Introducing Version 6: Tiger

Back to Overview

Graph Tracing - Oil and Gas

Graph Tracing - Oil and Gas


Digitizing Depth and Drill Records for Geo-modeling and Exploration.


  • Custom Models
  • Image Inference
  • Dropbox to store and retrieve images
Graph of the telemetry coming from drill head.


Data is essential to Oil & Gas drilling industry: depth, geologic layers, hardness, heat, etc all give clues into guiding drills toward successful finds. A major industry technology provider approached Iterate with a unique problem: the oil drilling machines in the field produce an ongoing record of the telemetry coming from the drill head, but that data is produces as an analog printout (looking not unlike an EKG):

To make this data work at scale, and in order to form a complete 3D model of the geologic layers in the area, our partner needed to digitize all of these printouts back into some digital form-- not just scans of the documents, but actually getting all the data as numeric data from the graphs.

Our data engineers trained several different Machine Learning (ML) models to do the following:

  1. Automatically learn to read the scale of each graph (e.g. what the x and y coordinates mean)
  2. Learn to find the undulating line of data within each graph (while simultaneously learning to ignore any printing chaff)
  3. Mathematically calculating and recording each linear graph into a series of numerical data points (i.e. something like a Fourier Transform on a massive scale)
  4. Pass these massive data tables to a common cloud storage
  5. Deploy Interplay to take those newly created data stores and produce data visualizations and reporting as requested

We tackled this problem with zest, and were able to produce a working app in a matter of a few weeks. Here’s the Flow of how we put this app together:

This project empowered our partner to not only digitize an ongoing operational need, it provided a great historical record which can further allow the end clients to synthesize larger trend data.

If you have a similar data problem, let us know.

Learn More