The Path to Analytics Maturity—Step 3: Leveraging Predictive Analytics
This is the third post in our four-part weekly series exploring analytics maturity. If you missed the beginning of the series, catch up by reading Step 1: Monitoring Critical Assets and Step 2: Applying Specific Analytics.
We’ve already reached the third stage of analytics maturity—where the customer shifts from using self-defined analytics to using machine learning techniques that predict future events. Predictive analytics come in different ways, like cluster analysis or model-based similarity. These kinds of analytics learn what normal equipment or process behavior is, then accurately predict future behavior.
This class of analytics leverages multi-variant analysis techniques where very complex relationships in data can be accurately modeled so they predict future states based on any variation of input. For example, load or ambient temperature can be taken into consideration when looking at equipment temperature to understand if the current temperature accurately reflects the modeled temperature based on these conditions.
Predictive analytics can help prevent significant problems. One customer, using similarity-based modeling, was able to identify unexplained vibration coming from the inside of a gas turbine. The customer thought that the problem could be explained away due to excess load and thought that the vibration was well within alarm limits. As a result, the customer didn't think the analytics were identifying a problem in anything but the analytics themselves!
Through continued monitoring and tracking, GE was able to convince the customer to shut down operations and perform more detailed diagnostics on the turbine, analyzing the vibration. In this case, visual inspection was needed via a borascope, and what the customer saw surprised them. One blade inside the turbine was cracked and about to break off (liberate) entirely! The crack and movement of the broken blade was seen by the analytics as excess internal vibration. If the blade had broken, it would have shattered the internals of the turbine—which would have cost them $30+ million in replacement costs and lost production costs.
This highlights the maturity element of using predictive analytics. At this stage, a customer has to be confident enough in the analytics to stop production on what looks like minor problems, and fix these before they become major issues. It is very hard to convince an operator to stop production and take preventative action when the margin of error is perceived to be well within normal limits. The operator must trust the data and analytic results, even when they contradict years of operational experience. At the end of the day, if an operator doesn't have this trust, the value of predictive analytics will be wasted.