2. Before You Start: Model Quality

Detection quality depends on the resolution of the underlying model. For reliable rock mapping, aim for a Ground Sampling Distance (GSD) of roughly 0.4–0.5 cm (0.15–0.20 inches). Coarser models lose the fine surface traces the AI relies on; finer models capture more of the true discontinuity network.

5. Reading Orientation Data: Stereonet & Stereogram

This is where Rock Mass AI turns thousands of individual measurements into a structural picture you can actually use. Both plots take the same input — the orientation of every detected plane — and present it as a density of poles. They differ only in how that density is drawn.

The stereographic projection in one paragraph

Imagine the lower half of a sphere centred on a discontinuity. The plane and its normal (pole) pierce that hemisphere; projecting those piercing points onto a flat disc gives the stereonet. Strayos uses an equal-area (Schmidt) lower-hemisphere projection, the standard for pole-density work. Every plane is reduced to a single pole, so a whole rock face becomes a cloud of points whose concentrations reveal the dominant joint orientations.

How to orient yourself on either plot

•      Azimuth runs around the rim. North is at the top (0°), then clockwise: East 90°, South 180°, West 270°.

•      Dip is read radially. For a pole: a flat-lying plane has a vertical pole that plots at the centre; a vertical plane has a horizontal pole that plots on the rim. So poles near the centre = shallow planes, poles near the rim = steep planes.

•      The pole sits opposite the dip direction. A plane dipping to the east places its pole on the western side of the net (180° away).

The Stereonet (contoured pole density)

On the Stereonet, each plane's pole is plotted as a point and the density of poles is contoured — coloured from few poles per grid cell (light) to many (warm reds). In the Strayos report the legend reads as poles per grid, with a grid of about 3.6° in trend by 0.9° in plunge. Each contour maximum (each red bullseye) is a concentration of parallel discontinuities — a joint set. Count the maxima and you have counted the joint sets; read the position of each maximum and you have its mean dip direction and dip angle. Great circles for the cluster mean planes can be overlaid on top.

The Stereogram (binned polar density)

The Stereogram shows the same orientation data on a polar grid of cells. The disc is divided into angular sectors (azimuth) and concentric rings (dip/plunge), and each cell is shaded by how many poles fall into that orientation bin. It is the discrete, histogram-style counterpart to the contoured stereonet: there is no smoothing, so you can read the exact distribution per orientation range and pick out minor or secondary sets that contouring might blur together.

Figure 6 — The same rock face shown two ways. Left: the Stereogram, a binned polar density grid (each cell coloured by pole count). Right: the Stereonet, a contoured equal-area projection — black dots are individual poles, and the warm contour maxima mark the dominant joint sets.

From clusters to joint sets

Each contour maximum corresponds to one cluster in the Clusters panel, where Strayos lists the set's mean orientation. For example, Cluster 4 in a sample dataset reports a Dip Direction of 13°, a Strike Direction of 283° and a Dip Angle of 57° — a single steep plane describing an internal sloped crack, not the outer bench surface. Reading the plots therefore reduces to three moves:

1.   Locate the maxima — each is a joint set.

2.   Read each maximum's azimuth (around the rim) and dip (radial position) for its mean orientation.

3.   Cross-check against the cluster list, then carry those sets into blast design, slope and bench-stability, and fragmentation assessment.

6. How Rock Mass AI Works

The module builds its results in two stages, then derives orientation geometry from them.

Stage 1 — Crack-polygon detection

First, the AI detects crack polygons on the surface of the bench from the 3D textures. At this stage it cannot yet distinguish true joints from random surface cracks — it simply finds every break it can see.

Stage 2 — Clusterisation (plane fitting)

Next, a clustering algorithm fits the detected cracks to planes of common orientation. Cracks that fit the same plane are grouped into one set. The approach has two known limitations: it assumes the majority of detected cracks are formed by joints, and because every crack also lies on the bench-surface plane, that surface direction can be over-fitted.

The illustration below makes the second limitation concrete. There are clearly three distinct joint groups in the rock (left). But because all cracks also sit on the bench surface, every crack can be fitted to that single surface plane (centre, red). Strayos therefore down-weights the bench-surface group: if other planes fit the majority of cracks well, they rank higher as genuine joint sets even when they contain fewer cracks (right, grey planes).

Figure 7 — Three real joint groups (left); the over-fitted bench-surface plane that is down-weighted (centre, red); the genuine internal planes the AI favours (right, grey).

Calculation steps

•      Predict joints on the 3D textures.

•      Map the predicted 2D polygons onto the 3D model to obtain 3D polygons.

•      Fit planes through those 3D polygons.

•      Compute strike and dip for each fitted plane (strike = azimuth of the plane–horizontal intersection; dip = inclination from horizontal; dip direction = strike + 90°).

•      Plot each plane's pole on the stereonet as a red point.

•      Cluster the poles and approximate each cluster as a plane — the named planes shown in the stereomap.

So, what formats can we export?

•      CSV — per-feature orientation and dimension data from the Data Table.

•      DXF — cluster and feature geometry for CAD and design workflows.

•      PDF — the formatted Rock Mass Report, including stereonet, stereogram and cluster orientations.