Model Predictive Control
Modern Model Predictive Control (MPC) methods were first developed in the petroleum refining industry in the early 1990’s. MPC creates a computerized cruise control system for your plant that attempts to match or exceed the performance of your best operator. Although petrochemical and refining are a core segment of our business, our model predictive control successes include ethanol plants, air separation, polymerization, batch, and mining and metallurgy
Soft sensors are a key technology in modern process control. The objective is to use readily available signals such as temperature and pressure to infer an intermittently measured value such as a lab quality test. Although predicting lab results is a common application, the same technology can be used to predict future process values, instrument errors, or imminent equipment failures. Our technology allows users to define several modeling approaches such as time series, state space, Kalman filters, neural networks etc. and the technology will automatically select the most reliable method for a particular operating regime.
Real Time Optimization
Traditionally, the job of process control has been to hold conditions steady. Modern model predictive control technology lets clients define a cost or profit function for their plant, and the controller will automatically nudge the plant operating point to a more desirable region. At a major Canadian oil company, the optimization layer had been proven to contribute over $5MM in additional annual profit to the client.
The focus of this presentation is to highlight the power and purpose of data reconciliation to help validate your measurement system before the data is used for production, yield, and loss accounting.
Advanced Process Control Project
This presentation describes the Advanced Process Control project lifecycle.
Industrial Plant Optimization
This presentation discusses some of the techniques used to optimize oil refining using advanced process control. How is real-time optimization implemented? Can the plant effectively be modeled using a subset of data?