Chemical Engineering Nov 10, 1986 pgs. 99-103 DETECTING ERRORS IN PROCESS DATA William A. Heenan and Robert W. Serth Chemical processes are often controlled by means of microprocessor-based controllers connected to a host computer. The controllers regulate the minute-to-minute operation of the process, using such measurements as flow rate, pressure, temperature and stream composition. The host computer uses the same inputs in optimization programs that determine the setpoints for the regulatory controllers, and in accounting programs that product material- and energy-balance reports for the pant’s management. Unfortunately, process measurements are never perfectly accurate and often they are quite wrong - because the instruments drift, plug up, or partially fail. What happens then? The optimization programs find the wrong optima and therefore waste money; this is one reason why such programs often end up being abandoned. Likewise, the accounting reports present a confusing picture of how the process is operating. If managers rely on the these reports, they may make the wrong decisions regarding how to run the process or where to invest in improving it. Gross errors - those that are well outside the instrument’s normal inaccuracy - are of particular concern because they may greatly distort an optimization or accounting scheme. Such gross errors are usually systematic ones - resulting from instruments that are consistently malfunctioning - rather than random inaccuracies. For instance, the steam or energy report will be seriously in error if a steam flow meter on a main header has partially failed and is reporting 50% too low a flow rate. At best, the production engineer will spot the discrepancy and give an estimate for the true steam flow. This can go on for days until the meter is repaired; meanwhile, another meter starts to drift. All this time the steam report is incomplete or erroneous, and eventually people stop relying on it. What is needed to overcome such problems is a computerized method for detecting gross errors, and reconciling the faulty readings with the more correct ones so as to closely approximate what the readings should have been. This article describes several methods for gross-error detection and data reconciliation, and shows how they can be applied to reconciling mass balances in a steam network. The methods also apply to other systems for which the balance equations are linear - e.g., mass balances in natural-gas distribution systems, total mass or energy balances for process steams, reaction kinetics for reactions subject to stoichiometric constraints. They cannot be applied (without modification) to non-linear systems such as component balances for process steams or simultaneous mass and energy balances. For continuing information of this article, please refer to: Heenan, W.A., and Serth, R.W., Detecting Errors in Process Data, Chemical Engineering, November 10, 1986, p.99-103.
|
|
* * * * * * * * * * * * * * * * * * * * * * * * *
* * * * * * * * * * * * |