System 1 Helps Maintain Turboexpander Operation at FCC Plant

System 1 condition monitoring and diagnostic software helped Ecopetrol’s oil refinery in Cartagena, Colombia (the Cartagena Refinery) determine the cause of sudden vibration increases in its fluid catalytic cracking (FCC) plant’s turboexpander that led to unit shutdown. In addition to helping the plant extend its operation four-fold, the effort has: lowered costs associated with uncovering the cause of failures, reduced production losses, effectively identified a solution to the vibration problems, and helped define mitigation actions needed to optimize the performance of the power recovery process.

Apolinar J. Moreno Arrieta
Rotating Equipment Engineer
Ecopetrol S.A.

Carlos A. Gómez
System 1 Technical Leader Latam



Due to process requirements, the plant’s turboexpander unit is forced to exceed the catalyst loading upper limit (established by design), resulting in high vibration after a short period of operation. As the vibration builds it creates a mass unbalance in the expander’s rotor that ultimately results in high vibration amplitude trips.

After each vibration trip, a thermal cycle must be performed to break the adhered catalyst before the expander can be restarted. This process results in three lost days of operation to the plant— translating to a big impact on the Cartagena Refinery’s production capability in terms of reliability and productivity.

On February 21, 2016, the unit was started and it exhibited normal vibration—with amplitudes around 0.5 mil pp—even when loaded with high temperature gases. After a few days in operation, the vibration started to increase, and information obtained from the System 1 online management system and the portable data acquisition system ADRE* Sxp/408 DSPi (both from the Bently Nevada product line), confirmed a mass unbalance. The unbalanced condition appeared to increase at a near-constant rate with occasional step changes—sudden increases in amplitudes— making it impossible to control using the routine abrasive cleaning procedure of injecting walnut hulls.

An analysis of the abnormal response concluded that the amplitude step changes were caused by adhered material loss at the front part of rotating blades and blade tips.

On March 21—only 30 days after operation began—material that accumulated in the blade tips led to rubs against material that accumulated in the shroud, causing reverse precession components prior to machine trip, with overall vibration amplitudes above 4 mils p-p.

After startup on March 22, the overall vibration values were similar to the ones recorded during the initial startup (0.5 mils p-p), with normal rotor dynamic behavior, discounting any mechanical problems (such as broken blades and shaft bow).


Figure 1. S1 Power Recovery Train, Turboexpander – Cartagena Refinery

Condition Diagnostics

The machine’s power recovery train consists of an electric motor coupled to a hydraulic clutch for starting, and a generator that can be used to generate electrical power or to provide mechanical power as an electric motor. The generator is coupled to a variable speed gearbox connected to a centrifugal compressor. Finally, there is a 17403 HP turboexpander for FCC flue gas in overhung configuration that runs at 5,500 rpm, with a resonance speed of 3,100 rpm and a rotor weight of 2,900 pounds (as shown in Figure 1).

The unexpected trip alerted the plant’s Reliability team to begin to explore the root cause of this abnormal vibration behavior (continuous increase followed by return to original vibration state after three shutdown events).

The turboexpander is equipped with 3300XL Bently Nevada displacement transducers in both radial bearings (DE and NDE), as well as two thrust position transducers that protect the unit against excessive axial rotor movement.

The overall vibration trend from February 21 to April 2, 2016, is shown in Figure 2. On March 4 vibration amplitudes started to increase and on March 21 the vibration amplitudes peaked, reaching 4 mils p-p and then tripping the machine train.  After restart on March 23, vibration values returned to the original low levels.


Figure 2. Vibration trend from February 21 to April 2 for Turboexpander transducers.

This unexpected vibration trip and subsequent restart to “normal” condition (with the possibility of a repeating cycle of vibration), was used to study the abnormal condition that was affecting plant operation and full refinery production.

The Cartagena Refinery was able to take high-speed photography at the rear side of turboexpander blades, but not at the front. In some pictures, material buildup at the blade tips was noted, but it was unclear how much catalyst had actually adhered to the rest of the rotor. It was not possible to draw a conclusion regarding the effect of rotor mass change based on operating conditions.

During the abrasive walnut hull cleaning process carried out on March 17, some peaks in thrust position were detected at the exact moment when the cleaning process started. Additionally, peaks were observed in the shaft centerline for bearing 9 (as shown in Figure 3). For reference, four cleaning processes were performed (a, b, c and d), with vibration amplitudes rising after midnight of March 18. After the 7:30 am cleaning, vibration increased up to 3 mils p-p, activating the vibration alarm.


Figure 3. Vibration and thrust position trend (gap) and shaft centerline plots from March 17 - 18. During this period, the turboexpander was cleaned four times, and vibration increased from midnight of March 18, reaching alarm levels at 8:00 am.​

This undesirable condition that occurred after cleaning created confusion to both the operators and the Reliability team. It was expected that the vibration would decrease after cleaning, instead of the opposite condition that finally resulted in a vibration trip.

The Cartagena Refinery and Baker Hughes, a GE company (BHGE), joined forces to perform a comprehensive data analysis using System 1 software along with the ADRE 408 instrument to collect additional vibration data and attempt to confirm the source of the observed change in turboexpander vibration. As previously mentioned, the other transducers in the machine train did not change during the turboexpander vibration change.

Steady state polar plots for the bearing 9 vertical probe are shown in Figure 4, as well as direct orbits from proximity transducers at bearings 9 and 10 of turboexpander, recorded during different conditions. The first condition depicts the machine on February 23, recently after it was first started, with vibration values below 1 mil p-p and the 1X vector below 0.3 mil p-p. On midnight of March 16 the vibration increased above 2 mils p-p at bearing 9, with a change in the 1X vector phase as well as amplitude to 1.5 mil p-p. On midnight of March 19 the vibration increased and the 1X vector amplitude went above 3.5 mil p-p, with the phase changing almost 180 degrees. The direct amplitude of vibration was now above 3.8 mil p-p, very close to the danger setpoint established by the OEM.

Figure 4. Polar plot for bearing 9 vertical transducer and direct orbits for bearing 9 and 10 of the turboexpander at various times. Note the change in orbit shape and 1X phase during this period of analysis. On February 23, the orbit in bearing 10 was circular, turning into a completely flat orbit on March 19.

Figure 5. High-speed photography captured with stroboscopic light from rear section of the turboexpander blades. The left picture, taken on March 18, shows evidence of blade and shroud deposits. The right picture, taken two days earlier, shows deposits only in the blades and roots.

Figure 6. High-speed photography captured with stroboscopic light from the rear section of turboexpander blades. The left picture was taken before a walnut hulls injection cleaning process, the right one was taken after the cleaning.

Knowledge Paid the Bill

After performing the analysis and discovering the mechanism of catalyst deposit formation that affected the rotor balance condition, it was necessary to take actions to increase the time between shutdowns. The number of shutdowns per year was estimated to be 12, based on actual operating conditions involving the amount of catalyst in the turboexpander, which was over specifications.

System 1 software was used to configure an acceptance region to track the 1X vector during operation and activate an alarm when the vector was out of range (as shown in Figure 7). The alarm would trigger an investigation to understand when it was possible to clean the turboexpander.

The final decision to clean the rotor was based on rotor balance theory that considered the effect vector (i.e., the difference between original 1X vibration and final 1X vector) to determine cleaning process effectiveness. When a vibration change due to catalyst accumulation causes an effect vector that reduces vibration amplitude, the Reliability team sends a notification to Operations to schedule turboexpander cleaning.

Figure 7. Acceptance region with the reference vector in orange and new vector out of this acceptance region. In red is the “effect vector” due to mass change inside the turboexpander blades after cleaning​

The vibration data collected on March 17⎯when the 1X vibration vector changed⎯was used to determine the existing System 1 asset condition monitoring system’s ability to detect this change. The vibration trend, direct orbits before and after the event, as well as polar plots for proximity transducers installed at turboexpander bearings are shown in Figure 8. The orbit change was over 400% in amplitude at bearing 9, while the orbit shape at bearing 10 changed from highly elliptical to completely flat. The flat orbit was caused by the reduction in the clearance between the shroud and the blade tips due to the catalyst deposit buildup. This information was confirmed after shutdown on March 23.

Figure 8. Vibration trend, orbits and polar plots during increasing vibration event on March 17. Direct orbits show the effect of vibration amplitude after this event. Polar plots show the phase and amplitude changes during the same period.​

This phenomenon was shared with Refinery Managers, and the new cleaning process was explained. The new process was based on dynamic behavior, as opposed to perception of vibration amplitude change, breaking the paradigm of strictly following manual recommendations even when an online monitoring system is in place to detect early changes in the dynamic condition.

On June 16 a “preventive” cleaning was carried out, after which vibration increased above alarm setpoints in three out of four proximity systems (indicating that vibration was real and not caused by false signals). The Reliability team recommended that Operations stop any other cleaning process until they established a course of action based on conditions (as previously explained).

The vibration trend from June 6 to July 6, 2016 was recorded for the vibration transducers installed at turboexpander (see Figure 9). On June 27 and July 1 a controlled cleaning was performed to decrease vibration. Overlaid in the plot are the direct orbits from the turboexpander bearings for the high vibrations level condition from June 23 (orange), and for the new data taken on July 6 (blue). The decrease in orbit size and, thus, vibration amplitudes was clearly observed, and confirmed the significant benefit of the analysis efforts carried out by Reliability team using the implemented System 1 online asset monitoring system to deliver a long-term solution to this operational problem.

Figure 9. Vibration trend from proximity transducers for the turboexpander from June 6 to July 6. Highlighted are orbits from June 23 (orange) and July 6 (blue) to show the effect of controlled cleaning. On June 17 a preventive cleaning was carried out that increased vibration amplitudes above alarm setpoints.​

This knowledge allows Ecopetrol to operate this FCC plant for more than four months without the need to apply a shutdown the plant for a three-day cooldown process for thermal cleaning. Now the company’s Operations and Production departments can comply with customer and country requirements for fuels.


  • The excessive vibration amplitudes experienced by the turboexpander were due to the process and not to the machine itself. The process involved catalyst deposit buildup in stationary and rotating components, leading to rotor mass unbalance and a synchronous or 1X amplitude increase when some of these deposits detached.
  • A causal relationship was identified between the non-linear response of the expander when the amplitudes of vibration increased and remained in high values (despite frequent cleaning using walnut shells), and the 1X phase changes. This suggested that the 1X phase changes were a fundamental variable to be monitored, along with the 1X amplitude, which provided direct information about the condition of catalyst accumulation and subsequent non-uniform distribution on the rotor.
  • The condition monitoring and diagnostic technology available from Ecopetrol’s support and rotating equipment Reliability staff were critical to detect the change in the dynamic behavior of the expander well in advance. With the additional help of both the OEM and BHGE’s specialized personnel, it was possible to achieve an understanding of the behavior of the system, as well as the failure mechanism.
  • The correct operation of the expander requires that the amount of catalyst being used remains within design allowable limits. For this particular case, a flow greater than two or three times the maximum allowable limit reduced the expected continued run time from two years to just one month.
  • The development of knowledge of this critical asset, along with existing monitoring technology has generated initial economic benefits of more than $5 million USD, both by extending the turboexpander’s mean time between failures (MTBF) four times and by eliminating the failure itself by following the defined actions that are based on the dynamic analysis and thermodynamic performance of the entire power recovery train.

Page bottom button