Condition monitoring is a maintenance strategy that uses real-time data to track the health of machines and detect faults before they cause breakdowns. Rather than waiting for equipment to fail or replacing parts on a fixed schedule, condition monitoring continuously measures physical indicators like vibration, temperature, and oil quality to catch problems early. The goal is straightforward: minimize unplanned downtime and extend the useful life of your equipment.
How Condition Monitoring Works
At its core, condition monitoring involves attaching sensors to machinery, collecting data from those sensors on an ongoing basis, and comparing that data against known baselines. When a measurement drifts outside its normal range, it triggers an alert so maintenance teams can investigate before the machine fails.
The international standard ISO 17359 lays out a general framework for this process. It starts with auditing your equipment to determine which machines are most critical, then selecting appropriate monitoring methods, establishing baseline measurements, setting alert thresholds, and using the collected data for diagnosis. The final step is deciding what maintenance action to take and reviewing the results over time. In practice, this means condition monitoring isn’t just slapping a sensor on a motor. It’s a structured approach that ties sensor data to actual maintenance decisions.
Vibration Analysis
Vibration analysis is the most widely used condition monitoring technique, particularly for rotating equipment like motors, pumps, compressors, and fans. Every machine vibrates during normal operation, but the pattern of that vibration changes when something goes wrong. The three core parameters measured are acceleration, velocity, and displacement.
Acceleration is the most sensitive to changes in machine condition, making it the best choice for catching problems early. Velocity is useful for identifying resonance conditions, where a machine vibrates excessively at certain speeds. Displacement measures how far a component physically moves during vibration, which can indicate looseness or damage.
What makes vibration analysis powerful is its diagnostic specificity. Different faults produce different vibration signatures. Misalignment between connected shafts creates a distinct pattern. Imbalance in a rotating part shows up at the fundamental rotation frequency. Bearing defects, slipping belts, and fluid turbulence each produce characteristic peaks at frequencies below the normal rotation speed. By reading the vibration spectrum, a trained analyst (or an automated system) can pinpoint not just that something is wrong, but what is wrong and where.
Thermal Monitoring
Infrared thermography uses thermal cameras to detect abnormal heat patterns on the surface of machines. Electrical connections that are corroding or loosening generate excess heat. Bearings running without adequate lubrication get hotter than their surroundings. Blockages in cooling systems show up as unexpected hot spots. Because temperature changes often precede mechanical failure, thermal imaging can catch issues that vibration analysis might miss, especially in electrical systems and static components that don’t vibrate in a meaningful way.
The technique is non-contact and can scan large areas quickly, which makes it practical for routine inspections of switchgear, control panels, and heat exchangers alongside rotating equipment.
Oil and Fluid Analysis
For machines with lubricated moving parts, the oil itself carries a detailed record of what’s happening inside. Oil analysis examines two things: the condition of the lubricant and the wear particles suspended in it.
Viscosity is the single most important quality indicator for lubricating oil. When viscosity drifts too far from its rated value, the oil can no longer form an adequate protective film between metal surfaces, accelerating wear. Chemical analysis can also reveal contamination from water, fuel, or coolant that has leaked into the oil system.
The wear particles are equally informative. Ferrography, a technique that separates metallic particles from oil samples using a magnetic field, can identify the type of wear occurring inside a machine. Large, rough particles suggest severe abrasive wear. Flat, smooth particles point to sliding contact between surfaces. The size, shape, and chemical composition of these particles, often in the 50 to 100 micrometer range, tell maintenance teams which internal components are deteriorating and how quickly. A machine typically progresses through a run-in period, a stable wear period, and then a severe wear period, and tracking particle trends over time reveals exactly where in that lifecycle a machine sits.
Acoustic Emission Monitoring
Acoustic emission monitoring detects high-frequency sound waves, typically between 20 kHz and 2 MHz, that are generated when materials undergo stress. These ultrasonic waves are produced by events like microscopic crack formation, crack growth, localized plastic deformation, and friction between surfaces. Because these frequencies are far above the range of human hearing, specialized sensors are needed.
This technique is particularly useful for detecting very early stage faults, like a hairline crack in a gear tooth or the first signs of bearing surface damage, before they produce noticeable vibration changes. It’s also applied to pressure vessels, pipelines, and structural components where crack propagation is a primary concern.
Condition Monitoring vs. Predictive Maintenance
These two terms are closely related but not identical, and the distinction matters. Traditional condition monitoring measures what is happening right now. When vibration levels cross a threshold, the alert means a fault has already developed or is actively taking place. In maintenance circles, this is sometimes called “condition-based reactive maintenance” because you’re responding to a problem that already exists, even if you caught it before it caused a breakdown. Overall vibration level is a lagging indicator.
Predictive maintenance takes the same sensor data and uses it to forecast failures 60 or 90 days in advance. The difference is in the analysis. By combining standard vibration measurements with high-frequency ultrasonic data and trend analysis, predictive systems can identify lubrication defects, bearing degradation, cavitation, and seal failures well before they would trigger a traditional condition monitoring alert. This gives operations teams enough lead time to plan and schedule repairs around production needs rather than scrambling to respond.
In practice, most organizations start with condition monitoring and evolve toward predictive maintenance as they build up historical data and analytical capability.
The Role of AI and Machine Learning
Modern condition monitoring systems increasingly rely on machine learning to automate fault detection. Rather than depending entirely on human analysts to interpret vibration spectra or oil reports, these systems train algorithms on historical sensor data to recognize patterns associated with specific fault conditions.
One approach that has shown strong results involves converting raw sensor signals into time-frequency images called scalograms, then feeding those images into a convolutional neural network, the same type of AI architecture used in image recognition. The network learns to classify equipment condition, distinguishing a new component from a worn one, for example, with high accuracy. This kind of automated monitoring is especially valuable in high-volume manufacturing environments where hundreds or thousands of machines need continuous oversight and human review of every data stream isn’t feasible.
The practical benefit is speed and consistency. An AI system evaluates every data point against the same criteria, doesn’t get fatigued during a night shift, and can flag subtle trends that a human analyst reviewing weekly reports might overlook.
Which Machines Benefit Most
Condition monitoring delivers the biggest return on rotating assets: pumps, fans, compressors, mixers, agitators, and conveyors. These machines have bearings, shafts, and moving parts that wear in predictable ways and respond well to vibration and oil analysis. But the approach applies broadly to any equipment where an unexpected failure would be costly, dangerous, or disruptive to production.
The decision of which machines to monitor and which techniques to use comes down to a cost-benefit analysis. A critical pump that would shut down an entire production line justifies continuous online monitoring with multiple sensor types. A small auxiliary fan might only warrant periodic handheld vibration checks. Matching the monitoring investment to the consequences of failure is what separates an effective program from one that generates data nobody acts on.