What if you could prevent brakedowns?
What if you could reduce unnecessary maintenance?
Problem: In motors and pumps, parts will need regular maintenance. Especially bearings are prone to failure. Maintenance work is usually triggered in one of two ways: Either by running equipment until it breaks or with periodic maintenance. The consequence of running equipment until it breaks will often involve extended downtime, and the whole motor can be damaged instead of just a bearing, resulting in more costly repairs. With periodic maintenance, the equipment is frequently maintained without any real need; this is both inefficient and expensive.
Solution: The factory couldn't reliably pinpoint what the pattern looks like just before motor failure. Instead, the maintenance engineer created an anomaly detection model in intelecy that learned how the motor behaved when it was in a known good state.
As months passed, they could see that the difference in predicted and actual temperature in the bearing was increasing. The factory had already implemented alarming on a high-temperature value. However, in long periods they would run the asset with reduced capacity, and even though the temperature increased, it did not reach the alarm limit. The same is true for the winter months where the surrounding temperature cools the bearing enough to prevent itreaching the hard-coded alarm threshold. The solution was to use machine learning to replace the hard-coded alarm threshold limit with a dynamic threshold that calculates the temperature of the motor in a “good state” and compares it with the actual temperature. When the difference is above 25 degrees, a warning will trigger a work order to schedule maintenance.
Result: Earlier, an incident led to several days of downtime in production as they waited for too long to do maintenance and the whole motor had to be replaced. As a result, periodic maintenance was scheduled frequently, and every time a few hours of production was lost.
Today, using intelecy, they can schedule maintenance when required. Since a machine learning algorithm is monitoring the process, they are not dependent on having humans monitor every motor in the factory.