In System 800xA APC extension there is a control module for an MPC controller in the AC 800M controller. Using this control module the MPC controller is easily connected to measured signals and to downstream PID controllers. Once this is done, and the application is downloaded to the AC 800M controller, the MPC can be operated manually using preconfigured operator displays and faceplates.

The connections between the MPC and the other objects are established using “control connections.” These are bidirectional multi-signal connections where not only signal values are transported but also the Boolean information about operational modes for the downstream PID.

System 800xA APC utilizes the 800xA infrastructure fully. Since an MPC controller can be computationally demanding the execution of the APC service for the MPC engine can be distributed to any server in the 800xA system. If desired, eg, for additional reliability, a redundant service can also be configured. Further, the System 800xA infrastructure provides all the necessary supervision, and all events and anomalies are recorded in the 800xA alarm and event functionality.

Other benefits of the System 800xA APC are:

• Built on already established ABB products
• Migration path for Predict & Control (P&C) and Expert Optimizer controllers
• A structure for the MPC application is enforced, which simplifies maintenance since all related artifacts are stored in one location

With this new product the control engineer can now concentrate on the control problem, leaving all other issues to the platform.

The MPC controller is packaged as an 800xA system extension with a library and a service. Configuration of an instance of the MPC controller in the 800xA APC starts in the 800M Control Builder. After connecting the PVs (measurements), MVs (controller outputs), and FF variables (measurable disturbances), the application can be downloaded to an 800xA controller. Normally the MPC MVs are connected to external set points for cascaded level-1 PID controllers.

The following has then been accomplished:

• The MPC can be operated in manual mode from faceplates. All signals can be visualized in faceplates. This is useful, for example, for plant testing to obtain data for empirical modeling.
• Supervision is automatically established for the data transfer between the control module and the 800xA service with the MPC engine.
• By using control connections between the MPC and the cascaded PID controllers the MPC will notice when a PID is not operating in auto mode with an external set point, and the MPC will also notice when signals in the PID saturate. The MPC is then able to take the correct actions when such situations occur.

The Model Builder is intended for, as the name indicates, creating the model that will be used in the MPC. The model can be created in three different ways.

1. One way is that a model can be obtained using empirical modeling, where a discrete time-state space model is calculated from logged data. Data should preferably be obtained from an identification experiment where the MVs are changed up and down. There are different ways to do this; the simplest is to make step changes in each of the MVs sequentially.
2. Alternatively a model can be defined by a set of low-order transfer function models, one for each input-output relation in the multivariable model. A typical loworder transfer function is defined by the parameters in: but more complicated transfer functions can also be defined.
3. A third possibility is to graphically build a first-principles model using predefined blocks. This is the most generic method that is supported in the Model Builder.

Although these are completely different approaches, a model can be merged together by using smaller models of any of the three types.

The Model Builder provides functions to analyze models. There are functions for step responses and also for model validation where the model is fed with logged inputs and the simulated model outputs are compared with logged outputs. Once a model is considered to be of sufficient quality for use in the controller, an MPC can be designed. This is also done in the Model Builder.

The influence of the chosen tuning parameters can be evaluated by simulations with different inputs. There are possibilities for simulation with steps in set points, in feed-forward, and in output disturbances. Robustness can easily be evaluated when using a different model for simulation than the one that is used in the MPC controller.

The controller parameters are stored together with the model as an xml file.

The final step is to deploy the designed MPC in the 800xA system. In System 800xA Plant Explorer an xml file, with tuning parameters and model, can be selected to configure the online MPC algorithm with one of the MPC controllers that was defined in the Model Builder. Now the MPC controller is fully operational in the 800xA system and can be switched to auto mode to fulfill its task.

Once the basic configuration is finished a number of tailor-made faceplates are generated by the system. These faceplates contain complete information for both the operator and the APC engineer.

In other words, not only set points and limits are available, but also the internal parameterization of the controller is accessible to authorized 800xA users. Most of the tuning parameters are available in the faceplates provided or operator displays. This is useful if further online tuning is needed.

In cases where several APC controllers are deployed in the same server, it might be necessary to spread the CPU load. This is achieved using the scheduling tool provided by 800xA APC, where each controller is assigned a time slot for its optimal starting point.

Additional functionality is available due to the integration with System 800xA. For example, 800xA offers integrated alarm handling, National Language Support and APC key performance indicator (KPI) tables.