Electric Submersible pumps are particularly challenging to monitor and surveil due primarily to the often inaccessible locations often in deep wells, reservoirs or sumps and relative scarcity of process data. Yet often due to the same geographical limitations these pumps are critical to the manufacturing or production process as they generally don't have backup. Effective, proactive surveillance can be an important strategy to optimize online factor and pump run life.
A different type of Electric Submersible Pump (ESP) was used by Sygnology's client on one of its projects to address specific issues involving the reservoir. This resulted in the ESPs being mounted in caissons constructed in the sea floor thousands of feet below sea level. These pumps operate in an extremely harsh 24/7 service environment and, as a result, experience a relatively small mean time between failures. The real issue, from a financial perspective, is not only the failure, but anticipating the failure so that a quick response for replacement is achievable and unplanned down-time minimized.
The pump environment considered in this case study was quite abnormal. Usually they are mounted at the surface, but the subsea setting of these caisson ESPs rendered them much more difficult and costly to repair or replace in instances of failure. This has additionally resulted in a multitude of communications challenges—for example, being located 8000 feet below sea level means the pumps are inaccessible due to the depths—and thus large-scale, high-speed data collection for numerous pieces of equipment isn’t always feasible or reliable. Instead, the only data available for evaluation was the slower sampled data collected by the general historian, which typically has a sampling rate between one and ten seconds. This is an extremely slow data rate in comparison to the speeds typically used to monitor equipment issues.
The customer had implemented various statistical approaches to attempt to identify when one of these pumps was experiencing damage that had the potential to end the life of the pump. Once those endeavors were deemed to be essentially unsuccessful, the client provided that data to Sygnology team members and asked if we could identify anything in the data that would indicate possible failure markers.
Sygnology quickly went to work identifying new potential analytical approaches and assembling test frameworks. Since it wasn’t feasible to monitor the pumps directly by using any additional instrumentation, historian data was evaluated in search of anomalies. Ultimately, applying Sygnology's signal processing and statistical techniques yielded promising outcomes.
Additionally, Sygnology was able to develop a system which used process data characteristics to detect and alert to both major and minor backspin events in near real-time and implement it on pumps that were currently operational. By analyzing the historical data for pumps that had experienced failures due to backspin damage Sygnology was able to positively correlate the damage and failure modes to instances and duration of backspin over the lifespan of the failed pumps.
During the evaluation period, it was determined that applying Sygnology's monitoring, signal processing, and statistical techniques revealed markers in the historical data that pointed to an issue which progressively developed prior to the ultimate failure of the unit. These markers coincided with time frames in which it was determined that the pump inadvertently spun backwards due to a malfunctioning downstream check valve that allowed flow to reverse through the pump when it was shut down. This backspin has been known to cause damage to the unidirectional thrust bearings in the pump, eventually leading to failure.
Once it was determined that there were some approaches that could be useful in seeing damage reflected in historical data, the client requested a review of additional historical data for indications to other ESP failures. As expected, there were additional historical markers found in other pumps. These techniques built confidence that some of these failures could potentially be predicted even in the restraints of low sample rate and data compression.
Using the Sygnology developed signal processing and statistical techniques, Sygnology was able to conclusively correlate the frequency and duration of both major and minor backspin events to the amount of backspin-related damage seen in failed pumps.