Curvaceous Customer Newsletter Issue 1 2005

Curvaceous chosen as finalists in
The Carbon Trust Innovation Awards 2005

Curvaceous are proud to have been one of only three finalists from over 250 nominations for The Carbon Trust Innovation Awards 2005. Curvaceous was chosen for the success of their new-technological approach to reducing carbon dioxide emissions from operating process plant by increasing plant efficiency.

From a trial in 2002 on a 10MW Combined Heat and Power (CHP) plant Curvaceous saved 13% of fuel cost equating to over £200,000 per year. Over 11,000 tonnes per annum of carbon dioxide emissions were cut as a direct result without any capital expenditure.

Curvaceous was delighted to reach the final of this prestigious award. Managing Director Dr. Robin Brooks

“Improving the efficiency of existing plants in the Process Industries ultimately means that they need to burn less fuel. This has a significant effect on reducing climate change, the effects of which are becoming more and more noticeable. Our focus on the efficiency of the process industries is helping many large process companies achieve better efficiency with short-term economic benefits to themselves and reduced emissions benefits to the environment. We are proud that The Carbon Trust has recognised Curvaceous’ invention and development of GPC and honoured GPC as a successful technology in the struggle to lower carbon dioxide emissions.”

TAP

Put the date in your diary today, make sure your TAP subscription is up-to-date and don't miss out.

To be held after Easter this year and with the promise of an exciting new agenda register now to show your interest.

Something new for 2005

The FREE training day will be held on Tuesday 26th April BEFORE the TAP Forum on Wednesday 27th April. This will include hands-on activities and is a great warm up for the Forum. One particular benefit is the opportunity to network with other Curvaceous users from a range of industries and disciplines but who very often share similar problems.

Places for the training will be limited on a first come first served basis so please register your interest now.

1. NEW AGENT IN NEW JERSEY

Cayuga Partners of New Jersey have become the latest addition to the Curvaceous Agent Team.

Walt Beadling, Managing Partner of Cayuga Partners, is looking forward to working with Curvaceous;

“The Curvaceous technology will enable us to bring many more benefits to our existing customers and new clients. Its unique ability to visualize, analyze and optimize processes from end to end is strikingly simple and provides a new understanding of operations and consequently how to improve them. We are especially excited about its potential for supply chain management and optimization.”

Cayuga Partners

Send to a friend

www.curvaceous.de.

2. THE POINT VALUE PREDICTOR

We invest substantially in Research to build our patent portfolio. This is one of our research projects now moving towards development which might be of interest to many of you...

A universal predictor of product properties from process measurements

The basic concept of Geometric Process Control (GPC) technology is the Best Operating Zone (BOZ). As we hope you know, this is the multidimensional envelope of a set of operating datapoints, all of which represent good operation. The selection of the BOZ dataset is a skilled engineering activity, which is undertaken using a larger collection of historical operating data that fully defines the GPC model.

During online operation GPC uses the geometry of the BOZ and the current values of the online process variables to calculate limits within which all process variables must lie for satisfactory operation. These are dynamic limits that change at every time-step.

Now consider the case when some key variables – usually product qualities – are not available online, but are determined after the event, e.g. by laboratory analysis. We call these “Quality Variables”. Historical values of these variables are included in the BOZ dataset.

We know that as long as the online process variables are kept inside the BOZ (exactly what GPC does), the values of the quality variables will be within specification. We know this because the BOZ dataset was chosen to meet those criterion. Further, the multi-dimensional geometry of the BOZ, together with the current values of the online process variables, provide a predicted range for the current value of each quality variable which is narrower than the specification.

It is often useful to have specific point predictions of the current values of quality variables in addition to their ranges. To provide this, GPC makes use of information contained in the interior points of the BOZ. The historical datapoints that are “close” (in a multi-dimensional sense) to the current (measured) operating point contain additional information that can be used to predict the (unmeasured) values of the quality variables.

In addition to making these point predictions, GPC uses another quite independent technique to predict narrow limits on the values of the quality variables. These are calculated using the multi-dimensional envelope of an inner “sliver” of the BOZ dataset. These are not confidence limits in the statistical sense, since confidence limits are a uni-variate construct, but instead show the (very small) range within which we are 100% certain that the actual result will be found.

There are a number of parameters of the prediction calculation that are user-settable at present. None has been found to be critical indicating that they can be discarded. This leads to a predictor that does not require parameters and hence may eventually not require to be individually ‘fitted’ to data. The method does not require any equation fitting for its quality value predictions. It is completely different from the neural-net, statistical and chemometric methods traditionally used for inferential predictions.

The big advantage that it may eventually deliver is that it will be the first universal inferentive predictor not needing to be individually calibrated for each variable to be predicted. It shows how much information is actually buried in existing data that was not accessible prior to GPC.

If you think you have an early application please contact us

3. ALL QUESTIONS HAVE ANSWERS

As mentioned in the last issue, Curvaceous has been working closely with Huntsman Petrochemicals and jointly produced some very good results. The IChemE's Chemical Engineer magazine includes the Huntsman-Curvaceous partnership in their new Question & Answer section giving an insight into the challenges and process problems of an operating paraxylene plant and the solutions that Curvaceous helped to discover. See the evidence here.

Curvaceous has featured in magazines such as Hydrocarbon Processing, Process Control News Europe and MPT Ireland over the past few months.

If you have had an experience you would like to share please contact us and we'll do our best to get your voice heard.

4. HINTS & TIPS

Here's a something new to try on Visual Explorer (CVE)

Visualisation of the Distribution of a variable

The distribution of data values for a variable can be of interest particularly if many points occupy single values or narrow bands.

To visualise this distribution, the technique is to first create a SortOrder variable of the variable of interest

(Variables > Add new function variable > SortOrder)

Then display a Scatter Plot of the new SortOrder variable against the variable of interest. The result being a plot of the cumulative distribution.

SortOrder is a multi-variate sort capability of Visual Explorer; it does not re-arrange the data, but creates a new variable with values from 0 to n-1 for n points. It allocates the value 0 to the lowest ‘value’ multi-variate point up to n-1 for the highest. The sequence of variables in its parameter list has positional significance, thus performing a ‘lexicographical’ sort.

For this particular requirement we are only passing the single variable of interest to SortOrder to create a new variable with content corresponding to the sort sequence of the target variable.

A classic cumulative distribution plot has the SortOrder variable on the vertical axis on a scale of 0 to 1; this linear transform maybe simply effected by plotting against a derived variable of ‘SortOrder/max(SortOrder) as in the example plot below.

A range query on the SortOrder variable can be used to quantify population percentage within ranges of interest. Many data points of similar value overlaid on the parallel axis plot will become apparent as a steep gradient on the distribution plot.

This technique arose from suggestions of the GPC User Group.
Watch out for more handy hints and tips next time.

5. USE IT OR LOSE IT!

The GPC User Group is used much less than it deserves to be so we are happy to promote it here as perhaps some of you are unaware of it. The Groups moderator is Mike Tyrrell of INEOS Chlor.

Anything can be discussed from the latest tricks and tips for CVE to the state of the weather. The group is entirely independent and therefore acts as a good networking tool for people from around the process industries in several different countries. So go on, pick each others brains, get independent advice and meet new people trying to get the same solutions to their problems as you!

Watch this space for more news from the Independent User Group.

www.curvaceous.com