How machine learning is transforming predictive maintenance

Traditionally, plant managers relied on preventative maintenance schedules provided by a machine’s manufacturer, including regularly replacing machine components based on a suggested timeline. However, these timelines are only estimates of when the machine will require service, and the actual use of the machine can greatly affect the reliability of these estimates.

On one hand, this means that you could be paying for unnecessary maintenance plans and replacement parts that aren’t needed. On the other hand, many things can go wrong between scheduled maintenance visits. For example, if bearings wear prematurely or a motor overheats, a machine may require service sooner than anticipated. Furthermore, if a problem goes undetected for too long, the issue could escalate to further damage the machine and lead to costly unplanned downtime. Predictive maintenance helps avoid these problems, saving time and costs.

A leading solution for predictive maintenance is condition monitoring which plays a key role in predictive maintenance by allowing users to identify critical changes in machine performance. One important condition to monitor is vibration. Machine vibration is often caused by imbalanced, misaligned, loose, or worn parts. As vibration increases, so can damage to the machine. By monitoring motors, pumps, compressors, fans, blowers, and gearboxes for increases in vibration, problems can be detected before they become severe and result in unplanned downtime.

Uniform measurement

Vibration sensors typically measure RMS velocity, which provides the most uniform measurement of vibration over a wide range of machine frequencies and is indicative of overall machine health. Another key data point is temperature change (overheating). Machine learning takes this information and automatically defines a machine’s baseline.

However, collecting machine performance metrics is only the beginning. The ability to accurately interpret and communicate this data is essential for system reliability, and that’s where machine learning comes into play. Condition monitoring with machine learning will recognise the machine’s unique baseline of vibration and temperature levels and automatically set warning and alert thresholds at the appropriate points. This makes the condition monitoring system more reliable and less dependent on error-prone manual calculations.

Wireless condition monitoring is easy to deploy quickly. The ability to monitor machines in inconvenient locations allows for more comprehensive monitoring and increased reliability.

Based on a white paper from Banner Engineering: "Predictive Maintenance Trends: How Machine Learning is Transforming Machine Maintenance”