1 Oil Health Monitoring Oil sampling analysis has been the backbone of all oil reliability programs for the past century. From early analysis technology used by oil labs to the more modern miniaturized benchtop equipment, these have transformed the way and speed to which we make decisions around preventative maintenance. The latest of these advancements has been new in-line, real-time oil quality monitoring sensors that deliver on the earlier promises of true online monitoring capabilities. The value from oil samples is centered around oil health itself. Properties such as oxidation, TBN, TAN, additive packages, viscosity, water, fuel, soot, etc. are all used to assess the remaining-useful-life (RUL) and identify preventative, as well as corrective actions. A few notable challenges with oil sampling are the time delay between sample and results, access to remote and mobile assets, infrequency of sampling compared to reliability events, Research 31 Tribologie + Schmierungstechnik · 70. Jahrgang · eOnly Sonderausgabe 1/ 2023 DOI 10.24053/ TuS-2023-0031 Combining Oil Health, Level, and Vibration to Achieve Complete Machine Monitoring Jeremy Sheldon, Mark Redding* * Jeremy Sheldon Poseidon Systems LLC, Victor NY, USA Mark Redding Poseidon Systems LLC, Victor NY, USA New digital technologies are transforming the way and speed at which decisions are made for preventative maintenance. These advancements include new in-line, real-time monitoring sensors that deliver on the earlier promises of true online monitoring capabilities when compared to traditional lab-based oil analysis, visual determination of oil level, and route-based vibration data collection. Notable challenges with manual methods are the time between sample and results, access to remote and mobile assets, infrequency of sampling compared to reliability events, and elevated risk of human error. Because of these challenges, many operators are moving towards online monitoring. Industries such as energy, mining, rail, marine, etc. have all started adopting online oil monitoring programs and it is expected to become standard practice over the next few years. Recent sensor advancements are now capable of measuring a variety of oil health related parameters including oil condition, relative humidity, temperature, and others. These new sensors can detect most, if not all, key oil events and project the remaining useful life of the oil while the asset is in operation. However, oil level, the most critical measurement is typically overlooked. While oil level sensors exist, their use is not widespread due to a variety of reasons, one of the main being sensor form factors. Vibration analysis has proven to be one of the most reliable and quantitative indicators of gearbox compo- Abstract nent failures. Acquiring, analyzing, and monitoring of a machinery’s vibration can reveal much about that machine’s health. Vibration is often the preferred machinery monitoring solution. Each sensing technology and its interpretation offer varying fault detection and isolation performance. Those implementing a condition monitoring system (CMS) must evaluate the sensors, fault detection performance, cost, and various other aspects before implementing a CMS. Cost and complexity of implementing both solutions may require selection of either an oil condition sensor or vibration sensing but not both, reducing the overall CMS performance. To help address the above-mentioned challenges, the authors developed a multi-sensor device that can be used to monitor different failure modes in a single small form factor design. The developed sensor monitors the oil level, quality, water contamination, magnetic flux, and vibration in real-time. Keywords Preventative maintenance, online monitoring, condition monitoring, multi-sensor device 1.1 Oil Condition Various technologies provide the ability to measure multiple properties of a lubricant. As mentioned above, two of the more widely used methods for online oil monitoring are electrochemical impedance spectroscopy and measurement of the dielectric constant or ratio (simply referred to as dielectric) of the oil/ fluid, see Figure 1. Dielectric is a measurement of how well a material conducts electric current compared to how the same field would conduct in a perfect vacuum [1]. The relative simplicity of the sensing technology is well suited for online applications. As oil degrades contaminates and chemical changes affect the measured dielectric of the oil. Dielectric may increase or decrease depending on the change to the oil. Typically, for many important health changes, for example depletion of additives, the oil dielectric will decrease. The developed sensor incorporates a dielectric measurement for tracking oil condition. Most oil sensors are best used to determine anomalous conditions and when a sample of the lubricant should be taken and sent to a traditional oil-lab for a more traditional oil analysis before any further action is taken or the output can be used to directly determine what action should be taken when significant changes are detected. Research 32 Tribologie + Schmierungstechnik · 70. Jahrgang · eOnly Sonderausgabe 1/ 2023 DOI 10.24053/ TuS-2023-0031 and high risk of human error. Because of these challenges, many operators are moving towards online fluid monitoring. Typically, online sensors are electrochemical impedance, dielectric, conductivity, or permittivity measurement devices detecting degradation of the oil. Some methods and sensors can detect degradation of not only overall quality, but estimating percent soot, total base number, relative humidity, additive depletion, etc. Regardless of sensing method, effective oil condition sensors must detect most, if not all, key oil events and have an output that trends with the life of the oil while the asset is in operation. While these sensors cannot duplicate lab analysis results, they can provide the necessary insight to make preventative maintenance decisions before damage occurs. Online oil quality sensors are key enablers to shift the oil sampling paradigm from periodic to true conditionbased sampling. To completely monitor the condition of an oil, multiple parameters are required. At a minimum a combination of sensors to monitor oil level, track overall oil health, and a monitor water activity are recommended. Figure 2: Example Oil Condition Sensor Data Figure 1: Notional RH vs PPM Comparison To illustrate, a series of laboratory sub-scale testing was performed to determine the performance of a commercially available sensor at varying oil contamination levels. Two separate testing phases were performed to assess the sensor response to coolant contamination and fuel contamination. sensors at varying oil contamination levels. Coolant and fuel were used as they are common sources of contamination, however the sensor performance is expected to be similar for other fluid contaminations, for instance with incorrect fluid addition. In the first test, coolant was added incrementally. One of the measure parameters trended very well with both coolant and fuel were added, as shown in Figure 2. In addition, the parameter also responded proportionally when increasing amounts of contaminates were supplied. Consistent and proportional response to oil condition of change of interest is a critical aspect to assess when selecting an oil condition sensor. The sensed parameters must behave in a proven way to allow confidence in the resulting decisions. In addition, one must ensure the sensor is sensitive to the failure modes of the target machine. 1.2 Fluid Level When implementing a condition monitoring solution, fluid level is often overlooked. Insufficient lubrication is obviously detrimental to the machine’s mechanical components. Additionally, low oil can also cause the remaining oil to degrade more rapidly. Various studies report inadequate, or no lubrication is responsible for 30- 80 % of bearing failures[2,3]. The authors have many examples of low oil or oil out events causing massive gearbox damage. Many of these would have been avoidable had the actual oil level been actively monitored. Therefore, monitoring the level of the fluid is a critical aspect to overall machine health. An integrated level sensing mechanism is built into the developed sensor. Online real time monitoring of oil level offers many benefits over traditional manual/ visual readings. Monitoring continuously allows trending of oil level with machine operation to confirm adequate lubrication at all operating conditions. Alarms and notifications can be raised at the instant oil levels drop avoiding the time delay associated with manual readings that may occur days apart. In addition, online monitoring of level removes the need for manual checks that can be expensive and dangerous depending on the location of machinery. 1.3 Relative Humidity The developed sensor incorporates a relative humidity sensing element that uses a hydroscopic dielectric compound in a capacitive sensing element to produce an output which correlates directly to the relative saturation of moisture in the working fluid. Thus, by monitoring the change in capacitance, relative humidity can be derived. Like water is detected in the atmosphere by monitoring relative humidity, relative humidity sensors in oil provide a measure of the dissolved water present. While in the dissolved state water likely poses relatively low risk to the system; however, as the level approaches the oil’s saturation point the risk of free water and emulsion formation greatly increases. When interpreting the data and setting limits one must consider the risk the free water poses to the specific system. In most laboratory reports water content (both free and dissolved) is expressed in parts per million (PPM). The acceptable PPM levels vary greatly by oil type because the saturation points of oils are highly variable. Oil saturation points vary due to the oil chemistry, additives, age, depletion, and other influences. Saturation points also vary with oil temperatures, meaning that for fixed PPM limit the actual saturation level of the oil will change with temperature. A more relevant to machine health way to express water content in oil is as a percentage of saturation, or relative humidity. Expressing water content in relative humidity accounts for the changing saturation point and at any given time directly indicates the risk of free water without needing to know the often difficult to find oil specific saturation curve. Oils can typically absorb more water with increasing temperature. So, for a given fixed PPM limit at lower temperatures the oil could be close to or above saturation limit while at higher temperatures the oil could be relatively dry. A notional example is shown in Figure 3. In this example an assumed 1000 PPM of water is present. It should be clear that for given fixed PPM water level results in a variable risk of free water or heavily emulsified oil across changing operating temperatures, thus the risk to the asset condition is variable. When implementing a remote monitoring solution specific to water content in oil, one must consider the entire operating range of the system. Research 33 Tribologie + Schmierungstechnik · 70. Jahrgang · eOnly Sonderausgabe 1/ 2023 DOI 10.24053/ TuS-2023-0031 Figure 3: Notional RH vs PPM Comparison part of this analysis, shaft speeds and machine geometry are often used to calculate component-specific frequencies useful in fault isolation. Spectral (typically Fast Fourier Transform (FFT)) and time domain analysis is an important part of any vibration monitoring system. Typical systems extract condition indicators or features from the collected vibration data that are used for trending and machine health assessment. For instance, a commonly used vibration indicator, or feature, is root mean squared (RMS). RMS is a relatively easy to compute statistical feature that is indicative of the overall broad band vibration level. One of the limitations of RMS is its lack of fault isolation capabilities when compared to methods involving frequency analysis, such as FFT analysis, which will be described later in the paper. Nonetheless, RMS is an important part of many monitoring systems, including integral to accepted standards, such as ISO-10816-21 [6]. This paper will not get into the finer details of vibration analysis but will instead focus on the application and results from real world examples. 2.2 Motor Magnetic Flux Measurement of shaft speed is important for knowing the operational state of an asset. In addition, knowing the shaft speed is critical for many vibration diagnostics involving frequency domain analysis. Vibration diagnostic techniques often use shaft speeds and machine geometry to calculate component specific frequencies useful in fault isolation. Typically, shaft speed is measured using a dedicated sensor directly affixed or with direct access to the target shaft. The requirements of these types of speed sensors can make installation a challenge due to the additional complexity of adding another sensor, proximity to the rotating shaft needed by many sensors, and the difficulty of retrofitting. To overcome these challenges, the developed sensor integrates a magnetic flux measurement for detection of electrical issues and estimating shaft speed. Magnetic flux is a measurement of how much magnetic field passes through a given area, in this case the sensor element. Research 34 Tribologie + Schmierungstechnik · 70. Jahrgang · eOnly Sonderausgabe 1/ 2023 DOI 10.24053/ TuS-2023-0031 1.4 Temperature Although seemingly trivial, machine temperature (bearing, housing, surface, or oil temperature) is a critical part of condition and machine monitoring. Measurement of machine internal or surface temperatures can provide an overall indicator of machine health, even if the indication may be late in the fault progression. Additionally, temperature measurements can often provide insights into secondary system performance, such as cooling systems. Prolonged periods of high oil temperatures can indicate poor lubrication condition, moderate to severe mechanical damage, or other abnormal operating conditions that require remediation. Oil condition can more rapidly degrade under abnormal temperatures. In addition, some of the parameters measured by sensors are temperature dependent and a real time oil temperature reading is necessary to ensure analysis conclusions are accurate. Therefore, temperature measurements are a valuable addition to any remote monitoring system. 2 Mechanical Health Monitoring To monitor the health of mechanical components, a wide range of sensors, data acquisition devices, and processing techniques exist. In fact, there are too many sensor types to list herein. Therefore, the authors selected vibration as the sensor type for the developed multi-parameter sensor. 2.1 Vibration Vibration analysis has proven to be one of the most reliable and quantitative indicators of gearbox component failures. Acquiring, analyzing, and monitoring of a machinery’s vibration can reveal much about that machine’s health. Typically, vibration is measured via an accelerometer type sensor mounted directly to the machine. The measured raw vibration signal must first be analyzed to extract useful information form the acquired waveform [4,5]. Many analysis techniques exist, usually applying one or more signal processing techniques to process the time waveform into usable reduced-order information, sometimes called features or condition indicators. As Figure 4: Example Magnetic Flux Measurement, Shaft Speed [60 Hz] Highlighted The magnetic field is generated by the motor. The sensor measures this field and estimates the shaft speed, compensating for slippage. Additionally, the measure flux signal can be used for motor diagnostics [7]. An example of the measured signal from an operating pump motor is shown in Figure 4. 3 Multiparameter Sensor Overview To simultaneously address needs for oil and mechanical health monitoring in a single sensors, the authors have developed an integrated sensor package that combines all these parameters into a single sensor. The integrated sensing device facilitates conditioned-based maintenance practices by combining multiple sensors into a single package to enable the early indication of lubricant issues and avoiding costly downtime. The device is installed in place of an oil level sight glass, as shown in Figure 5. As mentioned above, the sensors utilize capacitive measurement to monitor oil level, a dielectric measurement to track overall oil health, and a water activity measurement to monitor for water contamination. The sensor also measures vibration at the same location. The sensor does not require any consumables or complicated installation and it is applicable to a wide range of fluid types. The technology is well suited for online applications. 4 Example Applications The following section summarizes a few use cases of the sensor highlighting each sensor parameter. 4.1 Vibration Measurement Example To assess the vibration monitoring sensor integrated into the multiparameter sensor, the authors installed the sensor’s MEMS-based wireless vibration nodes, hereafter referred to as vibration nodes, on each of three different 1.5 MW wind turbine generators. The sensors were installed radially orientated near the drive (nearest the main rotor) and non-drive end (furthest from the main rotor) generator bearings. The focus of the installation was to assess the capabilities of the sensors and integrated system. As such, the sensors were installed on three generators of known health states: healthy bearing, incipient bearing fault, and severe bearing fault. The faulted bearings were earlier identified and confirmed via borescope inspection. Data from the vibration nodes was collected over approximately eight months to assess the vibration system’s performance on each generator, focused on the bearings. The monitoring period extended beyond the bearing replacement to compare before and after repair vibration levels. The three RMS trends over time are shown in Figure 6. As expected, higher vibration levels were observed on faulted generator bearings during the first few months. The generators with faulty bearings were inoperative while repairs were in progress, the healthy turbine continued operating. Once repairs were complete, all generators exhibited similar vibration levels, except for a short duration of higher vibrations during the first start up. From the clear difference in the before and after repair vibration levels confirms the Research 35 Tribologie + Schmierungstechnik · 70. Jahrgang · eOnly Sonderausgabe 1/ 2023 DOI 10.24053/ TuS-2023-0031 Figure 5: Developed Multiparameter Sensor Figure 6: Generator Bearing Vibration Trend First, the multiple oil changes are clear. With each new oil change the dielectric returns to roughly the same level. The repeatability of the sensor response as the oil ages is also clear. Reliably trending data enables confidence in the conclusions drawn and knowledge derived from the data. It also simplifies the analytics, requiring simple thresholds and slope calculations. Second, there was a clearly different trend in the oil change in January, highlighted in red. The dielectric rate of increase slowed and began to plateau. This is indicative of abnormal oil health. The asset owner was notified that an oil ample should be taken. The oil sample confirmed high oxidation was occurring due engine malfunction. Once the repair was performed, the sensor data indicated the oil condition returned to normal, thus confirming the corrective maintenance action was taken. 4.4 Water Intrusion Example The sensor’s relative humidity measurement from a pump system is shown in Figure 9. During normal operation, the oil water content is relatively high, above 50 %, but normal for this type of pump application. However as noted, the sensor measured a short duration of 100 % RH indicating an issue but quickly returning to normal levels. At this point in time the operator of the pump was notified of a potential seal failure. Unfortunately, before the scheduled maintenance the seal completely failed resulting in large amounts of water contaminating the oil. The pump seal was repaired and put back into service with sensor data confirming the oil was drying out as result of the repair and top off. Without the sensor, the pump would have likely operated with the heavily contaminated oil resulting in additional damage. 4.5 Dielectric Measurement Example The final example is from a laboratory-based experiment conducted to evaluate the level sensor and vibration sensor in combination. The developed sensor was installed on a pump gearbox sump to monitor the lubrication oil level and pump itself. The sump oil level was slowly reduced over time to simulate an oil leak. Oil was refilled the following day. The level measurement during the test is shown in Figure 10. Real time alarms would have quickly indicate the reduced oil level, alerting maintenance teams to take immediate action to avoid damage due to improper lubrication. The vibration trend from the test is shown in Figure 11. There was no discernable difference in vibration when the pump was running without lubrication. The level measurement provided the first indication, and vibration may have picked up after actual damage started vs immediate response from level sensor. Although no mechanical damage occurred during the short test, a CMS based only on vibration would have missed the oil out event resulting in increased risk to secondary damage Research 36 Tribologie + Schmierungstechnik · 70. Jahrgang · eOnly Sonderausgabe 1/ 2023 DOI 10.24053/ TuS-2023-0031 repair adequately remedied the underlying fault state. However, from the trend, it is difficult to tell the difference between the fault severities, which is important for understanding the likely fault progression and timeline in which repairs are required. 4.2 Level Measurement Example The level of an industrial pump was measured over the course of many months. While no critical anomalies were detected over this time, an interesting response was noted. A single day of data is shown in Figure 7. The monitored asset had an automatic oil fill system that is intended to automatically replenish the sump oil when necessary. From the data, the top off events are clear. Although not indicative of an issue, the condition monitoring practitioner is able to confirm normal operation just as much as abnormal operation. 4.3 Dielectric Measurement Example An example of the measured oil dielectric from a engine application is show in Figure 8. Over the course of approximately 8 months the asset oil health was monitored using a dielectric measurement like that incorporated into the developed sensor. Two observations can be made from the data. Figure 7: Pump Gearbox Oil Level Trend Figure 8: Oil Dielectric Trend and machine failure. An example such as this highlights the need for multiple sensors to detect the variety of important machinery failure modes and the risks associated with reliance on one sensor type. 5 Summary In condition monitoring, there is no single sensor type or analysis method that will work for every component and every failure mode. Combining sensors will certainly result in the best results for any condition monitoring application. However, one must always balance cost and complexity with the required performance. In addition, certain components and machines will require specific monitoring. For instance, oil debris cannot be used on grease-lubricated components, and vibration systems cannot detect clocking bearing failure modes. Unfortunately, budgets often restrict what is possible, often requiring selection of only one monitoring approach. In that case, one must be careful in selecting the approach as there is no single sensing solution capable of detecting every failure mode. The advent of the internet of things era has helped reduce the technological and cost barriers of deploying multiple sensors on a single asset. IOTera technologies have enabled combination of sensors in reasonably priced packages for every increasing number of machines beyond the most expensive of critical assets. References [1] https: / / www.machinerylubrication.com/ Read/ 28778/ dielectric-instruments-oilanalysis [2] SKF, “Bearing damage and failure analysis,” SKF Group, June 2017. Research 37 Tribologie + Schmierungstechnik · 70. Jahrgang · eOnly Sonderausgabe 1/ 2023 DOI 10.24053/ TuS-2023-0031 Figure 10: Oil Level Trend During Simulated Oil Leak Figure 11: Vibration Trend During Simulated Oil Leak Figure 9: Oil Relative Humidity Trend From Pump Seal Failure Phase 1 and Phase 2 Testing,” National Renewable Energy Laboratory, 2011. [6] ISO 10816-21, “Mechanical vibration — Evaluation of machine vibration by measurements on non-rotating parts — Part 21: Horizontal axis wind turbines with gearbox,” First edition, May 1, 2015. [7] Yazidi A., H. Henao, Capolino G. A., Artioli M., Filippetti F., Casadei D., “Flux Signature Analysis: an Alternative Method for the Fault Diagnosis of Induction Machines,” accessed online 12/ 01/ 2022. Research 38 Tribologie + Schmierungstechnik · 70. Jahrgang · eOnly Sonderausgabe 1/ 2023 DOI 10.24053/ TuS-2023-0031 [3] Booser, E. Richard, “Tribology Data Handbook: An Excellent Friction, Lubrication, and Wear Resource,” CRC Press, Boca Raton, FL 1997, pg. 900-907. [4] Sheldon, Jeremy S., Watson, Matthew J., Byington, Carl S., “Integrating Oil Health And Vibration Diagnostics For Reliable Wind Turbine Health Predictions,” Proceedings of ASME Turbo Expo 2011 GT2011 June 6-10, 2011, Vancouver, British Columbia, Canada. [5] Sheng S., H. Link, W. LaCava, J. van Dam, B. McNiff, P. Veers, and J. Keller, S. Butterfield and F. Oyague, “Wind Turbine Drivetrain Condition Monitoring During GRC