INTRODUCTION

INTRODUCTION. The centrifugal pump is one of the simplest pieces of equipment from the controls and instrumentation point of view. It is a two port device with a well defined characteristic. Its purpose is to provide the necessary pressure to move liquid at the desired rate from point A to point B of the process. Figure 1-1 shows a 'generic' process with a centrifugal pump connected to deliver liquid from A to B.

Figure 1-2 shows the characteristic curve of an actual pump (a single stage vertical turbine pump) together with the characteristic curve of the process, known as the system curve. The intersection of the two curves defines the operating point of both pump and process. It would be fortunate indeed if this operating point is the one actually specified for the process. It is impossible for one operating point to meet all desired operating conditions since the operating point is, by definition, exactly one of an infinity of possible operating points. In fact the entire point of controlling the pump is to modify its characteristic so that its actual operating point is the one that is required at every instance in time.

Several definitions are presented in order to discuss the diagram:

Po = Differential pressure, or head, at the operating point of the pump and also of the process.

Qo = Flow rate, at operating point, of the pump and also of the process.

Ppm = Maximum differential pressure across the pump (at shutoff).

Qpm = Maximum discharge flow of the pump.

Plm = Static (Minimum) differential pressure between points B and A of the process.

The minimum static differential pressure of the process is frequently zero, as in a closed, circulating system. If the pump is in parallel with other pumps that are maintaining the system pressure, then P lm is greater than zero. It is clear from the outset that if P lm is greater than Ppm, no amount of process control can force the two curves to intersect. The pump is simply inadequate. How is process control like cutting off a rope? You can always cut off more, but you can't cut off less.

Assuming the pump is more than adequate for the process requirements at the moment, what is the best way to trim it back to the desired operating point, P1, Q1? There are three possible locations to place a valve: At the discharge, at the suction, and as a recycle valve. Each will be discussed in turn.

DISCHARGE THROTTLING. Since the pump exists to serve the requirements of the process, and one of the primary purposes of instrumentation is to adapt the equipment to the process, let us consider the pump from the point of view of the process. It can be viewed as a constant pressure device with an internal restriction. It is the restriction that gives it the "curve". It seems natural to put a valve on the discharge to further restrict the pump. This has the effect of rotating the curve of the pump/valve system clockwise around Ppm, as can be seen in Figure 1-3.

At this point I must warn the reader that we are about to encounter a paradigm shift. (!) The combination of pump and valve will be presented as a "black box" with a single characteristic curve which I shall term the "modified" pump curve.

The more traditional way of looking at the situation is from the point of view of the pump. It sees the process system curve as having rotated counter clockwise around Plm. Figure 1-3 shows that the flow, Q1, is the same for both cases. The difference between the two pressures is the Delta P across the valve. Since the purpose of the pump is to serve the process requirements, and the purpose of the valve is to adapt the pump to the process, it makes sense to consider the valve to be part of the pump system and to use the modified pump curve rather than the modified system curve in our discussion. In any case it can be seen that a discharge valve can be used to achieve any operating point on the system curve so long as the point is below the pump curve.

SUCTION THROTTLING. The second possibility for control using valves is to place the valve in the pump suction line. This would have an identical effect on the characteristic curve, but the method has a fatal flaw cavitation. Cavitation is a phenomenon that occurs when the pressure of a liquid is reduced below its vapour pressure and brought back up above the vapour pressure again. Bubbles of vapour form in the liquid and then collapse upon arriving at the higher pressure region. The collapse occurs at sonic speed ejecting minute jets of extremely high velocity liquid. Wherever these jets impinge on a solid surface extreme erosion occurs. Over time even the hardest materials will be destroyed. Therefore it is of utmost importance that this pressure reduction never occurs. It is prevented by having sufficient pressure available at the pump suction so that the pressure drops that occur as the liquid is drawn into the eye of the impeller are at all times above the vapour pressure of the liquid at its current temperature.

An explanation of the term Net Positive Suction Head (NPSH) is in order. This is the pressure of the liquid at the pump suction in terms of feet or meters of liquid head above the vapour pressure of the liquid. The actual NPSH under operating conditions is called NPSHA and the minimum required by the pump to prevent cavitation is called NPSHR. Clearly NPSHA must be greater than NPSHR to avoid cavitation. It is safe to leave a margin of about one meter.

These peculiar definitions are very reasonable in terms of the pumps actual characteristic but they cause some problems to the controls engineer. It means that the gauge pressure equivalent of a given NPSHA is proportional to the density of the liquid and is also affected by its temperature. The vapour pressure can rise dramatically as the temperature rises. This means that the NPSHA can fall without a noticeable change in pressure.

Anything that would reduce the net positive pressure at the pump inlet below the NPSHR must be absolutely avoided. Thus suction throttling is never used to control pump flow.

RECYCLE CONTROL. The third remaining possibility for pump control with valves is to bleed some of the discharge flow back to the pump suction or to some other point on the supply side. Once again we can view the result as a modified system curve or as a modified pump characteristic. Figure 1-4 shows both. Each curve is a rotation of the original: The modified system curve as a clockwise rotation around Plm. Note the little "tail" at the left of the modified system curve. This represents the flow through the recycle valve before the discharge check valve opens to the process. The modified pump curve has a counter clockwise rotation around the hypothetical intersection of the pump curve with the flow axis.

This family of curves shows several problems with recycle control. Firstly, the pump is not rated to discharge more than the flow rate at the end of the curve. It is possible, of course, to run the pump with a wide open discharge, minimum D P, but it is unhealthy for this particular pump to run at such a high rate. Excessive flow may cause cavitation damage. (Excess flow cavitation is not caused by NPSH problems but by high velocity within the internal passages of the pump.) This restriction means that the minimum discharge pressure may not be lower than the one corresponding to the maximum flow. In other words, the modified pump curve cannot reach all points on the system curve.

Secondly, although many pumps are capable of operating near zero discharge pressure, the very flat pressure vs. flow curve for much of the lower range for most pumps means a change of flow has very little effect on the discharge pressure. Thus it would take a very large amount of flow to produce a small drop in pressure. In control terms this means that control would be very 'sloppy'. Discharge throttling on the other hand, allows the pump to develop the head that 'suits' it. The unwanted pressure is dropped across the valve. (Note that the curve for this particular pump rises rather steeply. It will be more easily controlled than most.)Thirdly, this method is often inefficient. Figure 1-5 shows a system curve, a pump characteristic, a discharge modified characteristic, and a recycle modified characteristic. Above these is a pump power requirement curve. In the case of discharge control, the pump is adapted to the process by dropping its discharge pressure. If one follows the flow line vertically to the actual pump curve and then beyond to the power requirement curve one arrives at its power requirement. In the case of recycle control, the pump is adapted by reducing the discharge flow. Following the pressure line to the right to the actual pump curve and then upwards to the power requirement curve one arrives at the power requirement for recycle control. Note that the power requirement curve tends to slope upward as flow increases. Therefore recycle control consumes more pump horsepower than discharge throttling when both achieve the same operating point. This is not always so. If the power requirement curve were flat, there would be no difference. Notice on the curve that there is a slight drop in horsepower near the right hand end. If circumstances were such that the operating point corresponded to a downward sloping power curve, recycle control would be more efficient. This is rare.

SPEED CONTROL. There is, of course, one other means of adapting a pump to the changing demands of the process: Speed control. The virtue of this method is that it reduces the energy input to the system instead of dumping the excess. Figure 1-6 shows a system curve superimposed on a family of curves for a variable speed pump. The curves reach all parts of the system curve below the full speed curve. Therefore this is an effective means of control. Note, however, that these curves have one feature in common with recycle control: At the far left end of the system curve the pump curve and the system curve are almost parallel. (The particular pump chosen for this example has a rather steeply rising curve near shutoff. Most are considerably flatter.) In mathematical terms this means that the intersection is poorly defined. In practical terms this means that it is difficult to maintain a precise operating point and that control is 'loose' at high turndown.

In practice, variable speed drives for centrifugal pumps are still relatively uncommon. For small pumps the power savings are not significant and for large pumps the associated electronics become very expensive. Also, they do not have the high reliability of valves. Variable speed steam turbine drives are quite common in the larger horsepower ranges. Electric variable speed drives are used in certain specialized applications such as pumps that are embedded inside a high pressure vessel. In such cases there are no alternatives.

RIDING ON THE CURVE. Last but not least: No control at all! The fact is that the majority of pumps in the world run with no control at all. The exact flows and pressures are not critical and the pump has been reasonably well selected. The discharge pressure will rise to partially compensate for increased back pressure. It falls as the back pressure decreases so that the flow does not increase as much as it otherwise might. The pump is allowed to "ride on its curve". When this situation is acceptable, leave well enough alone and don't try to fix what ain't broke. (Be careful though, the machine may still require minimum flow and other protections as detailed in the section on Machine Protection.)

MEASUREMENT. The appropriate measurement for the controller depends on the demands of the process. Flow control is a frequent requirement. Two rules guide the location of the flow measurement: Make sure that side streams are included or not, as required, by the measurement and make the measurement at the highest convenient pressure. The latter requirement is to avoid any possibility of flashing or cavitation within the measuring device. In general the best place to measure flow from a centrifugal pump is between the recycle Tee and the discharge throttling valve. The exception is when the discharge is at an extremely high pressure and the suction has adequate NPSHA. In that case a suction measurement may be best.

Level control of a vessel is one of the most common requirements1. The vessel may be either upstream or downstream. It is quite possible to connect the Level Controller directly to the discharge valve. Frequently, however, the vessel serves to buffer a downstream process from upstream flow variations. In that case it is not desirable for level control to be precise. Perfect level control implies that the flow out is exactly equal to flow in at all times. Often it is desired that the downstream flow remain as uniform as possible while keeping the level within bounds. In simple terms, it is desired that the flow out is the average of the flow in. The vessel absorbs the instantaneous differences. This simple requirement is more difficult to accomplish than it may seem and deserves a discussion entirely of its own. A simple arrangement that is often satisfactory and is widely used is to have the Level Controller cascade to a Flow Controller on the pump discharge. The flow loop keeps the discharge 'constant' while the Level Controller gradually raises or lowers the setpoint as the level in the vessel rises or falls.

Another common requirement is to control the pressure of either upstream or downstream equipment. The tap for the pressure transmitter should be connected at the point where it is desired to control the pressure. Note that a pressure tap between the pump and a discharge throttling valve is probably meaningless. A careful look at many pump curves will show that the characteristic near shutoff is quite flat and may even slope downward. Pressure control cannot be accomplished when the pressure curve is flat. If the slope is the 'wrong' way, control will work backwards and drive the valve away from the set point. In this case the minimum flow should be set so that the pump cannot operate in the positive slope region of the curve. (It is, of course, possible to reverse the action of the controller so that it can operate to the left of the peak. But in that case, what will happen if the operating point moves to the right? It is extremely difficult to design control systems that can operate continuously along a characteristic curve that has a local minimum or a maximum in it.)

There is a second, more serious, problem with pressure control. Centrifugal pumps are essentially constant head machines. The discharge pressure for a given pump rotating at a fixed speed is proportional to the density of the liquid. This means that if the liquid has a constant density, the discharge pressure is constant. The "curve" of the pump curve is produced by losses and other affects caused by flow. Unless there is a flow through the system, there is only one pressure and that is the shutoff pressure. If it is desired to control the pressure of a vessel being charged by a pump, it is best to pressure control a valve at the outlet of the vessel and let the pump ride on its curve. If the vessel must be dead ended, only recycle flow at the pump can control pressure to a setpoint.

ON / OFF CONTROL. On/off control is used in many situations where the object is simply to move a liquid from point A to point B and the exact pressure or flow rate is unimportant. A typical example is the sump pump. The simplest arrangement employs a level switch with a very broad deadband. This is used together with a Hand/Off/Auto switch to turn the pump on and off. The schematic is shown in Figure 1-7. The LSHL contact opens when the level is below its setpoints. "M" represents the motor contactor which energizes the motor whenever the contactor is energized. "M" also represents the auxiliary contact that is closed whenever the contactor is energized.

If it is important that the level never goes beyond the upper or lower setpoints, the Start/Stop arrangement is preferred. It is illustrated in Figure 1-8. The process sensing switch has a separate output for the upper setpoint (On) and the lower setpoint (Off). (Two switches may be required.) The manual switch consists of a Start and a Stop button or a combined Start/Run/Stop selector with a spring return to centre. The operator may start or stop the pump whenever the level is between the two setpoints. He cannot stop it when the level exceeds the high setpoint unless he locks it out. He cannot start the pump below the low setpoint. A variation of the circuit places the left connection of the start button to the left of the low level switch. With this arrangement it is possible to drain a vessel below the low set point by holding the start button on. The pump will stop as soon as the button is released.

With both of these arrangements, there must be sufficient deadband between the high and low setpoints to make certain that the pump does not cycle on and off too rapidly. Excessive wear of both the motor and its starter will result if this occurs. Rapid cycling is a sign of an over-sized pump.

MACHINE PROTECTION. Once the process requirements have been met, the attention of the process control engineer turns to protecting the equipment. Centrifugal pumps are fairly undemanding. In general they have only two requirements: that the NPSHR is met at all times and that a certain minimum flow is maintained. To meet the first requirement is generally a piping design problem. In cases of doubt, a low pressure shutdown switch may be added to the suction line. A second look at the explanations of NPSH, above, shows that determining the setpoint of the switch is not necessarily a simple matter if there is any possibility of the liquid density changing. Things get even more complicated if the vapour pressure is very sensitive to temperature. A rise in temperature that causes the liquid to boil will cause the net positive pressure to fall to zero even though there is an increase in actual pressure. LPG and LNG pumps are notorious for NPSHA problems. Fortunately most pumps can tolerate brief periods of cavitation without noticeable damage.

When a pump is taking suction from a vessel, a low level shutdown switch is essential. The switch, or transmitter, must be separate from any level control devices.

To meet the second requirement, minimum flow, is somewhat more difficult. A centrifugal pump adds energy to the liquid that the moving liquid carries away. If flow is blocked, the temperature within the pump will rise steadily until the liquid boils (net positive pressure is now zero). Damage to the pump is quite likely. For this reason some form of minimum flow is almost always included on larger machines. The simplest arrangement is a fixed restriction orifice on a line leading back to the supply side of the pump. The preferred destination of the recycle flow is back to the vessel from which it came. This allows the heat to dissipate before it is recycled back into the machine. Restriction orifices have two drawbacks: They waste energy when the process demand is sufficiently high to meet all minimum flow requirements and also they limit the maximum pump output.

A more efficient method of recycle control requires that the discharge flow of the pump itself is measured, and that a valve in the recycle line is opened when the process does not draw the required minimum flow. The most straightforward way to accomplish this is shown in Figure 1-9. Note that the recycle line tees off upstream of the control valve. It is precisely when the control valve is closed that the recycle is needed. There is a small problem with controlling the minimum flow in this way: The measurement orifice in the discharge consumes energy and also slightly reduces pump capacity. A second problem is that the actual signal being measured is the D P across the orifice plate. Since flow varies as the square root of D P, a minimum flow of 40% of maximum flow implies a controller whose set point is only 16% of the measurement range. A typical instrument accuracy is 1%. Therefore an error of 7% of the setpoint can be expected. Fortunately the minimum flow need not be held very accurately. Recycle control is sometimes accomplished using a local pneumatic controller mounted directly on the valve. Note: Always use a fail-open valve.

Various schemes have been devised to infer the required valve setting from the net discharge flow measurement. These require the flow downstream of the recycle Tee to be subtracted from the required minimum flow. The recycle valve is then opened in proportion to the difference, if it is positive. To do this accurately one must know the valve and actuator characteristics. There is no feedback to confirm that the correct flow is occurring. Since the flow is usually above the minimum flow, the valve is usually closed. This will cause the controller to wind up and be slow in responding when a low flow condition suddenly arises. Fortunately pumps can tolerate short periods of low flow so this is not a problem.

One method of minimum flow control that is occasionally proposed is to put a flow control loop on the recycle line with the set point equal to the minimum flow. This solution is worse than a fixed restriction. When discharge flow is high, the discharge pressure falls. Flow through a fixed orifice will reduce somewhat. A flow control loop will open the valve further to maintain constant flow precisely when it is not needed. At this point the operator will be tempted to manually close the valve. Then, when a discharge blockage occurs, there will be no minimum flow at all!

There are a number of devices available, called Automatic Recirculation Valves, or ARC valves, that combine the functions of net discharge measurement, recycle control, recycle valve and discharge check valve all in one device. These devices can be very effective but they suffer from one drawback: lack of flexibility. In cases where the pump and process characteristics are well known, they can be an ideal solution. Pipelines, for example, have many identical pumps operating under steady conditions. Once the correct components are known, application is routine. It must be kept in mind, however, that both the process and pump data provided to the controls engineer for a new facility are often tentative. ARC valves have very little margin for error when the reality turns out differently from the theory. One particular problem that can occur with the older style ARC valves that operate in an open/close mode, and even with some that modulate, is instability. It occurs as follows:

The discharge valve begins to close due to a reduced process demand. The ARC valve senses the reduced flow and opens the recycle valve. The pump discharge pressure drops. The discharge Flow Controller senses that it is being starved of flow and opens the discharge valve. The ARC valve sense the increased flow and closes. The pump pressure rises. The discharge valve closes. The cycle repeats itself.

Note that ARC valves are not positioned by conventional actuators. They are positioned by the process liquid itself and are capable of very rapid action. Instability results in violent slamming of the recycle valve, scaring the operators and severely damaging the reputation of the controls engineer. Very little can be done at this stage other than to remove the ARC valve and to attempt to modify its characteristic by changing the spring or boring out the recycle ports. The latter spoils the hardened seats required in high pressure drop applications and leakage is inevitable. Boiler feed pumps seem to be especially prone to these problems. Note that ARC valves are quite expensive and often cost more that a complete flow control loop. They are, however, extremely effective, and simple, under the right circumstances. Their use often simplifies the piping arrangement and essentially eliminates routine valve maintenance.

ARC valves are best bought as part of the pump package. In this way the responsibility for ensuring that they match the pump rests with the party that is most familiar with it.

The pump curves used in this article represent an actual pump but are by no means typical of all pumps. Multistage pumps, in particular, may have little quirks in the curves that can complicate controls. If the characteristic curve droops as it approaches the zero flow axis, (the shutoff pressure is less than their peak pressure) the minimum flow setting must be well to the right of the peak or severe instability can result. Boiler feed pumps discharge into a compressible volume. If they have a reverse slope near shutoff, they may experience surge much like a centrifugal compressor does. Note that API STD 6102, the American Petroleum Institute standard for centrifugal pumps, explicitly bans a drooping characteristic.

SEAL FLUSHING and COOLING. Pumps in certain special services require flushing and/or cooling fluids to be injected into the seals. The details are provided in API STD 6102, Appendix D. In general the instrumentation is rather simple, consisting of rotameters, pressure gauges and thermometers.

In certain hazardous services, sealing becomes a more complex issue. If the danger of a seal leak is sufficiently serious, specialized leak detection may be required. One simple method is the installation of a pressure switch, or better yet, a transmitter, between the tandem seals. This can then be connected to the plant alarm system.

SAFETY. Centrifugal pumps are not generally hazardous pieces of equipment. However, there is one special safety consideration whenever a pump is drawing volatile hydrocarbons or other flammable liquids from a vessel with significant capacity. (API RP-7503, defines this as five tons.) Volatile liquids have a low viscosity and seal leaks are not uncommon. The leaked liquid often catches fire and it is absolutely essential that the pump be shut down to prevent feeding the flames further. In such situations it is desirable to have a remotely controlled block valve between the pump suction and the source vessel. This valve and its actuator should be fire safe. Since closing the valve can cause low flow damage to the pump, it must have a limit switch to shut down the pump whenever the valve is not fully open. It should also have both opened and closed status indication in the control room so the operator can be fully confident that the valve is open when the pump is running and that the valve is closed when a hazardous situation exists. If the block valve has an electric actuator, it is a good idea to have an alarm on the main panel to indicate if there is a power failure, if the local switch is not in the 'Remote' position, or if there is any other reason the valve might not work when called upon to do so. In extremely critical processes, one may wish to interlock the pump so that it cannot start unless the valve is in working condition.

Any indoor pump in flammable service should have adequate fire detection in the building. Ultraviolet detectors are preferred because they are sensitive to flame. They are extremely fast acting since they do not depend on heat buildup or the generation of smoke. (There is an exception to this rule: If the flammable material produces a lot of smoke, it may obscure the vision of the UV detector. In such a case one might be advised to install both smoke and UV detectors.) A certain amount of care must be taken when UV detectors are installed. They are sensitive to sunlight and to welders. The sensitivity to welders is probably a good thing since it forces all welding to be co-ordinated with the control room. The sensitivity to sunlight means that they must be positioned so that they are unlikely to 'see' the sun. The usual position is high up under the eaves of the building in diagonally opposite corners. This is not always fool proof. The author is aware of one case where a pipeline compressor was shut down because a UV detector saw a welder working out in the parking lot. The welder was directly in line with a gap around a pipe that went through the building wall. "Smart" combined UV/IR detectors are becoming available that are able discriminate between sunlight, welding arcs and fire. This type is also suitable for outdoor use.

Fusible link sprinkler systems are extremely reliable and can contribute greatly in cooling down a fire that is too hot to approach. Their drawback is that they only become active once considerable heat has been developed. In critical applications they are best used together with a faster detection system.

It may be worthwhile installing flammable vapour detectors near the base of large pumps if leakage is a possibility.

Never overlook the placement of check valves. This is a safety related issue that should not be left to other disciplines as check valves are an integral part of the functioning of many control schemes. It is generally self-evident that parallel pumps need check valves on each individual discharge. This check is also needed downstream of the control valve on single pumps. When the pump is not running, the discharge valve will most probably go wide open. A reverse flow could have some peculiar effects on the upstream process. A check valve is also required downstream of the recycle valve if a fire safe valve is necessary. Any time the fire safe valve isolates the pump from the supply vessel, the recycle valve will open wide. ARC valves should be checked to make sure all necessary check functions are included.

ACCESSORY INSTRUMENTS. Centrifugal pumps require few accessory instruments. Since the purpose of the pump is to develop pressure, it is a good idea to have a pressure gauge on the discharge. If the application requires a low suction pressure interlock, a pressure gauge should also be provided at the suction. It would be nice to have a local flow indicator but they are invariably expensive and inconvenient to install so they are rarely used. A thermometer on the suction may serve to warn of cavitation if the vapour pressure is temperature sensitive.

PARALLEL PUMP INSTALLATIONS. Centrifugal pumps are frequently operated in parallel. Their smooth operating curve allows this to be done without complication. If it is intended that the pumps are usually operated individually and not simultaneously, it is sufficient to have a common discharge throttling valve and suction block fire safety valve. However it is essential that each have its own recycle arrangements. Do not be swayed by the argument that the two pumps will never be run simultaneously. The most obvious reason for simultaneous operation is to switch from one to the other so that maintenance can be done without shutting down the process. In this case the pump that is being started will be operating against a blocked discharge check valve and is in no position to make use of a common recycle valve. Remember that the throttling valve is there to serve the process but the recycle is there to protect the machine. You don't share seat belts do you?

Parallel variable speed pumps obviously have individual controls. The most effective arrangement is to provide constant flow controls to the majority of the pumps. The setpoints should be at the peak efficiency for each individual pump. The remaining pump should have its controller set to handle the swings. Actually this an example of the complex subject of Supply and Demand Control and deserves a discussion of its own. Note that is meaningless to have two pumps each on pressure control pumping into the same header. They will not share the load.

SERIES PUMP INSTALLATIONS. Sometimes centrifugal pumps are operated in series. The usual situation is when a multistage pump has an NPSHR greater than what is available. In such a case, a single-stage pump with a low NPSHR is used as a booster. This is common with boiler feed pumps especially if the pump is drawing hot water whose vapour pressure is already elevated.

Process demand control is applied to the high pressure pump. The booster pump should be on discharge pressure control. The author was involved in one situation where oil field injection water was drawn from a cistern connected directly to a river. In this case the booster pumps were pressure controlled by recycle back to the cistern. This allowed the recycle water to keep the water in the cistern agitated, preventing an accumulation of silt.

It is not unusual for a group of booster pumps in parallel to supply a group of high pressure pumps in parallel. In such cases care must be taken to ensure that the various operating combinations are matched in capacity.

Every individual pump in a series installation must have its own minimum flow arrangement.

SUMMARY. Figure 1-10 shows a complete set of instrumentation for a typical centrifugal pump application. The drawing illustrates a pump drawing volatile hydrocarbons from a large surge vessel. The following features are illustrated:

A level / flow cascade loop on the pump discharge to provide process control. A check valve on the discharge downstream of the control valve to prevent reverse flow when the pump is shut down. A fire safe motor operated valve (MOV) in case of seal leakage and fires. An interlock from the MOV to stop the pump if the valve is not fully opened. A low level interlock from the vessel to stop the pump if the vessel loses its liquid seal. A pressure gauge on the suction to indicate adequate NPSHA. A thermometer on the suction to indicate potentially high vapour pressure. A minimum flow recycle loop back to the vessel. A check valve on the recycle line to prevent reverse flow when the pump is shut down, especially when the fire valve is closed. A pressure gauge on the pump discharge to indicate that the pump is working.

INTRODUCTION. It would seem that controlling a vessel should be a very simple matter -- They don't really do anything! But then, if they didn't do anything why are there so many of them? And why do they have so many different names? Going through a typical set of Piping and Instrumentation Diagrams (P&IDs) I see the following vessels:

Degassing Drum Gas Separator Storage Tank

Feed Flash Drum Reflux Accumulator Day Tank

Surge Drum Suction Scrubber Slug Catcher

Lube Oil Separator Head Tank Deaerator

Although each of these is essentially a simple vessel or tank without any special internal structure, each serves a different purpose. Once it is clear what the purpose of a piece of equipment is, and how it functions, it will also be clear how to control and protect it. Different purposes require different controls.

SURGE TANKS. The most common function of a vessel or tank is to match two flows that are not identical in time but are expected to average out over the long run. Take a feed surge drum, for example. Flow into the unit is more or less steady but is subject to interruption. The flow to the processing unit should be as constant as possible, avoiding sudden change. Nevertheless, it, too, may be subject to interruption due to downstream conditions.

The purpose of the surge drum is to maintain sufficient inventory to feed the process and to maintain sufficient void capacity to continue receiving feed as it arrives. Clearly the tank must be large enough to accommodate any normal discrepancies between input and output over a reasonable period of time. Between the upper and lower bound, the exact value of the level does not matter.

Two separate control parameters are implied: Level and flow. Level control is no problem. Greg Shinskey 1 refers to "The easy element -- capacity". A high gain, level controller connected to a valve at either the inlet or the outlet will maintain the level very accurately at its setpoint. The only problem with this approach is that it absolutely defeats the purpose of the vessel. The same effect would be achieved by blocking in the vessel and bypassing the inlet directly to the outlet.

To control flow alone is also quite simple. A flow controller at the outlet, properly tuned, will maintain a steady flow to the process. Unfortunately, there is nothing to make this flow equal to inflow. It will not even equal the average inflow unless there is something to make it do so.

What we need is an instrument that measures the accumulated error between inflow and outflow. The tank itself is that instrument!

Level = Starting Level + (Inflow - Outflow) dt / Tank Area

(To a process controls engineer, every piece of equipment is just a big, non-tuneable instrument!) The level transmitter only transmits the process value to the control system. If we now cascade the output of the level controller to the flow controller, we have a system that has one process variable: Accumulated flow imbalance. It has only one point of control: Outflow to the process.

To start this simple process:

Fill the tank about half full.

Give the level controller the current level as its set point. (PV tracking does this automatically.)

Switch the flow controller to automatic with an estimated average flow as its setpoint.

Switch the flow controller to cascade.

Switch the level controller to automatic.

The control system will keep the flow "constant" but that constant varies in response to the imbalance between outflow and inflow. It is not important that the initial estimate of average flow be exact. A low guess will result in the tank level rising a little. A new, higher, estimate will result and the outflow will be adjusted accordingly. In the long term the average flow out is not an independent variable at all. It will be exactly equal to the average flow in. This can be accomplished at any arbitrary tank level. The level setpoint is based on the operator's estimate of the nature of the flow interruptions and whether the most probable upset will require additional flow or void capacity.

Should a pump be necessary to transfer the liquid from the vessel to its destination it should be placed between the vessel and the flow measurement. Further information on the control of pumps is found in Controlling Centrifugal Pumps2 . This article also includes a section titled "On/off Control" for less critical level applications.

There is a long discussion on the special requirements for level control of steam heat exchangers and condensate receivers in Controlling Steam Heaters3.

Surge drums are sometimes used for gas. The abrupt flow variations of a Pressure Swing Absorption (PSA) unit, for example, often need to be smoothed out before the tail gas can be introduced into a down-stream process. In these cases, pressure takes the role that level has in a liquid process. That is, a pressure/flow cascade is the appropriate solution.

TUNING SURGE TANK CONTROLLERS. Since the exact level of a surge drum is not important, the controller can be tuned very loosely allowing the level to rise and fall in response to any short term imbalances. This exactly serves the purpose of the surge tank; tight tuning defeats it. There is a non-linear control algorithm which specializes in the type of loose control required by surge tanks. One common name is the "gain on error squared" controller. Figure 6-2 shows its characteristic. The controller responds to small errors with a small gain; it responds to large errors with a large gain. This means that in the vicinity of the setpoint, the controller allows the level to drift freely and the flow to remain almost constant. With luck, the level will average out again before the deviation from setpoint is too great. If the level changes far from the setpoint so that the danger of running out of capacity exists, the controller responds with a strong signal and rapidly brings the level back to near setpoint.

Another form of non-linear controller is also available: The notch or gap controller. This algorithm has the gain divided into three segments by two break points. The middle segment, on either side of the setpoint, has a low gain to avoid excessive action while the outer segments have a higher gain for a rapid return. It has the advantage of allowing the user to set the breakpoints and gains below the setpoint differently from those above. Its disadvantage is that it has four tuning constants instead of only the one found in the gain-on-error-squared controller. Some gap controllers have a zero gain in the centre segment. This is totally useless as the controller will never bring the level back to the setpoint. (No gain, no action.) Instead it will tend to use either the upper or lower breakpoint as its effective setpoint and return the level with a high gain. It should be noted that an abrupt change in gain does not imply an abrupt change in valve position, only a change in the rate of movement.

A simple proportional mode controller is sufficient for many surge drum applications. A slow integral may be used to bring the level back to the setpoint during a prolonged change in flow rate, but it should be turned off if cycling results. Do not use the derivative mode! Besides amplifying noise, derivative provides tight control by cancelling out the integrating capacity of the tank and thus defeating its purpose. A tuning rule I have heard of, but have not tested myself is

K = D F/F * D L/LWhere K = controller proportional gain

D F/F = the proportion of flow variations in the uncontrolled flow

D L/L = the proportion of level available for surge. This is the distance between the level setpoint and the nearest alarm.

This formula attempts to put the loosest level control consistent with keeping the level away from the alarms. There is a catch, however: It is necessary to predict the amount of flow variation to be expected in the future. Of course it is also necessary to do this to a certain extent when the vessel is sized.

SUCTION SCRUBBERS. A compressor suction scrubber is an example of a vessel whose purpose is to separate, collect, and dump relatively small quantities of liquid from a gas stream. The following conditions generally apply:

Precise level control is of no value.

Smoothness of liquid flow is of no value.

The liquid flows to some form of drain.

The average liquid flow is quite small.

The pressure differential across the valve is high.

Relatively large slugs of liquid occur occasionally.

The last three conditions would result in a valve that is usually operating near its seat with a high D P. It would experience severe erosion resulting in a short, unhappy life. The solution is to control the valve in on/off or "snap acting" mode. There are several ways to accomplish this. The simplest is to tune the controller to a very high gain. This would cause the valve to spend almost all its time in the full open or closed position. Unfortunately the high-gain controller would also try to maintain accurate level control by rapidly switching the valve between these extreme positions. Any saving in seat erosion would be cancelled by a high rate of stem and packing wear. The same response can be achieved by using a simple level switch connected to the control valve via a solenoid. (Pneumatic level switches tubed directly to the valve actuator diaphragm are also available.) A level switch can be viewed as a controller with an extremely narrow proportional band (0%!) and consequently an extremely high gain (100% / 0% = ).

Selecting a switch with a broad deadband results in a great improvement. The valve now remains fully open until a significant reduction in level is achieved. It then remains fully closed until the level substantially rises. With this arrangement it is possible for the valve to have both long life and peak capacity. Recent experience indicates that transmitters are more reliable instruments than switches and also demand less maintenance4. If transmitter is used the deadband function is accomplished through logic in the control system. This would have the added advantage of allowing the operator access to the high and low setpoints. In some ways the suction scrubber acts as the exact opposite of a surge drum -- it collects slow dribbles of flow and releases them as intermittent surges.

Sometimes there is a third option -- specialized liquid dump valves. These behave somewhat like steam traps in their ability to pop open in the presence of liquid and snap shut in the presence of vapour. Since they are not general purpose instruments, it is best to use them only when there is an opportunity to test their performance; the vendor should be consulted. These devices might be very cost effective in packaged equipment such as on the discharge receiver of an instrument air compressor.

STEAM DRUMS. The purpose of a boiler steam drum is to provide space in which the water and steam may disengage. Since the drum serves at high pressures and temperatures, perhaps up to 3600 psi and 1000F (25 MPa and 540C), it is expensive to manufacture and there is considerable economic incentive to keep it as small as possible. The techniques of boiler feed water (BFW) control can be applied whenever extremely tight level control is a requirement.

The level of the feedwater in the steam drum must be kept above the bottom of the drum or a catastrophic explosion may result. It must also be kept below the steam outlet or liquid water will be carried over. Water droplets will damage superheater tubes, turbine blades, and other equipment. The diameter of the steam drum, and hence its cost, is determined largely by the ability of the control system to keep the water level within bounds.

Thus level control of a steam drum has exactly the opposite purpose of that of a surge drum: The water level must be kept within an extremely narrow band and tight control is of essence. It is a simple matter to maintain tight level control... use both the proportional and integral modes and turn up the gain! Figure 6-3, Single-Element BFW Control, shows this very simple arrangement. As always, there are problems. Firstly, high gain means extremely rapid swings in flow rate. The BFW pumps suffer under that type of abuse. There is a second problem, peculiar to boilers, called "swell". Swell is the phenomenon in which a rise in steam demand causes a drop in pressure. This in turn results in a rapid boilup within the tubes which causes the water level to rise. Paradoxically, an increased steam removal rate causes a rise in level due to the swelling of the steam bubbles. The level controller responds by reducing BFW flow at the very moment it is needed most. The swelling water soon collapses as the steam rises to the surface. Now the controller reverses its response and adds a large amount of essentially cold BFW into the system. This causes the water temperature to fall. The cooler water shrinks, lowering the level further. The use of single-element control is not very highly recommended for boilers!

The disturbance to the level is caused by a change in steam withdrawal rate. Since this is a measurable quantity, feed forward can be applied to the level controller output. Figure 6-4 shows how this is accomplished. The compensated steam flow is added to the output of the level controller. Thus a rise in steam withdrawal and the swelling of the water is accompanied simultaneously with a surge of cold BFW. Ideally the two cancel out exactly and the controller sees no change in level at all. They will not cancel out exactly for two reasons: Firstly, there is no reason why they should. One effect or the other will predominate. They won't even be simultaneous. Secondly, the BFW flow can only equal the steam withdrawal if the range of the valve is exactly equal to the range of the compensated steam flow. Since these two functions must be exactly equal over the entire operating range, it means that the valve must be perfectly linear and that its D P is absolutely constant. Not likely! So the level controller still has some work to do to keep the accumulated error at zero.

The rather farcical suggestion in the previous section, piping the inlet to outlet and bypassing the vessel, suggests a solution to the valve linearity problem: Use the measurement of the steam leaving the boiler as the setpoint to a BFW flow control loop. The level should remain constant once the shrinking and swelling have reached the new equilibrium. This simplistic solution overlooks a basic principle of process control: No two measured quantities are ever identical. In other words, the two flows will never be the same and the level will rise or fall at a rate proportional to the difference. Since level is a measure of the accumulated difference, a level controller is used to correct the BFW flow. What I have just described is the classic three-element boiler level control arrangement as shown in Figure 6-5.

The diagram also illustrates a few other features. Compensation has been applied to account for the effect of pressure on the steam density and its effect on the level transmitter. BFW flow is sometimes temperature compensated since it is most probably preheated and its temperature may vary. For a temperature change from 0C to 300C (32F to 572F) the specific gravity changes from 1.000 to 0.712 and a measurement error of 15% will result.

This detailed exposition of boiler level control is presented only to provide an example of how extremely tight level control can be accomplished when necessary. Boiler control is a rather broad subject and many articles and textbooks have already been published concerning it.

CONTROLLING LIQUID INTERFACES. It is generally assumed that level control refers to the control of a gas/liquid or vapour/liquid interface. It ain't necessarily so. An interface can occur between any two immiscible fluids. Since all gases are miscible with each other in all proportions, interface level control is always taken to mean the interface between two liquids such as oil and water.

Figure 6-6 shows an example of a boot on a crude oil separator. This vessel serves three purposes: It is a gas/oil separator, a feed surge drum and a water separator. A real vessel in this service probably contains inlet baffles and demister pads. Each of the three phases must be individually controlled. But it is possible for the gas phase to discharge to an externally pressure controlled system or even to atmosphere. (Possible yes, acceptable no.) The key to understanding the function of any separator is to think in terms of a constant inventory of each component. To repeat: each component must be controlled individually. The amount of gas flowing in must be balanced by gas flowing out. Changing pressure is a measure of the gas imbalance, therefore pressure control is the appropriate way of controlling the gas outlet. Similarly, oil level is an indication of oil imbalance and water level indicates water imbalance. None of the three streams may be controlled on flow, although a level / flow cascade is often used to smooth out flow variations to the downstream equipment. Pressure / flow cascade is unlikely to be used unless the volume of the vessel is large enough to serve as a gas surge drum. Level / flow cascade on the water is unlikely since the water probably drains to a collection system that itself serves as a surge drum to a number of separators.

Sometimes the ratio of water to oil is too great for a boot separator. In such cases a weir may be used to divide the vessel as shown in Figure 6-7.

Certain precautions must be taken to make sure that the level transmitter actually gives a true indication of the interface. There are basically two varieties of level indicating devices: The first measures the distance of an actual interface from some fixed point. Ultrasonic and radar devices belong to this group. These would be ideal for the purpose except that they are often not suited for installation in pressurized vessels. Furthermore they have difficulty "seeing" anything other than the very top interface. Even surface foam and condensation on the instrument "window" can confuse them.

The second, and more traditional, variety integrates some particular property, such as density or dielectric constant, over a span. Displacers, differential pressure transmitters, bubbler tubes, nuclear densitometers, capacitance probes, and even gauge glasses all belong to this variety. The key to successful measurement is that the level sensing device must sense only the two fluids bounding the interface. For a gauge glass this means that the lower tap must be in the lower of the two fluids and the upper tap must be in the fluid immediately above it. There may be NO intervening phases.

Figure 6-8 shows what happens when a gauge glass is connected to a vessel containing a vapour and two liquid phases. Assume that equal amounts of a liquid with Sg = 1.0, e.g. water, and a liquid with Sg = 0.5, perhaps propane, gradually flow into the vessel. Assume further that the span of the gauge glass is four feet, beginning one foot from the bottom of the vessel.

As the level of the propane rises, it flows into the glass. As both liquids rise further, water begins to enter the bottom of the glass. This is the state shown in vessel A. Up to this point, the glass shows a true indication of the level of propane in the vessel. Once water enters the glass, the propane is cut off. A constant plug, one foot thick, floats on top of the water. Its level no longer bears any obvious relationship to the actual level in the vessel. This is state shown in vessel B. The only relationship between the vessel and the glass is that the hydrostatic pressure is the same for both at the point where the glass taps into the vessel. A gauge glass is really nothing more than a manometer.

Once the level of the propane rises above the upper tap, it flows into the glass and the two interface levels adjust to the same elevation, as shown in vessel C. The gauge will continue to read correctly as long as its lower tap is in the water and the upper tap is in propane. If either fluid is withdrawn so that the upper tap is in the vapour space, the glass will once again read falsely.

This same analysis applies to any type of level indication based on density. Remember that a D P transmitter only gives a single reading, i.e. differential pressure. Therefore only a single quantity can be inferred. If the instrument is affected by only two fluids, it can yield the correct interface level between the two. If there are more than two distinct phases within the span of the two taps, it will give a reading based on the average densities of all the fluids within its span.

Capacitance or nuclear level transmitters will give similar results in multiphase situations, based on the average dielectric or nuclear absorption constants, respectively.

So... how can the process controls engineer be assured that the level readings are meaningful if even a gauge glass can't be trusted? Plan "A": Make the entire vessel out of glass. This isn't usually practical so we must fall back upon Plan "B": Every section of a gauge glass must have separate taps into the vessel so that each pair of taps has no "hidden" phase floating in between. Either that, or accept the fact that until the interface reaches its "normal" range, gauge glasses and transmitters will read falsely.

SLUG CATCHERS. Slug catchers are a special instance of three-phase separators frequently found in oilfield service. In addition to the usual separation functions, they are required to serve as surge tanks that can smooth out intermittent flow and also handle occasional very large surges in inlet flow. This is done by having two controllers connected to the oil-side transmitter. The oil overflow controller has its setpoint slightly below the top of the weir. In this manner, any surges can be accommodated by the large volume above the weir. This is in fact a non-linear, adaptive gain transmitter since transmitter gain = D output / D volume.

The inlet controller responds to the same level but has its setpoint just below the top of the vessel. It takes action only when the level rises to its set-point. This would happen if an unusually large slug of liquid arrived or if an upset in the downstream process caused the system to back up into the slug catcher. The facility would then be operating under "capacity control". Facilities lacking the capacity control feature are likely to experience a high level shutdown precisely when they are attempting to operate at maximum throughput. Not a very desirable occurrence.

It is common for level controllers to be tuned using both the proportional and integral modes. Since the inlet controller is normally functioning with the level well below its setpoint, reset windup will occur. This is a phenomena in which the controller attempts to raise the level to the setpoint by forcing an ever higher signal to the valve. This does not work, of course, since the valve is already wide open. If a sudden surge arrives that abruptly raises the level to the setpoint and beyond, the controller will be slow to close the valve since it has "wound up" in the opposite direction. Some form of anti-reset windup is required to prevent an unwarranted high level shutdown under these circumstances. It is probably a bad idea to use an equal percent valve in this application since it, also, is likely to respond slowly to a sudden demand.

It is possible to control the outlet and inlet valves with a single, split-range controller. This method accomplishes the required function of preventing high level shutdowns but has a serious disadvantage. If the setpoint of the combined inlet/outlet controller is set below the top of the weir, it will not take full advantage of the surge capacity of the vessel since the inlet will begin to close well before the top of the vessel is reached. If the setpoint is above the weir, it defeats its purpose by allowing mixed feed to flow directly to the oil outlet before it has time to separate. Thus a split-range controller will sacrifice either separation quality or surge capacity.

PRESSURIZATION SYSTEMS. A tank, vessel, or drum may require a pressurization system for any of a variety of reasons:

The surface of the liquid in a reflux drum consists of a liquid at equilibrium with its vapour. There may not be sufficient gravity head to provide the net-positive-suction-head required to operate the reflux pump without cavitation. Raising the vessel high in the air above the pump is one way of providing this. Unfortunately the condenser providing the liquid, drains by gravity so it must be raised even higher. The entire arrangement can become extremely expensive. An-other method is to use a canned pump which is sunk deep into the ground. This can also get pricey. A blanket gas pad may be a relatively inexpensive way of providing the necessary pressure.

The liquid in a storage tank is subject to oxidation, e.g. the surge tank of a glycol-based heat exchange system. A blanket of fuel gas will prevent the tank from breathing air as it cycles from empty to full and back again.

The liquid in a storage tank forms an explosive mixture with air. (A rather extreme form of oxidation!) A continuous gas purge may be required to prevent this.

The storage tank vents to a flare or other vapour collection system. A gas supply must be provided to make up any volume withdrawn when the withdrawal rate exceeds the fill rate. In other words, the pressurization system serves as a vacuum breaker.

A simple way of providing pressurization is to have a regulator connected to a source of pad gas and a second, back pressure regulator, connected to the vent. Care must be taken to set the back pressure regulator setpoint slightly higher than that of the inlet regulator. If there is no gap between the two settings, the pad gas will blow straight through to the vent. Remember that setting them to the "same" pressure is meaningless.

Often it is necessary to install a complete control loop including a pressure transmitter, a controller, and two valves. This has the advantage of allowing the panel operator to monitor and adjust a single setpoint. It also allows over- or under-pressure alarms to be easily provided. Figure 6-10 shows how the complete pressure control loop is arranged. For the most part it is pretty simple but there are two things to watch for: Firstly, there must be a gap between the action of the two valves. That is the reason for the split range values not meeting at 50%. Secondly, the failure mode of the valves must be taken into account. Since the two valves have the opposite effect, they must have opposite failure modes if they are to be operated by the same control signal. A DCS allows the output of the controller to drive two separate output modules, each characterized in its own way. This means that it is possible for the first 45% of the controller output to produce a 100 to 0% signal for the fill valve, and the last 45% of the output to produce a 0 to 100% signal for the vent valve. In this way both failure modes are accommodated and overlap of valve openings is impossible. The gap in the middle does not cause a problem for the controller. Integral windup will move the output quickly through the gap whenever there is a deviation from the setpoint. The reader should note that the split range control described in this paragraph is not at all the same as that described in the section on Slug Catchers.

It is possible to achieve the same effect by using a specialized, three-way valve that provides a gap in the middle. Most three-way valves are designed to have full overlap as they are intended for use in diverging/converging service. (If anyone knows of a centre-gap, non-overlapping valve, let me know.)

A number of vendors sell specialized gas blanketing systems capable of self-contained action. They consist of regulators with the very large diaphragms required to drive valves with pressures as low as 0.5" WC (125 Pa, 0.3 oz/in 2 ). Such systems are especially useful now that ever more stringent regulations concerning the emission of volatile organic compounds (VOCs) are being enforced. Figure 6-11 shows one typical arrangement.

Many factors enter into the correct specification of the setpoints and sizes for the various regulator and relief valves. These include:

The maximum allowable pressure of the tank.

The maximum allowable vacuum of the tank.

The vapour pressure of the stored liquid.

Inbreathing rate dependent on pump-out rate.

Outbreathing rate dependent on pump-in rate.

Vapour thermal expansion and contraction rate.

Tank surface area and insulation.

Table 6-1 provides setpoints applied in a specific case. It must be remembered that actual values differ widely. API 2000, Venting Atmospheric and Low-Pressure Storage Tanks 5 and tank vendors provide much information, however it may be advisable to consult a specialist in the field.

oz/in2"WCkPa

Maximum Allowable Pressure4.06.91.7

Manway Setting3.56.11.5

Relief Valve Pressure Setting3.05.21.3

Vent Regulator Setting2.03.50.9

Fill Regulator Setting0.50.90.2

Relief Valve Vacuum Setting-.05-.09-.02

Maximum Allowable Vacuum-1.0-1.7-0.4

Table 6-1. Typical Tank Blanket Pressure Settings

A brief sermon on tagging: According to ISA-S5.1, Instrumentation Symbols and Identification 6, all forms of relief valves including pressure, vacuum, spring- or weight-loaded, with or without a pilot are tagged "PSV". Common abbreviations such as "PVSV", "PVRV" or "PRV" have absolutely no official status and therefore are not acceptable as tags on P&IDs.

LEVEL MEASUREMENT. Level measurement is deceptively easy, yet it seems that more time is spent specifying level instruments than any other. The reason is that the correct installation of level instrument is an interdisciplinary effort involving Process, who set the basic functional requirements; Mechanical, who have various constraints such distance of taps from seams; Piping, who have accessibility and orientation requirements; and Instrumentation, who must select from a finite catalogue of available instruments.

Actually this task has become considerably easier in recent years due to the increased use of D P transmitters and other instruments such as ultrasonic and radar which do not have a predetermined span. There is no longer any significant penalty in either cost or accuracy if the instrument is specified to cover a broad span. For horizontal vessels the top connection should be vertical at the top of the vessel. The bottom connection should be horizontal a few inches from the bottom. This is necessary to prevent the accumulation of sediment. These connections no longer need to be in the same vertical plane nor do they require the same orientation.

Vertical vessels may still require a bit more attention. While a top-to-bottom span would be ideal, there may be trays, packing, or other internals that would cause a differential pressure in response to flow. It is also necessary for the level connections to remain clear of welding seams. This requirement may cause problems if alarms or other setpoints need to be near the bottom or top of the vessel.

The design process begins with the basic information on a P&ID in a form similar to that shown in Figure 6-12. A brief outline of the vessel including the bridle, if any, holding a gauge glass and a transmitter are shown. The desired values for the level alarms and the setpoint of the controller are also shown.

The Control Systems engineer must first decide if a D P transmitter is the appropriate choice. Let us assume it is. He/she must then try to find an appropriate span for the transmitter. A good rule of thumb is that alarms should not be set any lower than 10% or higher than 90% of the transmitter span. (Shutdown trip settings should not be closer than 5% from either end of span.) Since the two alarms are 42" - 6" = 36" apart, the span should be 36" x 1.25 = 45" thus allowing the alarms to be at 10% and 90% of span. This seems fine, but there is a problem. The first thing to determine is whether the vessel measurements are from the tangent line, from the seams or from some other reference point. In this particular case the vessel title block indicates that measurements are tan-to-tan. Since seams are generally 2" inside the tan lines, the lower tap of the transmitter is " above the seam. That is not acceptable. Mechanical considerations often dictate that nozzles may not be welded within 6" of a seam. This means that the lowest transmitter tap cannot be lower than 8" above the bottom tan line. The highest tap cannot be higher then 8" below the top tan line. This implies that the maximum transmitter span on a 48" T/T (tan-to-tan) vessel is 32". Alarms at 10% and 90% must be placed at 11.2" and 36.8". At this point, the Instrument Engineer becomes a broker between Process and Mechanical to help them find a compromise. Alarms at 11" and 37" are agreed upon. Don't forget to transfer this new information back to the P&ID!

It is a great convenience to the maintenance staff if the span of the transmitter is exactly equal to the span of the gauge glass. This is not always possible with displacers since both the gauge glasses and the transmitters come in fixed spans. However, it can easily be done for D P transmitters. The centre line of the sensing taps must be located at the tops and bottoms of the visible glass. Unfortunately, the associated valves bring the centre line of the gauge glass tap 4" below the bottom of the glass. Fortunately, in extreme cases, it is possible to place the lower gauge glass tap below the lower bridle tap. There can be no meaningful readings below the lower bridle tap of course, since the bottom of the glass can never drain back into the vessel. As long as the glass itself does not go below the lower tap, it's OK. Since gauge glasses come in fixed lengths, it may not be possible for the upper tap of the transmitter to match the top of the visible glass. Remember that transmitter calibration will be more difficult if the upper tap does not fall within the range of the visible glass.

Occasionally it is necessary to connect either the top or the bottom taps to interconnecting pipe instead of to the vessel itself. If the taps are attached to inlet or outlet lines, the level signal will be affected by flow rate. This effect can be seen in coffee percolators: The level in the gauge glass bobs up and down as coffee is drawn into a cup.

SEALS. Diaphragm seals have become a very popular accessory to D P-based level transmitters. A very thin metal diaphragm isolates the transmitter from the process. The space between the diaphragm and the sensor itself is filled with a fluid such as silicone. The pressure changes are communicated through the diaphragm to the transmitter via an armoured capillary tube. The volume change between minimum and maximum pressure is extremely small in a modern transmitter; the amount of flex in the diaphragm is correspondingly small. The net effect of this is that the introduction of a diaphragm into a measurement system introduces an error of only several centimetres or less. This is seldom significant in level applications. A second effect is that only an extremely small amount of liquid movement actually occurs in the capillary tubes. This, together with modern low temperature fill fluids, means that the instrument response does not slow down too much on cold days. Figure 6-13 is an example from one vendor's catalogue.

Seals should be considered whenever one or more of the following conditions apply:

Dirty service - Whenever the process fluid is liable to plug the impulse lines, a diaphragm seal may be installed. It should be isolated from the vessel by a full-ported valve, NPS 2 or 3. Note that two NPS taps are provided on the diaphragm housing for calibration and flushing connections. Seals are especially useful in sanitary service where all hardware in contact with process fluid must frequently be thoroughly washed. It is a good idea to use seals and capillaries of equal length on both the upper and the lower leg in order to maintain a balanced response to errors.

Corrosive service -- Diaphragm seals made of corrosion resistant materials originated in corrosive service where they were referred to as 'chemical seals'. While the use of full-ported connections is not required in corrosive service, it is a good practice to maintain even if it might look strange to have 'such a big valve' for 'only' an instrument.

Freeze protection -- Diaphragm seals may eliminate the problem of freezing impulse lines. However, in extremely cold weather it may still be necessary to heat trace the capillaries to prevent measurement response from being excessively slow. Self-limiting electrical heat trace is the only way to go! Any heat trace system involving a thermostat will introduce spikes into the measurement system as the heat is switched on and off.

Uncertain phase -- This is the most frequent of all seal justifications. A warm vapour in equilibrium with its liquid will undergo condensation in the upper impulse line. Cold equilibrium liquid may experience boiling in the lower impulse line. Thus the measurement will slowly drift as the tube fills. Depending upon service and ambient temperatures, condensation and boiling may even alternate throughout the day. If this situation exists, the measurement becomes worthless. The traditional solution is to fill the upper line with a non-volatile, process compatible fluid. Depending on process and ambient conditions this might be water, glycol, oil or something else. The use of fill fluids introduces maintenance problems because any attempt to 'null' the transmitter by opening the equalization valve will drain the upper fluid into the process. It can only be replaced by climbing to the top of the vessel and filling the tube again. Bubbles are also a source of error. Seals provide a captive fill fluid that cannot be lost, does not form bubbles and cannot contaminate the process. (Did I say foolproof?)

UNDERGROUND TANKS. A special requirement concerns underground (UG) tanks. Modern steel UG tanks have a double wall construction. Requirements are outlined in CAN/ULC-S603, Standard for Steel Underground Tanks for Flammable and Combustible Liquids 7. The two walls of the tank are approximately an inch (2 cm) apart. A vacuum of 51 kPa (7.5 psi) is drawn on the interstitial space so that the two surfaces are, in many places, actually in contact with each other. A vacuum gauge is connected to the interstice. It must read at least 42 kPa (6.1 psi) of vacuum before the tank may installed. If the reading is ever less than 34 kPa (4.9 psi) the tank should be removed from service and steps taken to determine the cause of the leak. These values are summarized in Table 6-2, below. If a facility has many UG tanks, it may be desirable to connect the tanks to the central control system by means of vacuum transmitters. Low vacuum alarms can then be configured to alert the operators of any cases of leakage.

Interstitial VacuumpsikPa

Required at manufacturing7.551

Minimum acceptable for delivery6.142

Minimum allowable in service4.934

Table 6-2 Interstitial Vacuum Requirements for Underground TanksVOLUME MEASUREMENT. Most vessel and tank content measurements are made in the form of level. When true volume is required for such purposes as custody transfer, the tank volumes are calculated taking into account all details of their geometry as well as dimensional changes resulting from the pressure exerted by the density of the liquid. The results of these calculations and calibrations are tabulated by the manufacturer in a form known as "strapping tables".

True volume measurement is seldom relevant for control purposes since setpoints for controllers and alarms are usually related to specific geometric features of the vessels. The level must often be kept below the vapour outlet or a weir. A frequent requirement is that a specific head be maintained to prevent pump cavitation. Sometimes the requirement is simply to maintain the level near midpoint in order to provide surge capacity. None of these applications benefit from true volume compensation. Figure 6-14, Volume vs. Height for Cylinders and Spheres, provides the correct mathematical relationship between level and volume for these two vessel styles if true volume measurement is actually required. It can be seen that between 10% and 90% little is gained by applying the rather complex calculations required for volume control.

Volume = (R2L/2)(2q - sin 2q ) Volume = -(p /3)h2 (3R - h)

Where h = height of liquid in vessel R = radius of vessel L = length of cylinder q is radians and cos q = (R-h)/R

Note: The volume contained by elliptical vessel heads is that of a sphere of equal radius.

SAFETY. Vessels and tanks are probably the most hazardous pieces of equipment in any plant. Duguid's database shows that 22% of all safety incidents are related to storage and blending. This may seem a little surprising until one considers that they store energy as well as material. For example:

A vessel holding a compressed gas can cause a tremendous explosion if it ruptures. That is why "hydrotesting" with air or nitrogen is far more dangerous than with water.

Storage tanks can hold a considerable amount of gravitational energy. The most notorious example of this energy being released is the infamous "Boston Molasses Disaster" which occurred January 15, 1919. A tank located at the top of a hill ruptured and released two million gallons of molasses down a narrow street. Twenty-one people were killed and 150 were injured.

The contents of the tank or vessel can be flammable. While a line rupture external to the tank may be the cause of a fire, it is the reservoir of flammable fluid inside a tank that turns a minor fire into a major one. API RP 750, Management of Process Hazards, specifically addresses this point, however, it does not offer much in the way of solutions.

The contents of a tank can be lethal. The February 1984 release of methyl isocyanate in Bhopal, India was the worst non-nuclear industrial accident in human history. Over 2000 people were killed by the toxic vapour.

The single, most comprehensive guide to the design of vessels is the ASME (American Society of Mechanical Engineers) Boiler and Pressure Vessel Code8. This rather large document deals with all aspects of vessel design, construction and operation. Section VIII, Parts UG-125 to 136, in particular, deal with the requirements for pressure relief.

The pressure relieving requirements for non-pressure vessels, i.e. tanks, are covered in detail in API Standard 2000, Venting Atmospheric and Low-Pressure Tanks5.

Most safety related design practices applying to vessels and tanks are beyond the scope of the instrumentation and controls engineer; relief valves are an exception. Their correct sizing and selection is too broad a subject to be covered in this article especially since there is already a lot of material in print concerning them. Items 9 through 14 of the references below deal extensively with this subject. The earlier section, Pressurization Systems, gives a typical example of pressure protection for an atmospheric tank.

INTRODUCTION. The positive displacement pump is in some ways an even simpler device to control than the centrifugal pump discussed previously1. It has the same function, namely to provide the pressure necessary to move a liquid at the desired rate from point A to point B of the process. Figure 2-1 shows a 'generic' process with a positive displacement pump (in this case a gear pump) connected to deliver liquid from A to B.

There is a great variety of positive displacement pumps. They are divided into two broad categories: Rotary and reciprocating. From the controls point of view, however, they are all similar. Their characteristic curve is so simple that it is rarely drawn. It is essentially a straight vertical line, as shown in Figure 2-2. (For some reason PD pump curves are usually shown with the pressure and flow axis exchanged. I will not follow that convention in this article.) All are constant flow machines whose pressure rises to whatever value is necessary to put out the flow appropriate to the pump speed. If the discharge is blocked, the pressure will rise until something yields -- preferably a relief valve. Close examination of the curve shows a slight counter clockwise rotation. This is due to internal leakage.

For positive displacement pumps the major cause of leakage is the small amount of reverse flow that occurs before a check valve closes and possibly past the check valve after it is closed. Leakage past the piston is negligible. Diaphragm operated PD pumps have no cylinder to leak past. Rotating PD pumps, such as gear pumps or progressing cavity pumps have internal clearances which permit a small reverse flow, called "slip" or "blowby". There is another reason why the curve may rotate to slightly lower flows at higher discharge pressures: The driver may slow down as the load increases. None of these have a significant affect in curving the slope of the characteristic enough that this slope can be used for control. For most practical purposes the slope is vertical. The system curve of the process is also shown on Figure 2-2. Its intersection with the pump characteristic defines the operating point.

As always, the process controls engineer has the responsibility of matching the capacity of a specific piece of equipment to the demands of the process at every instant in time. Rarely does the actual system curve fall exactly on the one used for design and selection. As with any two port device, there are three locations in which a control valve can be placed: On the discharge, on the suction, and as a recycle valve.

DISCHARGE THROTTLING. Discharge throttling does not work! Looking at the process from the point of view of the pump, discharge throttling rotates the system curve counter clockwise so that the modified system curve intersects the pump curve higher up. The additional pressure is dropped through the valve so that the pressure and flow to the process is (almost) exactly the same as before. The "almost" is due the small increase in internal leakage that results in an equally small reduction in flow. An increased wear rate and a shortening of the life of the machine are the only results of this approach. If the pump is seen from the point of view of the process so that the valve is considered part of the pump, the same result is obtained. To obtain a modified pump characteristic curve, the pump curve must be rotated clockwise around the intersection with the pressure axis. The problem is that this hypothetical intersection is far off the top of the operating range. It is the point where the pressure is so high that 100% internal leakage occurs. The machine would self-destruct from excess pressure if one were stubborn enough to attempt to find this point. The rotation of the curve can still be performed on paper and it amounts to a slight shift to the left. Shown in Figure 2-3, it is virtually identical to the unmodified curve. To cut a long story short, you can't control a PD pump with discharge throttling.

SUCTION THROTTLING. Suction throttling has the same effect on the characteristic curve as discharge throttling and doesn't work either. PD pumps have a Net Positive Suction Head Required (NPSHR) just as centrifugal pumps do. In fact their requirements are even more stringent. Therefore restrictions and pressure drops in the suction lines must be similarly avoided.

RECYCLE CONTROL. This leaves recycle control as the only means of using a valve to control a PD pump. The valve is installed in a line teeing off from the discharge and leading back to the source of the liquid, possibly a surge tank. It must be fail open , of course. Figure 2-5 shows its effects on the characteristic curves. Viewing the process from the point of view of the pump, its effect is to rotate the system curve clockwise around its intersection with the pressure axis. Note that the little "tail" at the bottom left of the modified system curve is due to the flow through the recycle valve before the discharge check valve has opened. The flow through the pump is essentially as before but the pressure to the process has been reduced. Process flow will, of course, also be reduced by the amount flowing through the recycle line.

Viewing the pump from the process gives a different perspective on the same phenomenon. This time it is the pump curve that is rotated counter clockwise around its intersection with the flow axis. This modified pump curve gives the effect of greatly increased internal leakage. From the point of view of the process, this is exactly what is happening. Note that I have not used the same operating points in Figure 2-3 as I did in Figure 2-5. It is simply impossible to show any significant reduction in flow on a curve representing the effects of discharge throttling.

Recycle control is an efficient method of control for PD pumps. Since the flow rate is essentially constant, the power requirement is roughly proportional to discharge pressure. Since the effect of recycle is to drop the discharge pressure, it results in significant reductions in power requirement. Nevertheless there is still wasted power in proportion to discharge pressure times recycle flow.

Recycle valves experience rather severe service if the pressure drop is high. Cavitation will destroy them if they are not appropriately selected. Two approaches exist to deal with this problem: The first solution is to drop the pressure in many small stages through the use of many twists and turns in the valve trim. The second is to tolerate the resulting cavitation by shooting the liquid as a jet through a small hole in the middle of a disk. The jet then blasts directly into the discharge piping. The line diameter is often increased immediately downstream of the valve and the wall thickness is also increased. In this way the jet cavitates down the middle of the pipe. It makes a terrific racket.

In either case it may be necessary to put a fixed restriction downstream of the valve. It should be sized so that the ratio of the high to intermediate pressure is the same as the ratio of intermediate to low pressure. Keep in mind that the restriction will reduce the rangeability of the valve by making it act like a quick opening valve. This is because the restriction becomes the dominant factor in the line once the valve is about half way open. From that point on, the valve has little control.

Recycle lines for PD pumps should be run back to the suction vessel. This allows any entrained bubbles to escape. If they do not, they can build up to the point where pump capacity is impaired. It may even vapour lock.

SPEED CONTROL. Speed control is an obvious method of controlling the flow rate of PD pumps since flow is essentially proportional to speed. Pressure can also be controlled by sliding up and down the system curve. Any point on the system curve can, in theory, be reached. Most drivers, however, have low speed limits which limit the turndown of the system.

Variable speed electric motors are somewhat modified versions of normal motors. They require special provision for cooling and lubrication at low speed. In addition, they require specialized electronic power supplies called "invertors". These units provide power of the appropriate frequency and voltage. They are, unfortunately, still quite expensive and do not have the reliability of control valves. There is another reason why large variable speed electric drives are seldom used with reciprocating pumps. The large inertia of the system means that speed changes cannot be made quickly. If it is possible for a valve in the process side to close suddenly, a variable speed electric cannot reduce speed fast enough to prevent a severe pressure rise. A recycle valve will be required to protect the pump, as detailed below in the section on machine protection. A more simple type of electronic control is frequently used for small chemical injection pumps.

OTHER MEANS OF CONTROL. The great variety of types of PD pumps results in a variety of specialized means of flow control. A pneumatic actuator may be used to vary the geometry of the crank arrangement of a reciprocating pump so that each cycle displaces a greater or lesser amount of cylinder volume. Direct acting diaphragm pumps driven by compressed air or some other gas can be controlled by regulating the gas supply. There is also a technique known as "lost motion" whereby the crank arrangement first compresses a spring or volume pocket before it begins to work on the piston or diaphragm. These specialized methods are usually integral parts of the equipment and the controls engineer simply connects a pneumatic or milliamp signal to the appropriate input port. None of these methods changes the essentially constant flow nature of the pump curve. (The flow is still "constant" but at a different value.)

The efficiency of hydraulic or eddy current couplings is about the same as that of recycle control. This is because the torque on both sides of the coupling is proportional to D P. The power lost in the coupling will be proportional to torque times the reduction in speed. In other words, all unused power is being dumped. If the pressure does drop with a reduction in net discharge flow, then there will be a power savings. A valve is a cheaper way of accomplishing the same thing.

"Stroke Counting" is a method used when fixed amounts of liquid must be injected at specific intervals such as in batch processes. An electronic device is used to count the number of revolutions of a PD pump. After a sufficient number has been counted, the pump is shut off. When this method is used for pH control, the correct number of strokes can be calculated from a titration curve.

MEASUREMENT. The most common application for PD pumps is in high-pressure service. The flow rates vary from extremely small to moderately large. Pressure control is very common. Since the control valve tees off the discharge header, it is not significant where the sensing transmitter is placed. Keep in mind that the discharge will be pulsating. The pulsations may be relatively small for a rotary pump or they may be extremely large for a simplex (single cylinder) reciprocating pump. The degree of pulsation also depends on the effectiveness of the hydraulic pulsation dampeners that are often supplied wit