Predicting oncoming failure is a snap with sealless pumps

You can predict incipient failures in this class of fluid handling equipment.

By Julien Le Bleu, Jr., Arco Chemical Corp., and James Lobach, Chempump Division of Crane Pumps and Systems

Centrifugal sealess pumps, both canned motor and magnetic drive, should be monitored to determine mechanical condition. In sealess pumps, the pumped fluid is the cooling and lubricating medium for the pump bearings. If intermittent monitoring is used, then the chance of detecting pump damage caused by process changes is very small.

Vibration monitoring techniques as applied to sealed pumps have been unreliable for detecting problems. The effectiveness of conventional monitoring techniques is limited by the time interval between measurements, the relative isolation of the inner pump rotor from the outer measuring location, and by the pumped fluid. Other factors such as fluid affects and process noises can make interpretation difficult.

This paper presents the synergistic combination of two relatively new methods of sealess pump monitoring. These methods considerably enhance the range and magnitude of mechanical problems that can be identified on this type of pump.

One of the goals of predictive maintenance is the reduction of maintenance costs by use of condition monitoring. Identification of off-design operating conditions or mechanical damage at an early stage enables one to correct the conditions before damage occurs, or to optimally schedule repairs. Continuous monitoring of sealess pumps reduces equipment maintenance costs and facilitates root cause analysis of mechanical failures and operational problems. This cost reduction is accomplished by the use of the monitoring system to immediately identify conditions that can lead to failure.

This article presents the results of experiments that attempted to determine reliable predictive condition measurement tools for sealess pumps. These experiments measured mechanical and operating parameters as the pump operating points varied. The data from individual measurements was examined for correlation with other measurements and also to mechanical condition and damage. Monitoring of mechanical condition can be used to schedule maintenance intervals, but the best use of monitoring is to allow the detection of the conditions that lead to equipment damage. The monitoring system should provide information early enough to allow the potentially damaging conditions to be changed. Thus, the possible cause for the response and potential for failure are eliminated.

The parameters recorded during lab tests were: power in watts, overall high frequency tracking (overall high frequency tracking), rotor position, suction pressure, and capacity in gpm. The testing was done on a canned motor pump with a commercially available rotor position monitor supplied by a major pump manufacturer. A "truth" table was generated for various conditions and the table was tested using a closed pumping loop.

Sealess centrifugal process pumps fall into two major categories. One is the magnetically driven design and the other is the canned motor design. The two types have certain similarities. For example, they both use the process fluid for cooling the drive mechanism and lubricating of the internal bearings. The designs differ in how rotation is induced in the impeller and rotor.

The synchronous magnetic drive utilizes two sets of magnets, one on either side of a containment shell made of non-magnetic material (usually stainless steel or Hastaloy). An electric motor moves the outer magnets and oil lubricated bearings supported them. The inner magnets follow the outer magnets by the attraction through the containment shell. A second non-magnetic covering protects the internal magnets from the process fluid. The internal magnets are attached to a shaft that drives the pump impeller. The shaft is supported on bearings that are lubricated by the process fluid.

The canned motor pump uses a single rotating element that is essentially the rotor of an electric motor but with an impeller mounted on the shaft. The motor and pump casings are sealed eliminating the shaft penetrations common to conventional pumps with mechanical seals. A non-magnetic containment shell protects the stator and motor rotor from the process fluid. The rotor is supported generally using fluid lubricated film bearings. A portion of the pumped fluid is circulated to the motor to provide cooling and bearing lubrication. Canned motor mechanical construction is less complex than magnetic drive designs, but catastrophic failures of the stator assemblies are expensive to repair and difficult to decontaminate.

We sought a method of getting reliable and timely information about the condition of sealess pumps. The ideal condition monitoring method should:

  • be non-intrusive, the monitoring devices should not penetrate the liquid containing parts of the pump;
  • be able to indicate process changes that influence pump operation as well as measure mechanical wear;
  • be reliable and proven technology;
  • be readily available;
  • the results should be easy to interpret;
  • contain enough parameters to indicate a problem so "false trips" are not an occurrence and detecting failures are assured;
  • be easily retrofitted to existing sealess pumps;
  • not require the monitoring sensors to be sacrificial or consumed during normal wear of the equipment;
  • be upgradable so that new pumps do not need to be purchased just to have the improved monitoring technology; and
  • be able to withstand a chemical plant environment.

Several clear and simple patterns indicate problems with sealess pumps. However, the sensitivity of the detection system generates many indications. Not all indications can be resolved and not all indications imply an immediate problem. Improved diagnostics based on multiple parameters should be able to determine a healthy or unhealthy pump and pumping environment. This thinking lead to the collaboration between a pump manufacturer and an end user to instrument a pump and manipulate many normally occurring operating parameters to get a known response. The collaboration resulted in the generation of a truth table. This truth table lets pump users determine the cause of conditions leading to the response measured by the monitoring systems.

Sealess pump users have been requesting rotor position as a measurement for a long time. Measuring the rotor position allows operators to know where the rotor is in relation to the stationary pump components. As long as the bearings are not badly worn, serious damage caused by rotor to stator contact cannot occur. Reliable direct measurement of the rotor axial and radial position is now available in canned motor pumps. Historically, rotor position monitoring, when available, was applied to silicon carbide bearings. These do not wear so the benefits of rotor position measurement as a predictive tool were minimal. When rotor position monitoring is applied to bearings made of softer materials that wear--such as carbon /graphite--the technique becomes predictive because it allows the user to track wear on the bearings and schedule maintenance before serious damage occurs.Viewing the position output continuously reveals the rate of change of the wear in the rotor bearing system. It allows safe operation of the pump for a time after detecting a problem.

Rotor position monitoring as a "go--no go" gauge is a maintenance scheduling device only because it gives no rate of change or process information. Axial rotor position, if monitored continuously, gives more process information than does radial position. Continuous monitoring of pump condition information allows one to correct the root cause of process related bearing wear before the bearings are worn to the point that they require maintenance.

One manufacturer developed a patented combination carbon/silicon carbide bearing to support the rotor. If process conditions cause the silicon carbide to fail, then a "catcher" bearing made of carbon supports the rotor and allows continued operation for a time, but with no metal-to-metal contact. The silicon carbide bearing is behind the carbon bearing so the pieces of the failed bearing are captured and retained in the area of the shaft. Because they are not allowed to move freely with the pumped liquid there is minimum damage to the rotating and stationary parts of the pump. The combination of rotor position monitoring with a dual bearing is a reasonable stopgap measure that allows the use of silicon carbide bearings in a sealess pump until bearing materials with slow and predictable failure modes can be found.

An explanation of overall high frequency tracking can be found in a paper by IRD titled Technical Report #11 "Using Spike Energy for fault analysis and machine condition monitoring." The document explains overall high frequency tracking and some of its uses but not as it applies to sealess pumps.

The previous work that led to these experiments are documented in "Monitoring Sealess Pumps For Metal to Metal Contact", Proceedings from the 11th Annual Texas A&M Pump Users Symposium, and "A New Method of Monitoring Sealess Pumps", Proceedings from the 19th Annual Meeting of the Vibration Institute.

The narrow trace produced by overall high frequency tracking at a low value is desirable because it indicates an acceptable operating range for the pump. The only exception is loss of liquid to the pump. This observation should be validated during the course of testing. Rotor position was previously monitored as a "go/no-go" indication of bearing condition. Provisions to monitor radial and axial position continuously were used during these tests.

Test setup
The pump was installed in a test loop consisting of instrumentation, a supply tank, and associated piping. The pump was subjected to conditions that attempted to simulate what can be encountered during plant operation. The pressure on the supply tank could be varied giving the ability to induce or eliminate pump cavitation to measure its response with the sensors. The canned motor pump supplied for the test was a 3,450 RPM, 3 X 1-1/2 X 6.

The pump was equipped with the manufacturer's rotor position monitoring device to monitor both axial and radial rotor position. Overall high frequency tracking was measured using two accelerometers connected to a dual channel monitor.

The reasoning was that measurements of rotor position, overall high frequency tracking, and power would provide sufficient information to determine pump mechanical condition. The expectation was that this combination would also provide advance warning of process conditions that would adversely affect the pump's health. After all, two or more indications of a problem should minimize or eliminate the possibility of false trips.

Procedures
Measurements were taken for the following pump operating conditions:

  • pump capacity range from shutoff to 30 percent greater than BEP with data taken at 20 gpm intervals in the range,
  • best efficiency point as part of the previous test and used as a base line,
  • a sudden large increase in pump flow,
  • air leakage into the suction of pump (introduced by injection),
  • dry pump operation (as part of the previous test), and
  • reduced NPSHa.

The following information was recorded for each of the operating conditions listed above. The data in parentheses are the plot scale factors for the data presented in the paper. The letters in brackets represents the letters on the graphs for that data set. The dimensionless unit of spike energy, gSE, is the acceleration in g's of spike energy.

  • flow [A] (1 volt = 0 gpm, 5 volts = 300 gpm)
  • overall high frequency tracking on the casing [B] (0 to 16 gSE: 1 volt = 0 gSE, 5 volts = 16 gSE)
  • overall high frequency tracking on the rear bearing housing [C] (0 to 5 gSE: 1 volt = 0 gSE, 5 volts = 5 gSE)
  • motor input power in watts [D](1 volt = 0 kW, 5 volts = 15.5 kW)
  • axial rotor position [E] (1 volt = 0.000 inch, 5 volts = 0.100 inch)
  • radial rotor position [F] (1 volt = 0.000 inch, 5 volts = 0.013 inch)
  • suction pressure [G] (1 volt = 0 psia, 5 volts = 30 psia)

An increase in voltage for the rotor axial position data represents a movement of the rotor towards the suction flange of the pump. There were two overall high frequency tracking monitors and accelerometer sensors installed on the test pump. Both of the sensors indicated identical patterns, but at different magnitudes. One sensor was mounted on the head of a bolt that attached the pump casing to the pump housing. It was oriented in the axial direction. This mounting method was chosen so that the sensor could be easily mounted on more than one pump if desired. The plots are marked with a B to indicate these readings. A second sensor was mounted on the outboard bearing housing in the axial direction. The plots are marked with a C for the output from this sensor.

Test results
The data indicates a direct correlation between a change in the axial rotor position and a change in the magnitude of the overall high frequency tracking readings. When the operating conditions such as cavitation or pump flow changed, similar indications were evident in the axial rotor position and the overall high frequency tracking.

For example, if cavitation moved the rotor toward the suction flange of the pump, it also increased the pump casing noise that was detected and indicated by the overall high frequency tracking sensor. Rotor axial position and overall high frequency tracking levels appeared to track each other closely.

The rotor on this pump normally operates in a position slightly closer to the back plate of the pump and away from the suction. This is a function of the hydraulic balance and the normal running clearances around the pump impeller. The indicated radial position of the rotor did not change unless there was bearing wear, catastrophic bearing failure, or dry running. A general correlation between overall high frequency tracking levels and repair costs suggests that operation at high overall high frequency tracking levels results in higher repair costs.

Changes in flow
The changes in data that result from variation in flow from 50 to 190 gpm are represented in Figure 6. Note that at flows exceeding the best efficiency point, the rotor begins to move toward the suction flange and overall high frequency tracking begins to move upscale and increase in width. The trace becomes wider in both rotor position and overall high frequency tracking as the fluid flow and rotor position responds to the changes in pumping conditions. Wider traces of both overall high frequency tracking and axial position are indicative of operating conditions to avoid in pumps and especially sealess pumps. Radial position was constant because there was no bearing wear.

Large flow decrease
Overall high frequency tracking, indicated by (B) and (C) on the chart, is at a high value and a wide trace at 200 gpm indicating cavitation and pump stress. The axial rotor position also has a wide trace indicating the rotor was "hunting" to find its hydraulic balance. When the flow is decreased from 200 to 150 gpm, the rotor moves to a normal and more balanced axial location and overall high frequency tracking decreases from 10 to approximately 4 gSE. When the flow is reduced to shut off, the rotor moves toward the suction of the pump and overall high frequency tracking is somewhat elevated over that of operation at the point of best efficiency. If left to operate at this condition for some time, overall high frequency tracking will increase as the liquid in the pump begins to flash. Testing that allowed the liquid to boil in the pump while measuring temperature and overall high frequency tracking resulted in high temperatures and high overall high frequency tracking levels. When the flow rate is returned to near best efficiency point overall high frequency tracking returns to a previously observed level for normal operation. Because it returned to previous levels, there was no permanent damage to the pump. Years of plant experience have shown this to be the case. When serious damage occurs, such as metal to metal contact or excessive operating clearances on the wear rings, the overall high frequency tracking levels do not return to baseline values. Experimentation and field experience have shown that overall high frequency tracking levels are repeatable at different capacities as long as the pump is undamaged.

Large flow increase
At very low flows the overall high frequency tracking value is high and the rotor axial position is near center. At the best efficiency point for this pump, 150 gpm, overall high frequency tracking is at a minimum level and rotor axial position is slightly toward the motor, relative to its position at BEP.

When flow increases to levels significantly higher than BEP, an increased width of signal in both the overall high frequency tracking and axial rotor position data indicate the onset of cavitation. Rotor axial position oscillates at the high flow level; the wide data trace (E) indicates the rotor is "hunting" for an equilibrium position. The rapidly changing pressure balance on the impeller during cavitation causes this oscillation. The overall high frequency tracking signal also becomes less stable and "nosier" when the pump is cavitating.

Suction side restriction
The suction valve was closed at a constant flow to identify the response in terms of the measured parameters. The width of the trace representing axial rotor position and overall high frequency tracking begin to increase. Both of these indicate instability--one in the rotor position and the other in pumped fluid. This test could represent plugging of a suction strainer or a valve that was not opened fully. It can also represent a fluid that has become too hot and is flashing in the pump suction.

This indicates that sealess pump failure attributable to an inadequate suction condition or other cavitation-inducing operation can be minimized through the connection of the monitors to appropriate operator alarms. The monitoring system also allows the user to quantify the severity of the operating condition as well as assess the pump condition after eliminating the cavitation. Indications of severe bearing wear or permanent damage will be indicated.

Air injection and dry run
Overall high frequency tracking immediately begins to decrease. The axial position does not change immediately because the loss of fluid is not immediate. The radial position of the rotor changes because the loss of fluid in the radial bearings and around the rotor reduces the stiffness of the radial support. The loss of coupling fluid and transmissibility between the pump casing and the rotor explain the low overall high frequency tracking levels.

If dry running operation is continued to the point of bearing damage and possible metal-to-metal contact, then the overall high frequency tracking level will increase. When liquid is re-introduced to the pump, cavitation exists for a short time, and overall high frequency tracking levels increase. The wattmeter shows no load on the pump and the flow has fallen to zero.

Continued dry operation will destroy the pump. This data shows that a dry run, as with a tank pump out, does not result in an instantaneous catastrophic failure. The effects of dry running and severe cavitation are cumulative.

Cavitation induced by low suction pressure
While NPSH testing at the factory defines the onset of cavitation as a 3 percent loss of head, cavitation effects are sometimes seen well before a measurable head loss occurs. Continuos monitoring of overall high frequency tracking and rotor axial position represents a practical method to measure the onset of cavitation through the direct measurement of pump response to hydraulic conditions.

A test was conducted wherein the suction pressure was reduced while holding the pump capacity constant at 150 gpm to observe the effects on the monitored parameters. The axial rotor position moved dramatically toward the suction flange as the suction pressure was reduced. Overall high frequency tracking levels increased with a widening trace width. Power decreased and adjustment of the discharge valve held flow somewhat constant. The discharge pressure was not recorded but was observed to decrease as suction pressure was reduced, as expected. When suction pressure was re-established at atmospheric normal, the rotor moved back to its normal position and the other parameters returned to normal operating levels as well.

Overall high frequency tracking
Pump testing revealed several new items regarding overall high frequency tracking. These items are:

  • The mounting of the sensor was not as critical as originally suspected. That is, if the sensor is mounted solidly to the pump casing, the orientation as to axial or radial did not significantly change the magnitude of the overall readings. This premise was confirmed by taking readings with the sensor mounted in the original position--radial at 11 o'clock when viewed from the suction end--and then in the axial direction. The overall readings were fundamentally the same for both transducer locations. It is, however, important to have the sensor firmly attached to the pump, ideally using a stud mount.
  • Lower baseline overall high frequency tracking readings result when the accelerometer is mounted on the upstream side of the cutwater. Higher base line readings resulted when the accelerometer was mounted downstream of the cutwater. Presumably, turbulence or hydraulic noise associated with liquid passing by the cutwater increased the baseline noise.
  • A rule of thumb when using overall high frequency tracking is that less is better than more. The quieter the pump is operating, the better and more trouble-free it will operate. The exception to this rule is dry running.
  • Wider traces of both overall high frequency tracking and axial rotor position are indicative of operating conditions to be avoided in pumps, especially sealess pumps.
  • A time interval of one second or less should be used as a sample rate for capturing overall high frequency tracking data. This capture all of the fast changing operating and mechanical data that can take place with the pump.
  • Mounting the sensor closer to the source of the stimuli--usually on the pump casing--is better.
  • Conditions that raised the overall high frequency tracking level caused the rotor to move significantly in the axial direction. Generally it was in the direction of the suction of the pump on our test pump.
  • Recently a pump was found that had a high baseline noise in its casing area that could not be reduced. Testing at the high gSE level could not detect pump problems because sensitivity was so low. If the baseline noise is much above 20 gSE, then the sensitivity suffers. This makes monitoring questionable and measurements are so desensitized that the data is not useful. The known patterns may not be easy to find and interpret.

Rotor position monitoring
A series of wound coils outside the primary containment protected from the process fluid by the stator liner monitors the rotor. Electrical signals from the coils continuously monitor the actual running position of the rotor. The device simultaneously detects changes in rotor position both in the axial and radial direction. Comparing the output of the instrument to the original factory test baseline for a new pump determines the condition of the radial and axial bearings. The control center of the monitor is a microprocessor that provides digital, analog, and relay outputs. Analog output was used for each variable during these tests.

After the initial calibration, wear on the radial bearing is determined by a change in the radial direction that is greater than the baseline data for new bearings. The amount of wear is proportional to the change in signal. Normal operation of a sealess pump does not promote wear of the radial and axial bearing surfaces. The process upset conditions leading to lack of lubrication and rapid temperature rise are the main causes of wear in the bearings. Continuous monitoring enables users to trend these damaging events to predict and increase the maintenance intervals.

Analysis of rotor position for many of the upset conditions encountered during the testing gave very good correlation with overall high frequency tracking. The amount of rotor movement will likely be different for various pump models because of differences in the stiffness of hydraulic balance and susceptibility to cavitation or air ingestion. Once baseline data is taken with the pump operating in its process, monitoring of the rotor position and noting changes in position provide data to develop a good predictive maintenance tool incorporating process condition effects on pump wear. This translates to an early warning of potential problems.

Conclusions
There are many benefits to monitoring sealess pumps continuously. Since the trend in process plants is to use distributed control systems, much of the plant equipment is being operated remotely. Presently, the board operator in a control room has little or no feed back on the operating condition of most of the pumps, other than possibly flow and discharge pressure. Most of these parameters are not instrumented with trip or alarm limits based on pump health. It is possible for the pump to be operating in an off design condition or have mechanical damage with no feedback being supplied to the operator. The feedback presently comes in the form of failed equipment and expensive repairs.

Because the feedback and the cause of the pump problem occurs over a relatively long period of time and are usually not immediate, the cause and effect relationship may never be discovered. With continuous monitoring, it is possible to get immediate feedback on the condition of the pump and on the process conditions. Conditions such as cavitation, dry running, and extreme operating parameters that result in pump failure can be detected immediately. Armed with this information, the operator can make decisions to improve operating conditions that prolong equipment life and maintain product quality.

For new pump installations, use the latest technological advances being offered by sealess pump manufacturers such as rotor position monitoring. It is best to use both radial and axial monitoring of rotor position. The data presented in this article demonstrate that there are many conditions that radial monitoring alone will not detect. This is especially true with silicon carbide bearings. Overall high frequency tracking appears to be applicable for older installed pumps that have had catastrophic failures that were not detectable early enough to prevent serious damage to the pump. Retrofitting overall high frequency tracking monitoring systems to sealess pumps helps to minimize maintenance costs and eliminate leaks from damaged containment cans or shells.

If metal-to-metal contact is taking place, then stop the pump and schedule it for maintenance. Other successful monitoring practices used with overall high frequency tracking, alone or in combination with a power meter, are as follows.

Before adding overall high frequency tracking to a pump, know the condition of the pump and record a baseline representing that condition. Overall high frequency tracking values should be fairly low in the ten to 40 percent of full-scale range. The full-scale reading should be in the 0 to 5, 0 to 10, or 0 to 15-18 gSE range. If the baseline level is higher than these levels after installing the monitoring system, then there may either already be a problem with the pump or overall high frequency tracking may not work on this application. The baseline must be fairly low or the system becomes so insensitive that the desired information will be missed. On occasion there have been pumps that are so hydraulically noisy that overall high frequency tracking could not be used.

Take a baseline set of readings with the pump operating at its normal capacity. Do this even if the discharge valve must be throttled to achieve normal flow when the impeller is oversized.

When you detect an increase in overall high frequency tracking, vary the process, if possible, to eliminate the noisy condition. This will help determine whether mechanical or process conditions cause the increase. This is especially true if recent changes in operation affect the pump. If you see a recognized pattern of mechanical failure, then the pump should be shut down for repair.

If adjusting process conditions reduces the overall high frequency tracking levels to normal pump design values, then the pump most likely does not have a mechanical problem but is probably not being operated on its curve. To test this possibility, let the process settle for a short while. Start the standby pump, if one exists, and look at its overall high frequency tracking readings. If these readings are substantially lower than for the questionable pump, leave the spare pump in service. Put the other pump in standby mode and schedule it for maintenance.

If switching pumps cannot reduce the overall high frequency tracking noise, it may be an indication that the pump is sized incorrectly for the process and will be a maintenance problem. Past experience has shown that two pumps seldom exhibit the same problem at the same time unless the cause is process related. The pump application should be investigated for proper sizing and adequate suction conditions. This has been a successful method of determining if the process or damage to the pump is the source of increased overall high frequency tracking.

Consider rotor position monitoring if a new pump is required as an upgrade or for a new application. Rotor position monitoring is an excellent way of tracking bearing condition. It should prevent breeches of the primary containment if monitored continuously.

If the pump rotor moves more than 50 percent past the limits imposed by the bearing's original position, schedule maintenance while monitoring the rotor position continuously. If the wear is more than 70 percent of the original clearance, then stop the pump because catastrophic failure is imminent.

If using abradeable bearing material, trend the radial position output over time as the basis for scheduling maintenance. Depending on process conditions, the wear rate may not be linear. Exercise care as limits are reached. Our testing indicated that continuous rotor position monitoring of both radial and axial positions detects some process problem conditions.

The greatest savings come from detecting conditions that cause pump problems early enough to eliminate them and thus prevent a failure. If early detection of off process operation is not possible, then the next best maintenance practice is to detect a problem at its inception and schedule the pump for maintenance while the problem will result only in minimal maintenance costs, business interruption, and no leakage.