly how events are accurately located along the development of the drive.

Quality Controls and transparency

The BEMIS approach is to transparently show uncertainties for all estimates (e.g., event locations). Additionally, the BEMIS catalogue exposes all data for each event, allowing users to see location uncertainty, arrival time distributions, waveforms, travel time graphs and more, all within the same user interface. This allows for smooth QA/QC and makes routine investigations of events very easy, building confidence in automatic processing results and any downstream analyses and decision making.

The figure below shows the BEMIS mXrap Extension user interface. Users can easily investigate all aspects of event location. The example also shows processed blast events being compared to the known coordinates of the blast, a comparison that gives users an easy to use QA/QC tool, available on a daily basis.

Being able to show location uncertainties for all processed events is meaningful for users in multiple ways. For example, the figure below shows a mirroring effect for the selected event, which is a clear indication that the sensor array provides poor coverage in this area. It also shows that the velocity model does not have sufficient calibration to render high precision location in the area. Both of which are valuable pieces of information when making decisions in the mine.