AI-Boosted Geological Facies Analysis from High-Resolution Borehole Images
- Shiduo Yang (Schlumberger) | Yinyu Wang (Schlumberger) | Isabelle Le Nir (Schlumberger) | Alexis He (Schlumberger)
- Document ID
- Society of Petrophysicists and Well-Log Analysts
- SPWLA 61st Annual Logging Symposium - Online, 24 June - 29 July, Virtual Online Webinar
- Publication Date
- Document Type
- Conference Paper
- 2020. held jointly by the Society of Petrophysicists and Well Log Analysts (SPWLA) and the submitting authors
- 91 in the last 30 days
- 91 since 2007
- Show more detail
Borehole images provide many different texture features for facies analysis and natural fracture identification. However, classification of most of these features is achieved manually. The workflow proposed here is to implement geological "facial recognition" from borehole images and other petrophysical measurements. The image segmentation is the first step to split the geological "facies" from continuous borehole images. Then the clustering based on texture similarity and petrophysical measurements is the second step to major facies categories. The major facies categories are labeled manually, and a deep learning model is trained to recognize geological facies on new borehole images in the same reservoir.
A borehole image can be visually recognized as a composition of successive zones; different zones have different statistical properties, which can be used to characterize the image and generate the zonation. The continuous histogram and variogram derived from image data are used for image segmentation. From the highresolution borehole images, the segments obtained are numerous enough to perform what is known as unsupervised classification. Among various methods of unsupervised classification, we choose to use the mean shift algorithm for the automatic clustering. It is a deterministic process, which is suitable in determining the number of clusters. The segments are assigned as facies with a local geological setting, then structured and formatted to build a library of multimodal data (image data and petrophysical log data) for a given facies. A deep learning model is trained to associate multimodal data to a given facies. This model is used to identify automatically the image features of another borehole, for continuous facies analysis in similar depositional environments.
We demonstrated this workflow in different depositional environments. Twenty-four facies were recognized from a water-based mud image in a braided river environment from China compared with 14 with core description Twelve facies identified from an oil-based mud image in a lacustrine system from the United States were then applied to another water-based mud image in the same reservoir with the deep learning model. The results from this approach were verified after comparing with a manual interpretation from cores.
Microfacies identification is the fundamental information for sedimentary analysis. The drilling core data is commonly used for microfacies classification. Because of the high cost of coring, the high-resolution borehole image prevails for facies or facies association analysis over that with core calibration. There are lots of successful case studies on facies modeling or deposition environment identification from high-resolution borehole images in combination with other logging technologies or seismic images (Lawrence et al., 2003; Blount, 2017). However, for most of these studies, the sedimentary facies identification was achieved manually and required significant effort.
|File Size||2 MB||Number of Pages||13|