Linear model-predictive control (MPC) techniques have been applied very effectively to large-scale continuous plants over the last three decades.

However the approach cannot easily be applied to high-value batch or semi-continuous processes with multiple products and frequent grade changes, such as polymer production.

One of the most exciting developments in advanced process control (APC) in recent times is viable non-linear model-predictive control for large-scale plants.

Nonlinear MPC

NL-MPC is now becoming a reality because of the availability of accurate, physics-based process models, the increased speeds and robustness of solution techniques, and the level and availability of real-time data delivered by recent digitalization initiatives.

NL-MPC setup for a polymer process

The diagram above shows a typical NL-MPC setup for a polymer process. The installed MPC comprises two major components

  • Product Quality Monitor. This uses the process model in conjunction with real-time plant data to continually monitor (soft-sense) product quality. These real-time measurements can be augmented by periodic laboratory measurements where necessary, and can eventually replace laboratory measurements if desired.
  • Production Optimizer. This is the nonlinear advanced process controller which maintains production continuously at optimal product quality. During grade change, it minimizes transition time and hence production of off-spec material.

Benefits

Polymer plants can benefit significantly from the application of APC. The nonlinear MPC provides the following specific benefits:

  • The Product Quality Monitor provides a continuous reading of product quality, allowing quality to be tightly controlled, and avoiding ‘oversteer’ resulting from operator actions based on delayed laboratory measurements. This improves production reliability and reduces mixing, repelletizing and off-spec production.
  • Once confidence is established, the measurements from the Product Quality Monitor can completely replace or significantly reduce the need for laboratory measurements.
  • The Production Optimizer ensures that first-pass yield is continuously maximized at steady-state operation while maintaining stable production. It reduces operator interventions and prevents the implementation of shift-specific process conditions.
  • The Production Optimizer increases plant capacity and first-pass yield through faster grade changes, effectively minimizing grade transition time. This reduces  the amount of off-spec product produced and minimizes the effort for mixing and repelletizing, thus maximizing production of profitable grades, providing a higher degree of plant flexibility and reducing recycle and associated energy costs.

Example

The example below shows the effects of applying PSE’s gNLMPC optimizer to a polymer process such as the one shown above for continuous quality monitoring, yield maximization and acceleration of grade change to minimize production of low-value off-spec material during grade transition.

The following plots show the difference that nonlinear MPC can make when compared with a typical recipe-driven approach:

Manual product grade transition

Manual product grade transition

Here the transition is effected in a series of steps according to the operations recipe, with feedback from laboratory measurements at two-hourly intervals. The entire transition takes around 20 hours.

Automated grade transition

gNLMPC Automated grade transition

With the Product Quality Optimizer implemented, and the Product Quality Monitor providing product quality measurements continually, the grade transition can be completed in around a quarter of the time.

Experience shows that the automated grade transition can reduce transition time by 50-75%, reducing off-spec and recycled material and increasing the production of high-quality grades.

For processes where grade change occurs at weekly intervals, this can result in an additional week’s production of desired grades every year.