Identifying poor performers while the process is running

Transcription

1 Idetifyig poor performers while the process is ruig KEYWORDS Michel Ruel P.E. Presidet, TOP Cotrol Ic 4734 Soseeahray Drive 49, Bel-Air St, #103 Hubertus, WI Levis Qc G6W 6K9 USA Caada Process cotrol, Performace, Variability, Key Performace Idex, Stictio, Hysteresis, Oscillatio ABSTRACT Plats are cotiuously dowsizig persoel to improve their profitability. To esure maiteace people are workig where their effort will really improve performace, a process performace moitorig system idetifies clearly which loops do ot perform accordigly to their goal. The performace moitor software prioritizes loops based o greatest ecoomic gai. Maiteace ad actios ca ow be pro-active istead of scheduled. The system idetifies poor performers, oscillatig loops, faulty equipmet, etc. Fially, the program icludes tools to detect, diagose ad quatify problems. KEY METRICS Oe of the keys to makig a performace moitor is to be able to quickly set up the system with metrics that are sigificat to your plat. There must be a template or cookie-cutter approach for settig up your system agaist a bechmark. All assessmet itervals ad metrics are calculated, but ot all metrics are importat for your plat. This is why you eed to determie which metrics are the most sigificat for your plat. For example, paper mills may wat to use variability as a key metric, sice variability throughout plat loops affects variability i their fial product, whereas chemical plats may cosider average error or itegrated absolute error more sigificatly. Variability has bee a buzzword for years, but performace is a lot more tha variability. The performace of a flow loop i a cascade strategy is measured differetly tha the performace of a level loop actig as ivetory cotrol. Most plats will feel oscillatio detectio is a importat metric. Some may wat to look at the amout of time the loop is i automatic or ormal mode. Loops put i maual mode are probably ot workig properly. Idetifyig some of the key metrics is a atural habit for most plats. The plat persoel ofte kow the importat factors affectig the product quality ad dowtime. After the importat key metrics are idetified, templates are built aroud these metrics. The templates are applied to a time period appropriate for bechmarks of performace. The ideal period of time would be after every loop i the plat has bee checked, optimized ad tued. However, the realistic delay for

2 the time period to be settled will represet a portio of time compared with future metrics. Agaist these bechmarks, there are thresholds to be cosidered for each importat metric i the plat. These thresholds combied with the bechmarks provide a compariso of this loop with other loops i the plat. They also provide a compariso of the loop, the uit operatio or the plat agaist previous time periods. A ecoomic weight is the applied o each value, depedig o its ecoomic sigificace. HOW TO DO IT The software is desiged to help you make the biggest impacts o your plat. It pipoits areas that will yield the greatest ecoomic returs. This determies which cotrollers i a plat are ot performig well, which would beefit from re-tuig ad which require maiteace. The software digests data comig from the plat ad geerates s, reports ad lists of loops outside predetermied performace limits. Remote access capability, loop tuig, process aalysis tools, equipmet aalysis tools ad simulatio tools are essetial. The loop moitorig system should provide, o demad, the cotrol egieers ad techicias a list of loops that would make the greatest icrease i profits if used optimally. Maagers also eed to kow how the plat is doig o a historical basis. There are may possible performace idices: variability, IAE (Itegral of Absolute Error), umber of set-poit crossigs, average error, Harris idex, valve travel, time i ormal mode, domiat oscillatio period, etc. You ca choose the oes that are importat for your plat. Figure. 1 Performace idex Some assessmets may ot be importat for certai types of loops. For example, the average error o averagig level loops may ot be a importat idicator of their performace. The performace moitor allows specific key assessmets to be removed from idividual loops or categories of loops. (See figure 1.)

3 LOOP MONITORING REQUIRES DATA GATHERING AND STORAGE The first step required for loop moitorig is to get the data ad save it for assessmets. The primary method of coectig to the process cotrol computer is via OPC. OPC servers are available for most process cotrol computers. I geeral, the faster is the data collectio, the better it is. Sample times of about 1 secod are ideal. However, although persoal computers ad etworks are extremely fast today, it happes frequetly that the process cotrol computer is relatively slow ad becomes the bottleeck for gettig process data quickly. Because of this limitatio, each loop ca be sampled at a differet sample iterval. This iterval should be from 1 secod to 1 miute. This way, the loadig time o the process cotrol computer ca be balaced. Assessor Prioritizer Reportig Moitor Historia PID Tuer / Loop Aalysis OPC Iterface DCS/PLC #1 Figure. 2 DCS/PLC #2 DCS/PLC #N System architecture ASSESSING THE DATA Each loop is assiged to a operatio uit, the same way it is assiged to the plat. Each uit operatio is assiged to a assessmet iterval that defies how ofte that uit operatio performace is assessed. Assessmet times ca vary betwee 1 hour ad 1 week. It may be advatageous to set at 8 or 12 hours the assessmet times i order to compare how differet shift times i the day may affect the performace. Oe assessmet every 24 hours should be appropriate. At the assessmet iterval, the moitorig assessmet service evaluates each loop i the uit operatio. There are may evaluatio poits icludig: Oscillatio Detectio ad Aalysis Loop Robustess Harris idex Settlig time Set-poit Crossigs Normalized Itegral of Absolute Error, Average Error Noise Bad, Variability ad Variace Valve travel, Valve reversals, Valve at limit (% time) Process model ad quality Time i ormal mode of operatio Number of mode chages per shift Etc.

4 PLANT HEATH? LOOP HEALTH? The hierarchy of performace maagemet is: Corporatio Site Plat Uit Operatio Loop Key Performace Idex (KPI) ad ecoomic sigificace Agaist these bechmarks, there are thresholds to be cosidered for each importat metric i the plat: Baselies (target or ideal value): Idicate a referece value to compare agaist i the future. Thresholds (upper or lower permissible value): Represet limits or boudaries betwee which the assessmets would remai if the plat is ruig well. These thresholds combied with the bechmarks provide a ormalized idex for each KPI. Sice the idices are ormalized, they ca be aggregated to form oe idex for the loop. A ecoomic weight is the applied o each value depedig o its ecoomic sigificace. They also provide a compariso of the loop, the uit operatio or the plat agaist previous time periods. For each loop, KPI are agglomerated ad ecoomic sigificace is determied. The result is a percetage value represetig the room for improvemet. The value is calculated for each assessmet but averages ca be obtaied for ay period of time. For each KPI withi each loop, values are ormalized: Idex Bechmark TowardsThreshold = 100 Threshold Bechmark For each loop, idices are aggregated ad ecoomically weighed: % Equatio 1 % TowardsThresholdi % TowardsThresholdEcoomicLoop = EcoomicSigificace Equatio 2 i =1 For each uit, a average is calculated: % TowardsThresholdEcoomic = j = 1 Loop j % TowardsThreshold Equatio 3 Uit For each plat (ad so o for other levels), a average is calculated: % TowardsThresholdEcoomic = k = 1 Loopk % TowardsThreshold Equatio 4 Plat For each level, a umber represetig the performace is obtaied. That umber will take i accout the ecoomic weight ad the KPI for that part of the process. For each umber, the value represets room for improvemet. Maagers, superitedets ad egieers have ow umbers to decide where they should use their resources. Also, for each performace idex, it is possible to calculate a average for a group of loops, a uit, a plat or ay other group (for example, for all flow loops). % TowardsThreshold SpecificPerformaceIdex Loop k % TowardsThreshold SpecificPerformaceIdex = Equatio 5 k = 1

5 For example, a maager could have, as part of a automatic weekly report, the percetage of time whe the valves (withi a group) reach their limit. WHICH LOOPS ARE THE BIGGEST PAYBACK IN YOUR PLANT? Take the Flow1 loop as a example i figure 3 (top left-ed of the diagram). It has a average ecoomic assessmet of early 52%; it is the largest average ecoomic assessmet for all the loops show. The higher the average ecoomic assessmet is, the greater the egative ecoomic impact the loop has o the bottom lie of the plat. Figure. 3 Greatest ecoomic impact o the operatio However, this represets a opportuity to egage this loop, fid out why it is ill ad fix the problem. It will allow you to create the greatest ecoomic impact for the plat. The Biggest payback loops report automatically set up the triage order of loops to focus attetio o. The system the determies where the problems are comig from; diagostics are made ad possible causes are preseted. Figure 4 shows a custom loop list icludig all the possible causes of oscillatio ad a suggestio of diagosis. This list has bee customized to show all the potetial suggested causes of oscillatio: hardware, load upsets or tuig. The first row shows our Flow1 loop, cofirmig that oscillatio is caused by the valve: 100% of the time, it will suggest that the cause of oscillatio is tuig. Figure. 4 Loop List icludig possible diagosis of oscillatio.

6 PERFORM FURTHER TESTING Oce the loop has bee idetified as havig a cyclig problem that is probably caused by the valve, you ca perform additioal tests o the valve to pipoit ad verify the problems. By usig the tools ad equipmet, cotrol strategies ad tuig parameters ca be aalyzed. Two of the suggested tests to perform are the stictio test ad the hysteresis test. Both stictio ad hysteresis are problems that ofte affect valves, ad both will cause the loop to cycle. The tests are made while the process is ruig ad calculated values are preseted. Figure 5 shows that two actios were take based o the examiatio result of the Biggest payback loops report: valve was repaired, brigig the hysteresis to 1% ad stictio to 0.2%. Figure. 5 Report of the hysteresis ad stictio Usig the optimizatio software itegrated i the process moitor, the Flow1 loop was re-tued ad so were two other loops i the uit: a level loop ad a pressure loop. Flow1 was the ier loop of a cascade. The results ca be see after lettig the plat rest for a day followig the repair ad re-tuig of the uit operatio. RESULTS OF REPAIRING BIGGEST PAYBACK LOOPS Figure 6 ow shows that the Flow1 loop is performig better the day after the repair. Figure. 6 Biggest Payback Loops after itervetio Whe we look at the loop history graph as show o figure 7, we ca see how the loop has improved sice the correctios were made. This graph shows the Flow1 loop previous assessmets. The shadig represets the overall or average percetage towards threshold for all the importat assessmets. This is made up of 4 assessmets, as idicated by the additioal lies o the graph. The oldest assessmet (25 hours ago) is o the far-left ed ad the latest is o the far right ed of the graph. At assessmets 22 ad 21, the loop was tested ad the assessmets all icreased. After this time, the valve was repaired ad the loop was re-tued. I cosequece, the assessmets after 21 steadily decreased. The two other lies show the oscillatio measure ad oscillatio diagosis. Together, they are

7 a average of the 10 previous assessmets; it takes 10 assessmets to reach 0. The average error ad ormalized Harris idex dropped immediately after repair. Oscillatig Oscillatig, Hardware Harris Idex (orm) Average error PLANT PERFORMANCE MONITORING Figure. 7 Loop history graph for Flow1. Testig ad improvemets made durig assessmet 22 ad 21. Improvemets startig to show i assessmet 20. Fially, it is iterestig to examie the plat moitorig results over a year, as show i figure 8. PM8 PM3 TMP Figure. 8 Plat moitorig over a year CONCLUSION Very quickly, it is possible to establish a bechmark of metrics or assessmets for a etire plat or a group of loops. Oce established, these bechmarks ad threshold settigs are used as a compariso with other loops i the plat. This will direct the efforts of the plat persoel ad mostly affect the operatio. It also allows a compariso with time that shows, from a ecoomic poit of view, how the plat operatio beefits from the work ad moey spet o the process moitorig system. The process moitorig system should allow plats to prioritize their time to make the biggest ecoomic impacts o the compay bottom lie.

8 ACKNOWLEDGEMENTS All the figures i this article were derived from PlaTriage software, from ExperTue Ic. REFERENCES HUANG, Biao ad SHAH, Shirish L. Performace Assessmet of cotrol loops, Spriger-Verlag, Lodo, 1999, 255 p. LIPTAK, Bela G. Ed. Istrumet Egieerig Hadbook, 3 rd Editio, Process Software ad Digital Networks, Boca Rato, FL, CRC Press 2002, 912 p. (5.6 Platwide Cotrol Loop Optimizatio, p ). ABOUT THE AUTHOR Michel Ruel is a registered professioal egieer, uiversity lecturer ad author of several publicatios ad books o istrumetatio ad cotrol. Michel has 28 years of plat experiece i several compaies such as Mosato Chemicals, Domtar Paper, Dow Corig, Shell Oil, Abitibi-Cosolidated, Petro- Caada, Norada, Degussa, Alca, Smurfit Stoe, Kruger, Pratt & Whitey ad Iteratioal Paper. He is experieced i solvig uusual process cotrol problems. Michel has preseted process cotrol lectures to over 4,000 egieers ad techicias i may coutries. He traslates his experiece i a very user-friedly presetatio ad teachig style, i Frech ad Eglish. Michel is presidet of TOP Cotrol Ic. He is a ISA Fellow member.