PPCL Logo Process Plant Computing Limited

Tel: +44 1753 893090 |

LinkedIn Twitter
A Master Alarm Database (MAD) is a necessity and there is plenty of product choice available. All use extracts of information from the real-time alarm log and have the ability to produce large varieties of alarm performance reports which identify alarm system problems but are of limited value in solving them. But, in spite of the abundance of reports, plants all over the world struggle to achieve effective operator alarm systems. Managing alarms effectively is only one of many requirements of a good alarm system.

The simple answer is yes! CVE lets you pursue individual variable objectives. You may need to think about the sequence in which you pursue them because maximizing A within B may not give the same result as maximizing B within A. An objective function simplifies the problem by giving weights to different objectives while combining them into one objective. If you have one then include it as a variable in your CVE analysis and you can visually compare the difference between maximizing the objective function and maximizing several individual objectives. CVE is very good for difficult optimizations because it doesn’t hide the non-linearities in your data and lets you use your intelligence to the full. It was used to find the optimal composition of mixed refrigerant in an LNG liquefaction process which couldn’t be optimized in any other way.

The two main problems with start-ups is that (1) they may span more than one shift so more than one process operator and a cross-shift memory is then required and (2) because start-ups occur infrequently it is likely that at least one of the shift operators will be new to the start-up and performing the procedure for the first time. There are many other contributing problems that you may have solved already, such as ensuring that everyone involved is using the latest version of the start-up procedure and not the outdated paper copy left in a drawer from the last start-up.  

Process data is extremely highly correlated by heat, mass and momentum conservation and reaction kinetics. It is further complicated by variable volumetric hold-up, back-mixing and recycle streams. It is broken into numerous discontinuous segments by events such as switching a valve between auto and manual; changes of operating practice between operating shifts; variations in arrival rates of lab analyses of process samples; and on-line analysers drifting out of equipment fouling and mechanical faults. Understanding all of this variability and being able to recognise and differentiate between "normal" and "unusual" requires not only the ability to break the historical data into its many discontinuous segments, but also considerable knowledge of the process and its local operating practices - and perhaps even the operating preferences and practices of particular process operators.

This implies that the people needed to do any analysis will be the process, control, maintenance and instrumentation engineers who work closely with the unit and its operations staff. These are hands-on engineers who are always busy and unlikely to have the time or wish to acquire the depth of advanced mathematical knowledge necessary safely to make use of “traditional” data analytics, including "black-box" algorithms, such as neural nets, where the reasons for model outputs are not always easy for the user to understand and explain to others.  

People often ask what system changes are required to implement any of our products. The answer is - usually none, it's that simple.

The GPC system works with process and quality data already captured in the existing plant historian or even in manual spreadsheets of data. This means that any plant, regardless of how basic or advanced the systems in place are, can make use of this technology. No expensive capital outlays are required: the software is provided on a CD from which the programs are installed.

For pharmaceutical companies, QbD and PAT are current hot topics. Fundamental to these concepts are design spaces, control spaces and operating spaces. These concepts are relatively easy to understand in two or three variable problems, but in real-world situations where many more variables exist, visualising these problems is infinitely more complex.

Why are some batches better than others? Why do two identical batch reactors produce different batches? Probably because there has never been a way to compare two or more batches, let alone several batches made in each of two reactors. The solution to this long-term problem has come not from complicated mathematics and statistics but from n-dimensional geometry developed by pioneers in their field, PPCL.

Do you ever find that your process historian is excellent at capturing valuable data but then you have little time to truly understand what the data is trying to tell you? It’s another thing to add to the list of tasks you probably won’t get done today. Eventually, you’ll sit down minutes before the monthly meeting and use your highlighter pen to go over some discrepancies in the monthly data overview. There’s information there - but you don’t have time to play detective with your spreadsheet.

Often the case with many things is that you don’t know what you’re missing. In the world of process control it is no different. You have survived all these years without alternate methods and you have coped just fine, so why try something new?

The root-cause of Alarm Rationalisation having been such a tedious and time-consuming procedure has been the lack of any capability for Alarm Performance Prediction (APP).  You have had to choose values for alarm limits and then implement and use them in the control room for days or weeks before knowing whether you chose the right value. With hundreds of alarm limits to manage the cycle of monitoring, adjusting, reviewing and re-implementing activity alone represented a substantial workload that in practice was rarely adequately resourced. Being able to predict alarm performance before implementation would eliminate most of this activity and free-up resources for more effective process stewardship.

Many product properties can only be measured in the laboratory. This is frustrating for the people operating the plant that makes the product because they get the results only after the event. What the production people would really like is the ability to predict from the process variables what the properties of the final product will be. Then they wouldn’t have to wait for the lab result to tell them they had it wrong and had achieved only mediocre product quality.

Geometric Process Control (GPC) technology can be successfully applied to DOE through the C Response Surface Visualiser (CRSV) tool. CRSV allows Experimenters or Formulators to perform their own Response Surface Analysis at their workbench in a fraction of the time currently required. It replaces the numerically intensive finding of multi-variable regression equations to describe the Response Surface with a simple visual method that anyone can use.

It is said that as much as two thirds of the processing capacity of the human brain is devoted to visual processing and, while this is hard to verify, it is an everyday truth that we all find pictures much easier to understand that words or numbers. In scientific work our ‘picture’ is a graph. Just think how often your first action on receiving a spreadsheet full of numbers is to click on ‘graph’ in order to get a first impression of what information might exist within the numbers.

When you lift your hand to catch a ball it is very unlikely that your brain actually formulates or solves the equations of motion in working out when and where to place your hand to catch the ball. It’s much more likely that you use, unconsciously, something that’s much more akin to the geometrical methods underlying the Geometric Process Control (GPC) technology described in this article.


Register Now!
PPCL Webinar: Alarm Rationalization
14th and 15th October

PPCL © Copyright 2011-2019. All Rights Reserved. | Terms of use | Privacy Policy | P.O. Box 43, Gerrards Cross, Buckinghamshire, SL9 8UX, UK.