Machine learning for oil & gas engineering - the digital gold mine

Jeanette Anderson-Koh |  January 24, 2019

With advances in PLC and DCS technology, data collection in oil and gas process control systems has become cheaper and more plentiful in recent years, and this looks set to continue. Within our vast data stores lies a gold mine of information that can help us better control, diagnose, and understand our industrial systems.

This enormous potential has gone largely untapped due to the difficulties involved in translating the right mathematical tools to industry in an easy-to-use form. Particularly noteworthy in this respect are machine learning techniques that have, in fact, been advancing in parallel to this technology and have already proven to be a breakthrough in other applied sciences.

The right machine learning tools – tailored uniquely to address process control issues such as noisy data and feedback delay, and made simple enough to supplement everyday engineering work – could provide very cost-effective, scalable means of improving our systems. Stable control, reduced equipment wear-and-tear, and improved reliability, profitability, and sustainability would be within our reach.

An approach to harnessing machine learning in improving our process control systems is illustrated in the diagram below. Input and output data from the process is drawn from the PLC or DCS system and machine learning techniques are used to create a best-fit trend or model that captures the key characteristics of the process and the relationships between its inputs and output. This is summarized in simple equations that allow for integration of the model information with the process PID control, for rapid and real-time stabilization of the process.

The challenges of building best-fit models for process control data have always been:
1.  Simplicity

  • The mathematics of the model-building must be easy to use and understand.

2.  Real-world usability

  • Models must reflect the real-world properties of the process and capture its trends accurately.
  • Models must be able to be integrated into the means of process improvement – in this case, the PID control algorithm.

Conjecture System Identification has been built to address these challenges. It provides an intuitive interface for building best-fit trends from industrial process data, capturing process dynamics in terms of inputs, output, and time. Conjecture System Identification is an inroad to the digital gold mine of your industrial systems, allowing for a simple digital representation of these systems for better plant diagnosis and understanding, and improved automated control.

Conjecture System Identification is built and licensed by Conjecture Data Solutions. We are a startup focused on creating data-driven software solutions for engineering. Check out the rest of our website to see how our best-fit data trending is used to dramatically stabilize PID control, and how this can improve productivity and profitability for your business.