Process monitoring and on-line re-identification

So far, we presented the general structure of the problem and the first main aim is been analyzed. The second objective has also to be included in the same structure, although it looks to the supervision of the control loop. It is well know that, in the class of plant that are object of our study, human supervision is important. Moreover, due to the complexity of the problem, the supervisor needs some support tool (generally based on probability and statistics theory) which can take into account in an easy way the big number of problems/variables that have to be supervised/monitored. It is first of all necessary the monitoring of the performance of the control loop, and in particular of the MPC controller. It has to be remarked that the term good performance means good economic performance since, particularly in the petrochemical processes, an inefficient production is a bad production.

Monitoring takes into account the evaluation of the performance as well as diagnosis, and follow three main steps: 1) determination of appropriate references, by means of which the controller under monitoring can be evaluated, 2) establishing some monitoring statistics or performance indexes to evaluate the controller, 3) introducing methods that diagnose the causes that originated the bad performance, once this performance has been evaluated. The main cause of bad performance in control loops based on MPC is the differences between the prediction model and the plant (model errors), the changes in materials’ properties, disturbances and bad maintenance. In Figure 2 is shown how monitoring is incorporated in the control structure introduced previously.

As already mentioned, model errors are the main cause of bad performance in MPC. Once this error is detected, it is necessary to carry out a new model identifications, with the aim of obtaining updated parameters of the model. However, the identification of such a complex system like a petrochemical one is not that simple. It would be necessary to stop the plant in to order to realize the identification and update the parameters, which represents an expensive loss for any company. A way to avoid this process is on-line re-identification: that is, identification carried out without stopping the plant. This is another field in control theory: that is closed-loop identification. Clearly, closed-loop identification is a very desirable alternative for petrochemical plants (as the attention of PETROBRAS for this topic shows).

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.