ctllp05044Mistakes.qxd 4/18/2005 11:29 AM Page 1 THE FOUR BI GGEST M I STA KES I N I N STRUM EN TATI ON . . . A 1 De spit e ongoing a dva nce m e nt s in m e a sure - m e nt a nd com m unicat ions t e chnology, inst ru- m e nt ing a proce ss f or f e e dba ck cont rol re m a ins a t e chnica l cha lle nge . Today s se nsors a re ce rt a inly m ore sophist icat e d t ha n eve r be f ore , a nd f ie ldbus t e chnology ha s sim plif ie d m a ny inst a llat ion issue s conside ra bly. N one t he le ss, m uch ca n st ill go w rong w it h a n inst rum e nt at ion proje ct . The proble m lie s in t he st ra ight f orw a rd nat ure of inst rum e nt at ion proje ct s: e a ch va ria ble t o be m e a sure d m ust be m at che d w it h t he m ost a ppropriat e se nsor; t he se nsor m ust be inst a lle d, ca librat e d, a nd int e rf a ce d t o t he con- t rolle r; a nd t he inf orm at ion ge ne rat e d by t he se nsor m ust be f ilt e re d, f a ct ore d, a nd f ile d in orde r t o give t he cont rolle r a n a ccurat e pict ure of w hat s going on in t he proce ss. This a ppa r- e nt sim plicit y is w hat of t e n le a ds t o a f a lse se nse of se curit y a nd m isst e ps in a m ine f ie ld of pot e nt ia l proble m s. W it h t hat in m ind, he re a re f our circum st a nce s you de f init e ly w a nt t o avoid. >> M ist a ke # 1 : Se le ct ing t he w rong se nsor Technology m ism atch A lthough its generally obvious w hat quantity needs to be m eas- ured in a flow , tem perature, or pressure control application, its not alw ays obvious w hat kind of flow m eter, tem perature sensor, or pres- sure gauge is best suited to the job. A m ism atch betw een the sens- ing technology and the m aterial to be sensed can lead to skew ed m easurem ents and severely degraded control. This is especially true w hen m easuring flow rates. A ll flow m eters are designed to m easure the rate at w hich a gas or liquid has been passing through a partic- ular section of pipe, but not all flow m eters can m easure all flow s. A m agnetic flow m eter or m agm eter, for exam ple, can only detect the flow of electrically conductive m aterials by m eans of m agnetic induction. N on-conductive fluids like pure w ater w ill pass through a m agm eter undetected. M agm eters also have trouble distinguishing air bubbles from the fluid in the pipe. A s a result, a m agm eter w ill alw ays yield an artifi- cially high reading w hen bubbles pass through because it cannot sense the decrease in fluid volum e caused by the presence of the bubbles. In a feedback loop, this occurrrence w ould cause the con- troller to throttle back the flow rate m ore than necessary, preventing the required volum e of fluid from reaching the dow nstream process. The problem gets even w orse if the pipe is so full of air that it is only partially filled w ith liquid, a condition know n as open channel. A lthough recent technological inno- vations allow certain m agm eters to w ork in such a challenging environ- m ent, m echanical sensors such as turbines yield artificially high read- ings, since a trickle of fluid w ill m ove the m eters m echanism just as m uch as a full-pipe flow travel- ing at the sam e speed. O n the other hand, m echanical sensors are not affected by the conductivity ctllp05044Mistakes.qxd 4/18/2005 11:29 AM Page 2 N . . . A N D HOW TO AVOI D THEM 3 1 2 of the fluid, so they w ill som etim es w ork w here m agm eters fail. A n even m ore challenging appli- cation is the m easurem ent of pH in a caustic liquid such as the slurries found in paper m ills. A general-pur- pose pH probe m ade of corrodible m aterials m ight not only generate inaccurate data, it m ight die alto- gether, som etim es w ithin a m atter of days. Som e probes, such as those offered by A B B , are specifi- cally designed for such tough envi- ronm ents. They can double, triple, and even quadruple probe life in m any applications. The trick is to find the right tech- nology for the application, or to choose instrum ents that span a broader range of solutions. For exam ple, new digital technologies allow som e flow m eters to solve m any m ore flow problem s than their predecessors. Instrum entation vendors can be of help in avoiding the technology m ism atch m istake. The best ven- dors train their sales people to assist w ith sensor selection and provide clients w ith easy-to-use selection guides. Som e even offer extensive look-up tables based on product num ber, application, and serial num bers of past installations an especially useful service w hen replacing older products. Finding all the right parts can also be a challenge. Som e instru- m ents require specific housings, m ounting hardw are, and transm it- ters to forw ard the sensors data to the controller. The right vendor can m ake all the difference by provid- ing the entire assem bly under a single catalog num ber. W hen it com es to tem perature instrum enta- tion, for exam ple, training costs and purchasing effort are reduced w hen then vendor offers com pati- ble probes and transm itters togeth- er as a package. Paying too m uch (or too little) C orrect sensor selection is also a m atter of balancing cost against perform ance. W hen theres a choice of equally effective tech- nologies, the right choice is gener- ally the cheapest one that gets the job done. Tem perature instrum entation is a classic exam ple. The tw o dom inant technologies are resistance tem - perature detectors (RTD s) and therm ocouples. A n RTD consists of a m etal plate or rod through w hich a current is passed. The resistance that the current encounters varies w ith the tem perature of the m etal. A therm ocouple consists of tw o dis- sim ilar m etal w ires joined together at one end. The voltage betw een the unjoined ends varies w ith the tem perature of the joint. B oth yield voltages that can be electronically interpreted to indicate the tem pera- ture of the surroundings. Therm ocouples are generally cheaper, though less accurate than RTD s. If the application does not require particularly tight tem pera- ture control, an inexpensive therm o- couple and a w ell-tuned PID loop should do the trick. B ut for process- es that w ill only w ork correctly at very specific tem peratures, it w ould be a m istake not to pay for the greater accuracy that an RTD affords. The cost of scrapping a batch of under-cooked or scorched products w ould eventually dw arf any savings in equipm ent costs. A fast sensor can also be w orth the extra cost. If the process requires a rapid succession of heating and cooling cycles, the tem perature sensor m ust be able to generate a reading before its too late to be of any use. D espite their cheaper pedigree, therm ocouples tend to respond faster than RTD s so if speed is the only im portant perform ance issue, choose a therm ocouple. >> M ist a ke # 2 : I nst a lling se nsors incor re ct ly Placem ent The best sensor can yield disap- pointing results if not installed cor- rectly. M agm eters, for exam ple, tend to generate noisy signals if the flow theyre m easuring is turbulent. B ends, junctions, and valves in a pipe can all cause turbulence, thus m agm eters w ork best w hen installed in sections of straight pipe. Tem perature sensors are also sensitive to placem ent. Even a highly accurate RTD tucked in the corner of a m ixing cham ber w ill only be able to detect the tem pera- ture of its im m ediate vicinity. If the m ixing of the m aterial in the cham - ber is incom plete, that local tem - perature m ay or m ay not represent the tem perature of the m aterial elsew here in the cham ber. Local tem perature issues are the classic m istake that hom e heat- ing contractors often m ake w hen installing household therm ostats. A ctllp05044Mistakes.qxd 4/18/2005 11:29 AM Page 3 THE FOUR BI GGEST M I STA KES I N I N STRUM EN TATI ON . . . A m ounting location closest to the furnace m ay be convenient for w iring purposes, but if that spot happens to be in a hallw ay or other dead air space, the therm ostat w ill not be able to determ ine the aver- age tem perature elsew here in the house. It w ill only be able to m ain- tain the desired tem perature in its im m ediate vicinity. The rest of the house m ay end up roasting or freezing. C ontroller perform ance Poor control also results w hen a sensor is installed too far aw ay from the associated actuator. A dis- tant sensor m ay not be able to m easure the effects of the actua- tors last m ove in tim e for the con- troller to m ake an educated deci- sion about w hat to do next. For exam ple, consider the process of flattening hot steel into uniform sheets by m eans of tw o opposing rollers (see Figure 1). A thickness sensor dow nstream from the rollers gauges the sheet and causes the controller to apply either m ore or less pressure to com pensate for any out-of-spec thickness. Ideally, the thickness sensor should be located adjacent to the rollers to m inim ize the tim e betw een a change in roller pres- sure and the resulting change in the thickness m easurem ent. O therw ise, the controller w ill not be able to detect any m istakes it m ay have been m aking soon enough to prevent even m ore of the sheet from turning out too thick or thin. W orse still, an appreciable dead tim e betw een the controllers actions and the resulting effects on the steel can cause the controller to becom e im patient. It w ill see no results from an initial control m ove, so it w ill m ake another and another until som e change begins to appear in the m easurem ents reported by the sensor. B y that tim e, the controllers cum ulative efforts w ill have already overcom - pensated for the original error, causing an error in the opposite direction. The result w ill be a con- stant series of up and dow n sw ings in the roller pressure and a lot of steel ruined by lateral corrugations. O f course, overall process per- form ance considerations arent lim - ited to how w ell the sensor feeds data to the controller during opera- tion. O ther factors to consider include ease of installation and tim e spent on the selection process, set up routines, and any labor-intensive m aintenance. Fortunately, som e instrum entation vendors design their sensors to accom m odate such challenges, thereby im proving perform ance before the system even goes on- line. A B B , for instance, offers a sw irl flow m eter that significantly reduces the need to install special upstream and dow nstream devices to accurately m easure the flow through a pipe. Protection A steel m ill is also a classic exam - ple of a harsh environm ent that can destroy inadequately protected sensors. Fortunately, the hazards posed by a m anufacturing process are generally obvious and can often be overcom e by installing a shield or choosing a rugged instrum ent. O ften overlooked, how ever, are the effects of w eather. O utdoor instrum ents can take quite a beat- ing from rain, snow , hail, and falling CONTROLLER D PISTON ADJUSTABLE ROLLER FIXED ROLLER OPTICAL THICKNESS GAUGE FI GURE 1 . POOR SEN SOR PLACEM EN T In this steel rolling exam ple, D is the distance betw een the steel rollers and the thickness gauge dow nstream . If D is too large, the controller w ill take too long to correct thickness errors and m ay even m ake m atters w orse by becom ing im patient. ctllp05044Mistakes.qxd 4/18/2005 11:29 AM Page 4 N . . . A N D HOW TO AVOI D THEM 5 Instrum ents m ust be grounded to provide a reference voltage for the data signals they generate. R elying on earth ground is risky since not all of the earth shares the sam e electrical potential. The resulting currents w ill interfere w ith the sensorssignals. ice. O ver tim e, outdoor instrum ents can fail slow ly unless enclosed in appropriate housings. B ut even the housings them - selves can cause problem s for the enclosed instrum ents, particularly tem perature sensors. If an RTD or therm ocouple is m ounted on the sam e piece of m etal that supports the housing, the housing w ill w ork like a heat sink w hen the am bient tem perature drops low enough. It w ill tend to draw heat out of the sensor and artificially low er its reading. The heat-sink effect w ill also tend to reduce the benefits of any internal heat that has been applied to prevent an instrum ent from freezing. C onversely, if a housing is equipped w ith fins intended to draw heat out of the enclosed sen- sor during w arm w eather, the fins m ust be m ounted vertically. O therw ise, the w arm air around the fins w ill not be able to rise aw ay from the housing (see Figure 2). G round loops W hile its generally a good practice to insulate a sensor from the ther- m odynam ic effects of its surround- ings, its absolutely critical to estab- lish electrical isolation. The m ost com m on electrical problem s due to poor installation are ground loops. G round loops occur w hen an extraneous current flow s through the instrum entation w iring betw een tw o points that are supposed to be at the sam e voltage, but arent (see Figure 3). The resulting electrical interference can cause random fluctuations in the sensorsoutput and m ay even dam age the sensors them selves. A s the nam e im plies, ground loops m ostoften occurw hen instru- m ents and theircables are ground- ed im properly or not at all. Interestingly, the best w ay to isolate a plants instrum ents from ground loop currents is to connect them together at one m aster grounding point. If thats not possible, a grid of grounding points m ust be spread throughout the plant, m aking sure that all points on the grid are at the sam e electrical potential. Insecure connections and inade- quate w ires can cause a voltage im balance in the grid and ground loops betw een the instrum ents connected to it. Even the orientation of an instrum ent can affect its perform ance. H ere, the sensor is enclosed in a housing designed to dissipate the heat it generates. The fins m ust be m ounted vertically to allow w arm air to escape. FI GURE 2 . POOR M OUN TI N G } Data signals to I/O panel } Earth grounds at unequal voltages } Instruments Ground loop currents FI GURE 3 . POOR GROUN DI N G ctllp05044Mistakes.qxd 4/18/2005 11:29 AM Page 5 3 THE FOUR BI GGEST M I STA KES I N I N STRUM EN TATI ON . . . A 4 >> M ist a ke # 3 : Ge ne rat ing gibbe r ish N oise G round loops are not the only source of noise that can distort a sensors readings. R adio frequency interference (R FI) is even m ore com m on in plants that use w alkie- talkies, pagers, and w ireless net- w orks extensively. R FI also results w henever a current changes, such as w hen an electrom echanical con- tact or a static discharge generates a spark. The sources of R FI noise m ust be elim inated or at least kept aw ay from the plants instrum entation if at all possible. R eplacing electro- m echanical equipm ent w ith solid- state devices w ill elim inate arc-gen- erated R FI. O r, it m ay be sufficient to sim ply relocate sw itch boxes and relays to instrum ent-free areas of the plant. If all else fails, it m ay be possible to passively shield the source of the interference or the instrum ents being subjected to it. Ignoring the problem is not an option, especially w hen the source of the noise is ordinary house cur- rent. A t 60 H z, house current oscil- lates slow ly enough to have an appreciable effect on som e processes. C onsider the steel rolling appli- cation again. A 60 H z noise super- im posed on the output of the thick- ness gauge w ill pass through the controller and induce a 60 H z oscil- lation in the roller pressure. If the sheet exits the rollers w ith a veloci- ty of six feet per second, those oscillations w ill appear as bum ps in the sheet appearing every tenth of an inch. W hether those flaw s are appreciable or not w ill depend on the am plitude of the original noise signal, the inertia of the rollers, and the tuning of the controller. PID controllers tuned to provide appreciable derivative action are particularly susceptible to the effects of m easurem ent noise. They tend to react aggressively to every blip in the m easurem ent signal to quickly suppress deviations from the setpoint. If a blip turns out to be nothing but noise, the controller w ill take unw arranted corrective actions and m ake m atters w orse. Filtering U nfortunately, it is not alw ays pos- sible to elim inate noise sources altogether. It is often necessary to filter the raw sensor data by aver- aging several sam ples together or by ignoring any changes less than som e sm all percentage. M any digi- tal instrum ents, like A B B s FSM 4000 flow m eter, com e equipped w ith built-in filters. H ow ever, it is a m istake to think that num ber crunching alone can fix all m easurem ent noise prob- lem s. Filtering tends to increase the tim e required to detect a change in the m easured value and can even introduce spurious inform ation into the signal. W orse still, it can m ask the actual behavior of the process if it is overdone. It is generally m ore cost-effec- tive in the long run to install sen- sors correctly and m inim ize the sources of interference than to rely strictly on m athem atics to separate the data from the noise. W hen con- structing a control loop, data filters should be applied in the final stages of the project, just before loop tuning. M ist a ke # 4 : Quit t ing t oo soon Even w hen the data filters are in place and the last loop has been tuned, the project isnt over. There are som e com m only neglected chores that should continue as long as the instrum entation system is in place. C alibration M ost instrum entation engineers know that a sensor m ust be cali- brated in order to associate a num erical value w ith the electrical signal com ing out of the transm itter. Yet all too often, the instrum ents are calibrated just once during installa- tion then left to operate unattended for years. The result is an insidious prob- lem know n as drift. A sensors out- put tends to creep higher and high- er (or low er and low er), even if the m easured variable hasnt changed. D eposition on the sensing sur- faces, corrosion in the w iring, and long term w ear on m oving parts can all cause an instrum ent to begin generating artificially high (or low ) readings. A s a result, the con- troller w ill gradually increase or decrease its control efforts to com - pensate for a non-existent error. A nalog instrum ents are particu- larly susceptible to drift, m uch like old FM radios. The slightest nudge on the dial could cause the radio to lose its signal. W ith m odern digital radios, the one true frequency for each station is digitally encoded at ctllp05044Mistakes.qxd 4/18/2005 11:29 AM Page 6 N . . . A N D HOW TO AVOI D THEM a fixed value. Sim ilarly, m odern instrum ents that em ploy digital sig- nal processing cant be nudged. They m aintain the sam e calibration in the field as in the lab. D rift can also be reduced by the choice of sensing technology. Tem perature sensors w ith m ineral insolated cables, for exam ple, are less prone to drift. D rift due to w ear can be elim inated entirely by choosing instrum ents w ith no m ov- ing parts, like A B B s sw irl and vor- tex m eters. A nd even w hen drift cannot be elim inated, recalibrating every sen- sor in the plant at intervals recom - m ended by their m anufacturers can accom m odate it. U nfortunately, project engineers are often so anx- ious to finish a job and get on w ith operating the process that they neglect such basic m aintenance. A rguably the m ost challenging sources of drift are those that vary over tim e. D eteriorating probes and m oving parts beginning to w ear out can slow ly change an instrum ents accuracy. So m aintenance calibra- tion is required periodically even if there are no know n issues w ith the instrum ent. Som e m anufacturers are recog- nizing the tim e and efforts involved in traditional recalibration exercises and are designing instrum entation products to sim plify m atters. For exam ple, the C alM aster portable calibrator from A B B provides in-situ calibration verification and certifica- tion of A B B s M agM aster electro- m agnetic flow m eters w ithout requir- ing access to the flow m eter or opening the pipe. Instead, the operator sim ply connects a C alM aster to the flow m eters transm itter and a PC . A W indow s interface guides the oper- ator through a series of tests to evaluate the status of the transm it- ter, sensor, and interconnecting cables. The tests are com plex, but so autom ated that the w hole cali- bration routine can be accom - plished in 20 m inutes. O nce the tests are com plete, C alM aster w ill evaluate the m eas- urem ents taken. If all satisfy the cali- bration requirem ents, then a calibra- tion certificate can be printed either at that tim e or later. These certifi- cates can then be catalogued in order to m eet auditing and regulato- ry requirem ents such as ISO 9001. A n added benefit of C alM aster is that it can be used as a diag- nostic and condition m onitoring tool. It autom atically stores all m easured values and calibration inform ation in its ow n database files for each m eter, thus m aintain- ing a calibration history log and m aking it easier to undertake long- term trend analysis. D etailed observation can give early w arning of possible system failure, enabling the m aintenance engi- neer to anticipate problem s and take proactive rem edial action. Such autom ated system s m ake routine verification of flow m eter cal- ibration and the traceability of infor- m ation m uch less cum bersom e and costly than in the past. In the w ater industry, for exam ple, such tasks form erly entailed m echanical excavation of the flow m eter result- ing in a disruption of the w ater sup- ply and a substantial investm ent in m anpow er and equipm ent. Pla nning f or t he roa d a he a d A ll too often, an expansion proj- ect begins w ith w eeks of w onder- ing w hy the existing instrum enta- tion system w as constructed the w ay it w as and w hy it doesnt m atch the projects original plans. To avoid this, future planning should be a part of your im ple- m entation process and also include thorough docum entation of w hats been done before. Som eone w ill eventually w ant to expand the project and w ill need to know exactly w hich instru- m ents have been placed w here, w hat the instrum ents w ere sup- posed to be accom plishing, and how they w ere installed and con- figured. Even if the instrum entation sys- tem is never expanded, it w ill even- tually have to be repaired. W ires break and sensors w ear out. A good inventory of the system com - ponents w ill indicate w hat needs to be replaced, but thats only half the battle. R eplacem ent parts m ust be acquired along w ith the technical specs necessary to install them correctly. A n ongoing replacem ent parts program is a m ust. Either the origi- nal vendor m ust m ake provisions for stocking replacem ents (or upgrades) for all the instrum ents theyve provided to date, or the project engineers m ust continue to m onitor their suppliers to m ake sure that spare parts rem ain avail- able. For hard-to-find instrum ents, it m ay even be necessary to m aintain an in-house supply of replacem ent parts, just in case. 7 4 ctllp05044Mistakes.qxd 4/18/2005 11:30 AM Page 7