The key values that drive alarm system performance are where the alarm limits are set. Alarm standards recognize this importance, and direct that ‘they should be put at the boundary of where you normally operate’. This was the only advice you ever had for correctly positioning alarm limits. It sounded easy but then you found it was actually very difficult, if not impossible in practice to define the boundary on several hundred process variables corresponding to operating objectives defining ‘normal operation’ specified on maybe only 10 or 20 lab, quality, KPI, performance and equipment constraint variables. You will be relieved to know that PPCL’s C Visual Explorer (CVE) makes it easy to use envelopes of 1,000 or even more variables so that, for the first time ever, you can see a ‘feed to product’ envelope spanning several process units and gain new knowledge of what process activities in which unit lead to you meeting or missing end-of-line targets, KPIs and specifications.

How do you get from limits on ‘result’ (dependent) variables to limits on process (independent) variables? It’s not easy, even if you have a sophisticated first principles process model because process models are almost always in the form of y=f(x) where the y’s are the results and the x’s are the independent variables. You really want a model that gives you x=f(y), but such models don’t generally exist even for simple examples, let alone for the general case of a large operating plant. You could run many cases of a y=f(x) model to populate a solution space as in our paper, but it is generally only practical to do this with a very small model. See Brooks, Drury and van Walsem, Choosing Cut Points to Optimise Product Yields, Hydrocarbon Processing 1999, 78(11), 53-60).

But although algebra can’t deliver, geometry can so that you can see x as a function of y! That’s why CVE’s novel visualization method is such an important discovery. It can combine the dependent and independent variables from process history data into one picture and, by putting the operating objectives and specifications that define ‘normal’ on the dependent y-variables, CVE will show you the actual envelope that you would have used on the independent x-variables to achieve those specifications, targets and KPIs on the y-variables. Or put operating limits on some or all of the independent x-variables and it will show you all the results you would have obtained on the dependent y-variables from which you can immediately calculate the yield of desired results that you would have achieved and decide whether it is better or worse than you are achieving now.

And in the next 5 minutes you can make up ‘better’ and ‘worse’ specifications and add those to the picture so that you have a ‘simple response surface visualization’ or ‘contour map’ to help you understand which independent variable changes will improve or worsen the yield.