Product category:
Process Monitoring and Optimisation
News Release from: MatrikonOPC | Subject: BOC Hartford
Edited by the Processingtalk Editorial
Team on 27 July 2007
Faster, safer MPC implementation for BOC
Gases
In 2006 BOC decided to apply a radical new methodology using Tai Ji technology in testing and modeling their Hartford air separation plant, to replace a historically poorly performing MPC Controller
When implementing a new Model Predictive Control (MPC) project or maintaining an existing MPC application, model identification is the most difficult, time-consuming and labour-intensive part of the process Typically, traditional testing, modelling, integration and commissioning procedures take three to four months of around-the-clock work to complete
This article was originally published on Processingtalk on 9 Jul 2008 at 8.00am (UK)
Related stories
MatrikonOPC user group Asia-Pacific and Australia
Matrikon announces the latest MatrikonOPC User Group meeting and conference, to be held in Singapore on 29 July, followed by an optional three day hands-on training workshop
Matrikon Product Expo North - 18 June 2008
Matrikon Europe have recently announced they will be holding a series of regional product events in the UK: the first event, called 'Product Expo North' will be held in Middlesbrough on 18th June
When BOC Gases, a global industrial gas company, made the decision to replace the poorly performing MPC Controller on their Air Separation Unit at their Hartford, Illinois site, they sought out a method to deploy the new system faster and at lower cost, while maintaining safe and effective plant operation.
In 2006 BOC partnered with Matrikon, a veteran of 49 conventional MPC implementations at other BOC sites, to apply a radical new methodology using Tai Ji technology in testing and modeling the Hartford plant.
The BOC Hartford Air Separation Unit makes both liquid and gas products.
Further reading
Alarm management a hot topic in process industry
The recent alarm management seminar held in Aberdeen by Matrikon Europe proved to be a huge success, with papers provided by Shell and EEMUA, plus ABB and Matrikon
AES Wind Generation brings data together with OPC
Collecting data from numerous disparate wind farms into one display system for availability and performance has been made possible by use of MatrikonOPC technology
The next generation of alarm management
Matrikon announces the release of Matrikon Alarm Manager Version 4.1, its premier windows-based alarm system software application, compatible with EEMUA and ISA guidelines
On average, the plant produces 600 tonnes of gaseous oxygen and 650 tonnes of liquid nitrogen and oxygen.
The site also has gaseous oxygen pipelines which feed nearby customers and a GAN pipeline.
BOC Gases is a unique company in that it monitors most of its plants from remote operation centers (ROC) around the world while the actual plant sites have only a few personnel working at any given time.
Since they monitor most of the sites from ROC, their operators are watching 5-6 plants at any given time.
In order to ease the operating load for operators, BOC decided back in October, 2001 to implement a model predictive controller (MPC) for the Hartford plant.
In five years of operation, the performance of this original MPC consistently fell short of specifications.
Operators complained of constant controller issues which regularly led to the MPC being switched off and the plant running under operator control.
With operators already burdened with monitoring numerous other sites, it was crucial that the MPC be re-vamped as soon as possible.
The Hartford plant, according to the operators, was one of the most difficult plants to run outside MPC.
Having reliable MPC was a must.
The Tai Ji implementation method.
It was critical that the new system be implemented quickly in a much faster and safer time frame than conventional industry practice provides.
The process at Hartford is very nonlinear, especially the low pressure column purity control which is extremely critical.
Running the plant under automatic (regulatory) control is difficult and can often require operator intervention to maintain product purities: and even with manual intervention product purity, especially GAN or CLAR, can be lost.
This causes major issues, since the operator is also responsible for operating five or six other sites at the same time.
The time the plant spent in automatic needed to be reduced to an absolute minimum.
BOC and Matrikon decided to make use of Matrikon Control Performance Monitor Tai Ji to greatly accelerate the commissioning process.
Based on the Dr Yucai Zhu industry-proven Tai Ji ID technology, this method provides an alternative to traditional step testing and model identification methods.
The advantages BOC saw in this method were threefold:.
1) It is automatic: Testing and model ID are done automatically rather than manually by engineering personnel.
2) It is multivariable: Multiple MVs are tested simultaneously, making test time much shorter than single-variable tests.
3) It is closed-loop: Tests can be performed closed-loop (MPC and/or PID), resulting in fewer disturbances and fewer operator interventions.
In short, this new method of MPC implementation allows the modeling, step testing, integration and commissioning phases to overlap - at the peak of the procedure all four phases are, in fact, under way simultaneously - while requiring fewer personnel resources, at a greatly reduced risk of process upsets.
Step by step, the methodology ran as follows:.
Closed-loop operating data was obtained from the plant -- approximately one month's worth of data, when there was a lot of movement in the plant.
The closed-loop operating data was put through Control Performance Monitor Tai Ji to obtain initial models.
The initial models were put into the MPC, and the controller was brought online.
The plant was commissioned for two days to ensure everything was running correctly and that the MVs and CVs were functioning as they should.
Once the initial commissioning was complete, a Tai Ji step test was performed for 3-4 days.
This test involved moving all the manipulated variables simultaneously by perturbing CV target values every minute.
At the same time the step test was going on, modeling passes were conducted every 12-16 hrs to give the implementation team direct feedback on how the step test was going.
As we obtained direct feedback, step size changes were made to the CV targets (higher signal to noise ratio) in order to better develop the models.
Toward the end of the step test a lot of key models had converged, and the decision was made to put them online with the step test still going on.
With the new models online and the step test running, commissioning of the new models began.
This is the new methodology at its peak, with all three phases -- step testing, modeling and commissioning -- occurring at the same time.
Once the step test was over, commissioning and fine-tuning of the plant continued.
The end result of this methodology was the commissioning of a reliable, grassroots MPC controller in 12 work weeks rather than the industry standard 4-5 months.
The greatest challenge faced in this implementation was neither technical nor operational, according to Matrikon automation and optimisation engineer Zul Bandali - it was achieving confidence in the results of the new method.
"It is so much easier to see results when step testing is done one MV at a time versus moving all MVs together," says Bandali: "BOC had a tough time in the beginning in believing the results from Tai Ji but eventually their confidence grew and they accepted the results".
Even then there was a challenge in getting operations staff, conditioned by five years working with an under performing MPC, to trust the new controller.
"The operators were so used to seeing issues with the old MPC controller," says Bandali, "that during the step testing and commissioning phases they were always quick to jump in and take a loop out of MPC and put it into automatic.
"When checking with them on why a certain action was taken, they would reply 'Well, the old MPC would tend to bury this purity, so I don't want to take any chances and have an upset'.
An upset could mean anywhere from four to eight hours in lost production.
It took a while but the operators finally developed enough confidence in the new MPC".
The Benefits.
When you add up the savings in engineering coverage, operator interventions, testing time and time spent analysing data - each cut by approximately 50 per cent - the real bottom-line benefit of implementing an MPC controller in this compressed time frame is enormous.
This is in addition to the well-known and industry-proven benefits of having a reliable, well-modeled MPC.
At Hartford, the key areas of benefit were:.
1) Increased LP column stability which leads to better CLAR recovery, reduced operator intervention and reduction of downtime or loss producing events.
2) The ability to change production rates very quickly, without upsetting the plant and the key purities.
3) Optimisation of evaporation tower.
4) Better control of power demand.
The MPC is set up such that it can drive production while also maintaining the power demand targets set by the power company.
5) Better load following on pipelines.
The MPC controller can quickly adjust plant production to meet changes in pipeline demand.
6) Better constraint handling.
Improved ramping: overall average ramp rate improvement of 220-650%.
"LMPC has done a much better job then our previous APC system in maintaining the ever-so-critical low pressure column purity control," says Mike Golinsky, BOC, ROC Engineer: "The LMPC system has been in place five months now, and our downtime on argon has been reduced by over 75%.
The plant will run for weeks or longer without operator intervention being necessary on the LP column purity".
None of these benefits were being realized on the old, underperforming MPC.
While specific numbers for the Hartford plant are not yet available, Golinsky estimates USD20,000/year savings from reduction in upsets alone.
Industry experience shows that a solid MPC implementation will provide USD80 - 85k annual benefit to a plant.
After its demonstrated success at Hartford, this methodology will be carried forward on subsequent MPC implementations at the remaining eight plants in a nine-site rollout with BOC.
"Using Tai-Ji to step test the plant in a closed loop mode was great," says Mike Golinsky: "It reduced testing time, improved model quality, and most importantly reduced the risk of a loss-producing event occurring during the testing process". Request a free brochure from MatrikonOPC ...
• MatrikonOPC: contact details and other news
• Email this article to a colleague
• Register for the free Processingtalk email newsletter
• Processingtalk Home Page
Search the Pro-Talk network of sites
