In the banded iron formation-hosted iron ore deposits in the Hamersley Range of Western Australia, the stratigraphic boundaries are generally modelled using data from exploration drilling. Accurately identifying the locations of boundaries is important when modelling and mining ore deposits. Exploration drilling typically has a coarse horizontal spacing (~ 50 m), resulting in inaccuracies in the modelled boundaries on a bench scale. Measure-while-drilling (MWD) data from blast holes has a much denser spacing (~ 5–7 m), and can be used to locally update boundaries with a finer resolution. However, MWD measures the relative effort required to drill through the rock, and this does not directly correlate to the boundaries. Also, MWD measurements can be affected by factors such as blast damage, end-of-hole effects, different equipment, equipment wear, and settings and drill operators. Therefore, an accurate method of identifying the boundary points in the MWD was required. MWD data produces a signal that is significantly different from those in geophysical logs, without areas that can be clearly related to a single rock type, and the data is significantly noisier. Identifying boundary points in the MWD required initial preprocessing and cleaning of the data, including thresholding and applying a continuous wavelet transform. The cleaning allowed a boundary to be more clearly observed visually and reduced the number of incorrect boundary identifications. A Gaussian process (GP) model was trained to successfully identify the transition between the unmineralised West Angelas Shale and mineralised ore. The MWD-based boundary points were then used to train a surface GP to fit the boundary. When boundary points from multiple blasts were used, the GP surface produced a reasonable surface with more detail than the initial, exploration-based surface. Therefore, the GP surface can potentially be used to update geological boundaries in the deposit model to increase their accuracy within localised areas.

在西澳大利亚州哈默斯利山脉的带状铁矿床铁矿床中，地层边界通常使用勘探钻探数据建模。在对矿床进行建模和开采时，准确识别边界位置非常重要。勘探钻井通常具有较粗的水平间距（〜50 m），从而导致工作台规模上的建模边界不准确。爆破孔的随钻测量（MWD）数据具有更密的间距（〜5–7 m），可用于以更精细的分辨率局部更新边界。但是，MWD会测量钻穿岩石所需的相对工作量，而这与边界没有直接关系。此外，随钻测井的测量值可能受到以下因素的影响，例如爆炸破坏，井眼效应，不同设备，设备磨损，和设置以及演练操作员。因此，需要一种准确的方法来确定随钻测井中的边界点。MWD数据产生的信号与地球物理测井中的信号明显不同，没有明显与单一岩石类型相关的区域，且数据噪声较大。识别MWD中的边界点需要对数据进行初始预处理和清理，包括阈值化和应用连续小波变换。清洁使得可以在视觉上更清楚地观察到边界，并减少了不正确的边界标识的数量。训练了高斯过程（GP）模型，以成功识别未矿化的西安吉拉斯页岩和矿化矿石之间的过渡。然后，基于MWD的边界点将用于训练曲面GP以适合边界。当使用来自多个爆炸的边界点时，GP表面产生的合理表面比初始的基于勘探的表面具有更多的细节。因此，GP表面可以潜在地用于更新沉积模型中的地质边界，以提高其在局部区域内的准确性。