We can learn a lot about the meaning and value of threshold and anomaly analytics from something we rarely think about until a little light goes off on our car dashboard – engine oil pressure.
Maintaining the correct oil pressure is vital to engine functionality. Oil pressure ensures that rotating components and their bearings don’t touch, minimizing wear and tear. The oil pressure indicator in a car might go off when a car is left idling in the heat and pressure drops, or when pressure rises dramatically because of a filter blockage. This type of problem detection, a threshold detection, can often result in false alarms, as oil pressure naturally dips and rises but doesn’t require an oil change or further intervention. We can improve the error detection by understanding and monitoring the normal behavior of the sensor rather than relying on hard-coded thresholds. This tier of alerts would sense that while the oil pressure is still in the normal accepted range, it is now consistently low, signaling an abnormal operation. This is the next wave of analytics – anomaly detections.
Among current manufacturing service buzzwords, there are few buzzier terms than "data analytics" and "predictive service". Before we explore what threshold and anomaly detection analytics can do for your service organizations, we should explain how data analytics and problem detection can come together to add value to your service business.
Level 1: Static Thresholds and Threshold Alerts
These "starter" analytics begin with a basic dashboard that allows for simple visibility to current conditions. In threshold analysis, a team will set top and bottom condition values, such as high and low engine temperatures. The benefit of threshold detections is that they allow for easy automation of detection. When a condition exceeds the threshold, there is an alert. This works well for data or condition sets that fall into a narrow band of predetermined values. However, for moving multivariate systems, especially large capital assets such as turbines or engines, threshold setting becomes more difficult.
The drawback of threshold detection alone is that this doesn’t take system context into account once the high and low values are set. In addition to this, deciding on those thresholds must be done methodically. Set too wide a value band, and you may miss negative conditions. Set too narrow a value band and you’ll trigger false alarms. False positives and false negatives are equally bad in service contexts. A false positive might cause a manufacturer to needlessly dispatch a technician, incurring costs. A false negative will cause a manufacturer to miss a legitimate cause for concern and result in unscheduled downtime.
In the case of our oil pressure gauge, a high-low threshold alert would indicate a momentary condition that doesn’t require a trip to the mechanic the same way it would indicate a critical oil pressure rise or drop. But, it won’t alert you if your oil pressure is consistently high instead of fluctuating normally.
Join us on Monday for Part 2 of this series, where we discuss the next steps for industrial analytics, and what they mean for manufacturing. Click here for Part 2