As an integral part of onshore and offshore drilling, mud pumps circulate drilling fluids to facilitate drilling oil and natural gas wells. Mud pumps stabilize pressure and support the well during the drilling process and drilling fluids provide friction reduction and a means to remove cuttings. We created a leak detection system for hex pumps. The hex mud pump (see Figure 1) has six pistons, six suction valves, and six discharge valves. The six pistons are driven by a rotating, asymmetric cam. We designed a patented leakage system based on CompactRIO in house. The system monitors the suction and discharge valves using accelerometers.
The Case for an Automated Monitoring SystemValve leaks in piston pumps are often discovered at a late stage when the leaks are so severe that they induce large discharge pressure fluctuations and create washout damages. When a severe leak is detected, we localize it manually by listening to the fluid modules while the pump is running, but it is difficult to uniquely localize the leak and distinguish between a suction valve leak and a discharge valve leak.
Human exposure to hazards is the main disadvantage of manual detection, verification, and localization. Mud pumps convert large amounts of power and often output high pressures up to 350 Bar discharge. Additional equipment in pump rooms also generates high acoustic noise pressure levels that can exceed 100 dBA and cause health and hearing damage if humans are not correctly protected (see Figure 2).
Valve leaks often develop quickly, so manual detection gives very little time to prepare for exchanging the defective valve(s) after the leak is detected. If the leak source is uncertain, searching for the defective valve(s) can be costly and time-consuming.
To overcome these drawbacks, we needed a remote system for detecting and localizing pump leaks.
Discovering the Vibration MethodDuring a vibration monitoring project for hex pumps, we discovered the possibility of detecting leaks using accelerometers. We recorded vibrations at different locations, both on the pump and on the discharge line, along with suction pressure, discharge pressure, and pump speeds for different pump conditions. We used a 20 kHz sampling frequency and recorded 5-second snapshots with intervals of a few minutes. On one occasion, the vibration signature significantly changed during a 15 minute period. We soon realized the spot was a growing valve leak.
After the initial discovery, we performed more tests to further explore the leak detection possibility. Figure 3 shows vibrations of all six valve blocks when discharge valve 2 (D2 valve) has a severe leak. The trace numbers indicate the accelerometer/valve block number. The high intervals of the dashed help curves represent the theoretical suction phases that happen when the suction valves are closed. These curves offer easy interpretation of the vibration signals and are derived from the proximity of the sensor signal (not shown). The low values of the help curves represent the theoretical closing of the discharge valves, which happens when the respective pistons retract. The leak intervals have a lag time shift relative to the theoretical intervals. This time shift is on the order of 25 ms and comes from 1) valve inertia causing delayed valve closing, and 2) fluid compressibility causing a finite piston stroke to compress and decompress the fluid.
Analyzing the frequency spectra indicated that the leak induces strong, broad banded noise from 3 kHz up to the Nyquist frequency of 12.5 kHz (half the sampling frequency of 25 kHz). The overall noise level increases by a magnitude of 30 dB.
Leak Detection SystemBased on that encouraging experience, we wanted to include this condition-based maintenance system as a standard feature on all hex pumps. We developed the system as a stand-alone module to add to the existing hex pump control system (see Figure 4). Slightly simplified, it consists of the following components: accelerometers (one per valve block), a proximity sensor picking up pump speed and phase, a discharge pressure sensor, an embedded monitoring system (CompactRIO with NI 9234 acquisition modules for powering the accelerometers and acquiring high frequency data), signal processing software and alarm logics implemented using LabVIEW software running on the CompactRIO monitoring system, and an HMI user interface developed in LabVIEW.
The data acquisition and signal processing is briefly described by the following steps:
1. Capture high-rate data (25 kHz sample rate) from all sensors during a short time interval covering at least one pump cycle
2. Bandpass filter the acceleration signals to minimize the influence from ambient pump vibrations
3. Analyze the timing signal to find pump speed and cam angle
4. Construct adjusted window functions that selectively pick the filtered acceleration signal in every valve closing phase (adjusted here means narrowed and time lag corrected so that valve closing spikes are excluded)
5. Use these windows to calculate the RMS vibration level for each valve closing phase
6. Normalize the vibration levels through division by the median vibration level
7. Set a leak alarm if one or more of the normalized vibration levels exceed a specified threshold during a certain time interval
The default sampling frequency of the signals is 25 kHz, but the system can handle higher rates if necessary. The bandpass filter is optional, but experience shows that it improves contrast and detection sensitivity. Signal strength normalization by the median vibration level makes the detection nearly independent of the inherent ambient vibrations, which increase rapidly with increasing pump speed and discharge pressure. The last requirement, that the detected leaks last for a set time, eliminates erratic alarms caused by debris or large particles that can cause temporary seal malfunction.
We can remotely verify the leaks detected automatically by signal processing in several ways. First, the operator can view and interpret the vibration signals directly from graphs. Second, the operator can selectively listen to the recorded acceleration signals as audio signals to hear the leak sound. Third, the operator can check to see if the mean discharge pressure is stable or dropping. Lastly, the operator can see if the lowest pressure harmonics are growing.
The human ear/brain is an extremely sensitive instrument for picking up abnormal sounds. If the leak sound is too far up in the treble frequency range to hear, we can play the signal back with a lower sampling rate, thus transforming the leak noise into a more audible frequency band for the human ear.
We can use the desktop application shown in Figures 5 and 6 on a terminal to review the LDS and read raw logs and trend files directly from the LDS. This additional feature gives the operator the chance to get a closer view of the vibrations and perform audio playback to the user. Also, we can view the high-rate log of the discharge pressure to reveal a cyclic variation drop. This helps provide a better understanding of what is happening with the valves.
Figure 5 shows a 1.5 second snapshot of the vibration signatures after a severe leak developed in the D3 valve. It shows the filtered vibration signals from all six accelerometers during a 1.5 second snapshot. Acceleration signal 3 has enhanced noise amplitude during the D3 phase. A closer look at the other signals reveals that the leak induced vibrations are transferred to the other accelerometers during the same time intervals. However, the vibration transfer is relatively low, actually less that -20 dB for neighboring valve blocks and even less for the other blocks, so vibration transfer is not a serious problem in hex pumps.
Based on the field experience of the new leak detection system, we concluded that our leak detection method offers many advantages over current practices, including the following:
CompactRIO and LabVIEW proved to be fast tools for prototyping our system and gave us an embedded deployment system that we can reliably retrofit to existing pumps. In comparison to other leak detection methods, which are based on analyzing discharge pressure, we found our vibration-based methods to be more robust and reliable, especially when it comes to localizing a leak. Our studies show that an alternative method can be applied for shaft-driven piston pumps having either an integrated valve block or split blocks with a high vibration transfer. Leak localization for this kind of pump is mainly based on the phase of the pulsating vibration level. We can use it to localize one dominating leaky valve at a time.