Creating a Digital Twin for Process Simulation and Modeling

To better diagnose the performance of a refinery, a specific refinery unit, a process plant, or a power-generating station, many companies have replicated the physical process by creating a software representation of the process at an offsite location, usually in the cloud, but it may reside in the corporate headquarters or at a regional office. The purpose of a digital twin is to have a mirrored copy of the process in a safe environment, where models and analytics can query the data and new online calculations can be developed and tested against this digital twin without affecting the actual plant process. By running digital twin data through different models, statistical programs, ML, and other analytics, technical people can detect if the plant is running in line with expected outputs, using engineering principled models, statistical data-based models, or other algorithms used to predict future behavior.

One way to create a digital twin is to simply transfer all sensor data collected via control systems, together with the calculated (derived) variables, so that a near real-time version is available for analysis. Additionally, the data can be updated in real time by transmitting real-time data from the EIDI to the digital twin, so that the digital twin mirrors values in the plant in near real time. Engineering models, ML algorithms, and other predictive analytics are run in this offline, safe environment. Any insights or predictions made by modeling software or predictive analytics can be transmitted back to the actual plant EIDI, so that operations people can visualize how the actual plant process is tracking relative to model predictions that are stored in the EIDI as future data.

The team is now looking at deploying soft sensors to augment data collected from their control systems to enhance ML, digital twins, and data analytics efforts, which they hope will

  • • Improve their process control performance;
  • • Evaluate and resolve process bottlenecks;
  • • Use predictive analytics to more accurately forecast equipment failures, for example, adding sensors to their critical rotating equipment enables more robust, accurate calculations; and
  • • Generate more useful real-time alerts using the PF-to-F curve strategy referenced in Chapter 6, for pumps and feeders that encounter issues, such as increasing vibration.
< Prev   CONTENTS   Source   Next >