NOVEL PROCEDURES FOR RETROFITTING AND DESIGNING 

SENSOR NETWORKS IN PROCESS PLANTS

Accurate data are important for process monitoring as well as for proper production accounting, notwithstanding the inherent value for process control. Without data reconciliation, the accuracy of each variable is given by the accuracy of the instrument that measures it. Therefore, if increased accuracy is desired, it can only be achieved by increasing the accuracy of the instrument that measures it.

With data reconciliation becoming popular in industry, the accuracy of all variables can be improved through redundancy and several unmeasured variables can be estimated. The practitioner discovered that it is possible to influence the accuracy of a certain variable by increasing the accuracy of the measurement of other variables. It was soon discovered that this last option can be cheaper than increasing the accuracy of measurement of the variable in question.

Some commercial packages provide information that can be used to choose good revamping alternatives. However, these packages are not able to provide any useful information regarding the possibility of choosing new measurement points. The attached paper deals with the theoretical and practical aspects related to the revamping of sensor networks for improved accuracy. The whole procedure is coded and it also addresses several issues related to the robustness of the optimal network. For this purpose three new properties are introduced:

• Error Detectability: Gross errors of a certain pre selected size should be detectable.
• Availability: If gross errors are detected and the corresponding measurement is eliminated, a pre selected level of accuracy is still guaranteed
• Resilience: If gross errors are present and are not detected a certain pre selected level of accuracy is guaranteed.

The algorithm is set up so that the best (cheaper) alternative is picked. Sub optimal alternatives can also be made available. Reliability and process fault detectability are being added to the software that is being developed.

Lately, an unconstrained optimization model where the costs are being minimized is being developed. This is based on our recent work on the value of precision and accuracy.

Software

• A software that implements efficiently the techniques developed is under preparation. A first preliminary version is available.

      Contact me for access to all published work