Curvaceous control system alarm management

Alarm management is a both a worldwide hot topic and a widespread problem. With the advent of digital control rooms, where alarms are available and selectable for every input, there can be up to 2,000 – 4,000 alarms per console, potentially providing thousands of alarm events in an incident, which simply cannot be evaluated by the operator. How can the over-whelmed operator decide which alarms are safe to be ignored?

Investigations into alarm management systems began in the mid-1990s, due largely to legislation, fear of litigation, and initiatives by OSHA and HSE.  However despite these drivers, quantifying the economic value of an alarm system, or even the value of rationalising it, has rarely been attempted. Few, if any, plants actually know the value, as opposed to the cost, of their alarm systems so they cannot begin to justify initiating additional expenditure on alarm rationalisation or on-going continuous improvement. Sadly focus falls on alarm system performance only when companies are compelled by legislation or coerced by the need to be seen to have adhered to Best Practice.

Following his career at ICI, and then as Programme Director of the Abnormal Situations Management Consortium in the USA, Ian Nimmo formed UCDS Inc, a consultancy to help provide ‘Best Practice’ solutions for reducing the frequency and severity of abnormal situations in process plant control. After presenting a paper at the London Control Room Conference last December, Ian said: “At this conference I saw some outstanding alarm work by Dr Robin Brooks at Curvaceous Software. This was the first alarm management paper in the last ten years that got me excited as he explained how alarms can achieve management objectives, how to understand alarm settings in a new light, and how to change them correctly.”

Following up on this introduction, Ian Nimmo attended the conference for users of Curvaceous Software held in March 2010, to hear more about the latest version of the CVE 2.5 software for Alarm Rationalisation and management. Robin Brooks, Managing Director of PPCL, the new name for Curvaceous Software, explained that the plant operational performance data available from historians provides the profile of previous plant operations.

Maintaining quality and saving £000s in Plastics Production

In the Curvaceous Conference user group discussions, George Forrest, Managing Director of Altremis Limited, the only Multi-Variable Analysis consultancy in the UK, outlined the use of CVE in solving process performance and quality problems in a UK plastics company, which helped to realise around £180k of savings per year.

The company made hundreds of products across three manufacturing lines, which used extruders to push hot molten plastic through a heated pelleter plate to create long strands of plastic. The pelleter plate is heated by hot oil, and the process is similar to forming spaghetti strands. On the first line, as these strands exit the pelleter plate they are cut into pellets by high speed rotating knives.

The facility was experiencing occasional excessive knife wear on a specific family of products resulting in poor pellet quality. This led to excessive operational down-time, operator intervention, off-quality material and customer complaints. By using CVE software, data from the last ten product specific runs over a 12 – 18 month period was collated. This included measured weighing and process variables, set points and quality data. Knife wear was measured and indicated by the current amperage of the motor turning the knives. Within a very short time it was discovered that the excessive knife wear was related to the temperature of the pelleter plate. When the plate temperature was low, knife wear increased.

The CVE analysis recommended increasing the temperature set point conditions for the pelleter plate along with loop turning work to reduce the temperature variation. The facility then went from an average of seven to eight knife changes per product to just one knife change at the start of each run. In terms of reduced downtime, knife wear and off-quality generation the facility estimated a saving of around £80,000 per year and a reduction in customer complaints to zero.

In a related study, the same facility saved a further £100k by using CVE to solve a ‘plant mystery’ – why a high quality product would suddenly go off-quality for no apparent reason. Data was collated from three specific product runs: one excellent, the second average and the third poor. Additionally all the data from the other two manufacturing lines, from the period when the product runs had occurred was assessed. CVE analysis quickly discovered the root cause of the problem to be a drop in pelleter plate temperature: eventually this was traced to subtle disturbances in the hot oil boiler system on start-up of the second manufacturing line, because two of the three line pelleter plates were heated using a common hot oil system.

From this work, it became clear how important the pelleter plate temperature was in relation to the product quality and how sensitive it was to temperature variation. Pelleter plate temperature is now recognised as a key process variable within all the company’s other facilities worldwide.

Refinery applications

Much of the current interest in the use of the PPCL approach, using Visual Explorer (CVE) software for improved alarm management, arises from petrochemical processing and oil refining operations. An example quoted by Robin Brooks used data from a refinery hydro-desulphurisation (HDS) unit.

 

 

 

Figure 1: CVE plot of plant data over 3 months for 26 variables (vertical pink lines). Alarm set points are triangles, production of in spec kerosene plots are in blue, with out of spec operation in pink.

By using a CVE display graph showing 178 process variables as recorded at 5 minute intervals over three months of operation, it was possible to identify the patterns for the three different modes of operation: Standby, Kerosene desulphurisation and Light Gas Oil (LGO) desulphurisation. Like most plants, they had one set of alarm limits to cover all three modes – not necessarily optimal for any mode. Necessarily this leads to a lot of standing alarms, and a high alarm annunciation rate. For example there were 41 standing alarms in the standby mode, and a minimum of 3 at all times in other modes of operation. Alarm annunciations peak at 22/hour but are typically 8 per hour.

In order for the operator to use the process monitoring alarms to enable him to tune the plant for best operation, for example in Kerosene desulphurisation, then the plant alarms must be defined for this operational mode only.  Taking the plant CVE plot for the Kerosene desulphurisation mode only, it was seen that the starting positions for the alarms included almost all of the previous plant operational states, which produced both in-spec and out-of-spec kerosene.

By moving the alarm limits to the edge of the in-spec product profile only, the operators will be triggered with an alarm whenever the process strays into an area where previous data shows a history of out-of-spec performance. Previously the operators did not have this view of the operating envelope of plant performance available, and so had not realised the incorrect settings of the alarms that could have advised them of the poor performance.

Bringing the alarm limits in to the boundaries of this good operational zone without improving and tuning the model based controls does lead to unacceptably high standing alarm counts and annunciation rates – which can be seen in CVE – so improvement of the plant operating controls in parallel with the tightening of alarm limits typically sets a medium-term plant process improvement programme in place for the operation.  This is made possible and practical by the new-found ability with CVE to literally ‘see’ operating envelopes

The same challenge had previously been faced by operators at Mallinckrodt Chemicals, where some CVE alarm limits that triggered a pump shut-down were adjusted to be tighter, and significantly within the normal operational limits, to see whether the operators would be able to learn how to control the plant to a different style. In their para-aminophenol (PAP) process, these operators soon learned how to operate within these new tighter limits, encouraged, it was reported, by their desire to avoid going out into the cold and rain to the field-mounted pump starters!

“Alerts” provide operator guidance

A major refinery in North America has successfully used the PPCL alarm rationalisation processes within the Visual Explorer (CVE) and Process Modeller (CPM) software to reset their alarm limits.

A new class of alarms, known as “Alerts”, implemented by the real-time CPM product, were considered to fulfil the role of lower priority alarms and effectively give early warning of impending problems, but without themselves having a high annunciation rate. Alerts are similar to alarms except that they are not subject to the same rigorous Management of Change procedures, and they do not have fixed values – these depend on the current stage of the process.

The Manager leading the project said: “Alerts require Alert limits. They are related to the current process operating point so don’t have fixed values, unlike HiLo alarm limits, and there has previously been no way to establish Alert limit values, that would be usable for more than a few days at best. In this project, we have introduced the new PPCL GPC technology, based on an operating envelope of the process that re-calculates the Alert limits in real-time as an inter-related set of limits. The way we find the operating envelope guarantees that the process is always inside the HiLo alarm limits.”

The use of CVE delivered a significant increase in the ‘No Alarms Present’ time for each segment of the process. Using CPM to find Alert limits also generated corrective advice to assist the operator, enabling him to avoid approaching the HiLo Limits, and thus reducing their Probability of Failure On Demand. The existing HiLo Limits were rationalised one-by-one using the experience and deep process understanding of Operators and Support Engineers, with the objective of minimising the number of visible alarms and the annunciation rate at any time. This maximises the significance of those alarms that do remain visible.

Time variation of Alerts

The alarm management software in CVE2.5 can take the data plot and use an algorithm to set the operator “Alerts” at the edge of the area defining previously known good performance. So new operating Alert levels can be set when the process moves into the next phase, which might be triggered by elapsed time, or by the initiation of a valve sequence when the process moves on a step. This can be configured within the CVE software, where the Alert levels themselves move up and down in time, in step with the process.

Current pressure on oil refineries operating within the USA, after the Texas City explosion and investigation, numerous incidents since then investigated by the Chemical Safety Board, and calls for tighter OSHA regulation under the new administration, has meant that all process control alarm system procedures are under scrutiny: this style of intelligent Alert system, based on previous operational data, combined with separate ESD systems, shows potential for providing effective control: this is why there is so much interest in PPCL systems from the petrochemical and oil refining industries.

Curvaceous Software is now known as PPCL, and can be found at http://www.ppcl.com

Ian Nimmo at UCDS can be found at http://www.mycontrolroom.com

Advertisements

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: