Predictive maintenance (PdM) is trending as organizations implement industrial IoT (IIOT) systems for production control. Yet, implementing cost-effective and operationally efficient PdM programs requires thought to provide a return on investment. Additionally, PdM should not be considered a new maintenance strategy.
When targeted at critical equipment, PdM contributes to a layered maintenance technique, offering tools to complement run-to-failure, condition-based, and preventative tactics. Given the cost of adding PdM to your maintenance arsenal, using appropriate data sources is key to achieving expected cost savings and productivity.
Main PdM Data Sources
Most effective PdM programs consist of limited sensor inputs coupled with in-service operational data. The most common sensor used is for vibration, yet the effective use of any sensor requires understanding its strengths and limitations. Other monitoring and diagnostic techniques may provide a nuanced understanding to aid predictions. Understanding which data sources to use is critical to an optimized PdM program.
Vibration sensors have two constraints. The first is the use of microprocessor-based single-channel steady-state data collectors; the second is the limited applications in which the sensors are used.
Most microprocessor-based instruments capture single-channel data despite many sensors incorporating a second channel for input from a tachometer. This limitation prevents their use on variable speed equipment or more complex applications as the data acquisition assumes a constant vibration profile.
The single-state data collection creates an inability to capture important information such as load or speed modifications. Also, filters remove transient signals or impact data, despite such signals providing useful clues to abnormalities and potential failure.
These constraints aside, vibration sensors are often used solely for rotating equipment, even though their application extends to more complex machines. In “An Introduction to Predictive Maintenance“, R. Mobley gives an example of using a vibration sensor to track transients on a hydraulic ram by using the sensor time-domain function. Analysis can then identify leaking seals or scored cylinder walls.
Ultrasonics and vibration analysis both analyze noise but are complementary due to operating at different frequencies. Vibration analysis occurs from 1 Hertz (Hz) to 30,000 Hz, while ultrasonic analysis occurs at noise frequencies above 30,000 Hz to 1 MHz.
It’s not unusual to find companies using ultrasonics to monitor bearings where vibration monitoring is the norm. This decision may be due to the lower cost of ultrasonic sensors. Yet, ultrasonics are less suited to the task, with bearing or machine noise either operating below the monitored frequencies or compromised by similar frequencies from other sources. No meaningful input to predictive analysis will occur and some experts do not recommend their use on bearings.
Ultrasonics are ideally suited for high-frequency noise such as pressure or vacuum leaks, high ambient noise levels, or expansion or contraction of a medium passing through an orifice.
Thermographic systems monitor the emission of infrared energy from an object and capture the emitted, reflected, and transmitted energy. The reflected and transmitted energy must be filtered from the raw data to be useful for PdM.
There are several thermographic options, but the two useful for PdM are spot radiometers and infrared imaging. The spot radiometer monitors a single point on a machine, such as a gearbox bearing. When used with other monitoring techniques such as vibration or tribology, it allows the correlation of trends.
Infrared imaging is used on more complex equipment to scan the entire machine. The technique depends on the storage and recall of historical images for interpretation by a trained and competent analyst. Requiring more qualitative than quantitative skills, the trained analyst must compensate for the variables that alter the image at each scan.
Correctly applied, thermography can alert to electronic, electrical, and mechanical equipment problems, including trends in processes that transfer or retain heat.
Tribology is the study and application of friction, lubrication, and wear. A complex and comprehensive science, wear particle analysis and lubricating oil analysis are useful for PdM programs.
Both techniques begin with a sample of lubricating oil from in-service equipment, with wear particle analysis providing information on the current wear state of the machinery. Wear particles are analyzed using spectrographic and ferrographic analysis and tracked over time to understand wear rates and isolate possible failure modes.
In contrast, lubricating oil analysis determines the oil’s suitability for continued use, predicting a suitable change interval. The cost of premature oil changes is considerable when measured plant-wide. Conversely, leaving oil unchanged when its lubrication properties have degraded accelerates equipment wear and wastes energy. The analysis can also pinpoint lubricants more suitable for the application.
As to limitations, tribology requires the use of third-party laboratories, which imposes an ongoing cost to have each oil sample analyzed. Also, the sampling regime can threaten the value provided by the tests. Sampling after filtration or from a reservoir may not indicate the oil’s true condition, with particulates filtered or settled out, while poor sampling hygiene can cause contaminated data.
Managing the limitations and tribology is an excellent condition monitoring and preventative maintenance technique. The data from such analysis assists operational decisions and, when overlaid with other PdM monitoring, allows a comprehensive understanding of trends in machine health and possible failure modes.
Data Sources To Avoid
The challenge in choosing PdM sensors and data sources is the number of choices available, making data relevancy your primary consideration. We’ve spoken of not using ultrasonics for bearing monitoring, as it doesn’t provide actionable data. It’s not relevant. Similarly, a thermographic line scanner for temperature monitoring is less useful for PdM than spot or imaging systems.
Yet, relevancy is also applicable to data sources and their structures. If you monitor rotating machinery secondary to production continuity, maintenance costs inflate without tangible returns on investment. Using a failure mode effect and criticality analysis (FMECA) at the plant level and then for each identified piece of critical equipment, identifies and ranks systems and subsystems critical to safety or production. By identifying areas essential for monitoring, you rapidly narrow your choices of relevant data sources and appropriate sensors.
Implementing PdM on essential equipment promises enhanced reliability and reduced maintenance costs. However, it does not replace traditional time-based or condition-based maintenance tasks; it simply adds additional layers of analysis and prediction. Understanding your critical equipment failure modes and the most appropriate sensors and data sources is crucial to achieving your expected returns.
|Bryan Christiansen is the founder and CEO of Limble CMMS. Limble is a modern, easy-to-use mobile CMMS software that takes the stress and chaos out of maintenance by helping managers organize, automate, and streamline their maintenance operations.|
- Implementing predictive maintenance without machine-learning skills
- Choosing the most suitable predictive maintenance sensor
- Open-source kit supports predictive maintenance
- Using IIoT for predictive maintenance isn’t a panacea
For more Embedded, subscribe to Embedded’s weekly email newsletter.