Model-based automation

Enhancing operation by using first-principles models online

Much of today's online process control relies solely on observed plant data and has little predictive capability.

Where "predictive" models are used - for example the linear models used by Model-based Predictive Controllers (MPC) - these tend to be quite simplistic or are valid only for a small and specific set of operating states.

The benefits of model-based automation are all the benefits of better control, and more:

higher on-spec production
tighter quality control
continuous optimisation of economic performance
faster and better-controlled transitions between operating points
safer operation closer to process constraints
better monitoring of complex processes and equipment
better and more up-to-date management information
greater understanding of processes.

Unlike past approaches, the concepts described here can also be applied to batch process control.

The use of first-principles models - as opposed to linear models constructed from plant step tests, or "inferred" models built using techniques such as system identification - opens the way for much more sophisticated process control.

First-principles models bring many benefits to the operational context. The most obvious is the ability to keep the plant closer to the desired operating state by anticipating deviation rather than just reacting to it.

However, once such a model is available, it can be used for many applications, such as inferential ("soft") sensing, equipment and process health monitoring, transition optimisation, and many more.

The Master Model concept

High-fidelity process models developed within PSE's gPROMS modelling and solution environment can be exported to automation environments, where they act as a Master Model for applications providing extended process information and on-the-spot decision support.

The diagram below shows how a model developed originally for process design can be placed at the heart of a model-based control system, with little or no rework. PSE model-based automation approach

The Master Model approach

The pre-requisite for the online master model approach is a model of sufficient accuracy to predict behaviour during all typical operating states.

Increasingly frequently, this type of model is available from the process design stage. Ideally the model incorporates first-principles representation of reaction, separation, hydraulic and other key phenomena.

This model is then tuned against laboratory or pilot plant data to get accurate values for parameters - such as reaction kinetic coefficients - that will not change with day-to-day operation. The resulting tuned model is the Master Model, which can then be implemented online to sit at the heart of operations.

The online model is brought up-to-date on a regular (for example, daily or hourly) basis via a state-estimation operation to reflect change in plant parameters such as heat transfer coefficients or catalyst selectivity. This estimation process also provides reconciled "soft-sensor" information, a significant benefit in itself.

The up-to-date Master Model can then be used for many different activities:

Model-based automation activities using the Master Model

PSE activities

PSE are working with partners Honeywell, ABB and IPCOS Technology to apply the approaches outlined here.

PSE and IPCOS Technology have worked together to apply the Master Model approach to both continuous and batch polymer processes, and large-scale crystallisation processes for sugar and artificial sweetener production.

An example is the batch polymerization of Expanded Polystyrene (EPS) at BASF, where a reduction in batch time of nearly 30% was demonstrated.

Monitoring activities

Model-based automation applications can provide an unprecedented level of plant monitoring:

Soft (inferential) sensing

Reconciled data for any variable in the up-to-date plant model can be used as a measurement for display to the operator consoles or in control to provide a high degree of insight into the process. If required, parallel simulation runs using data from the up-to-date model can provide higher-frequency measurements.

Equipment health monitoring

Regular parameter estimation can provide highly accurate values for key equipment parameters, such as efficiencies and heat transfer coefficients. This is particularly important in monitoring of remote processes such as undersea installations.

Process health monitoring

In a similar way, parameter estimation can be used for general process health monitoring - for example, to provide key process performance indicators (KPIs) that may be calculated from a combination of real and inferred measurements.

Model-based Predictive Control (MPC)

The up-to-date master model can generate a linearised plant model at any operating state instantaneously. MPCs that can take advantage of multiple models can thus control current plant operation very closely, approximating and gaining all the benefits of non-linear control but using much simpler and lower-cost linear techniques.

Transition optimisation, including look-ahead simulation and optimisation

Much of the off-spec production of any plant occurs during transitions such as grade or feedstock change. Model-based automation techniques can reduce this significantly.

Trajectory optimisation. The Master Model can be used to generate the optimal trajectories - based on an economic objective function - for plant controller set points during transitions, using dynamic optimisation techniques.

Real-time optimisation. The linearised models generated for the MPC can be used in an online dynamic trajectory optimiser such as IPCOS' Pathfinder, to implement the optimised trajectories online during transitions.

This optimised transition control allows changes in operation to be achieved with maximum efficiency. Because it uses a linearised model, the optimiser is capable of running in a real-time even for the most complex processes.