PRANA and the Quest for Unobvious Relations

The human mind is unable to track the correlations between hundreds of diverse values. Instead, it can develop a prognostic system to help. Learn how PRANA finds concealed relations and reveals equipment defects.

The more information we have, the easier is to find spurious correlations. This is what Tyler Vigen, the author of Spurious Correlations, is passionate about. For example, per capita cheese consumption perfectly correlates with the number of people who died by becoming tangled in their bedsheets:

The correlation is 94.71%! But such a relation is highly improbable.

Another example is per capita consumption of margarine and the divorce rate in Maine:

The correlation is 99.26%, and in this particular case there could be a relation: margarine consumption depends on people's wealth. Maybe the divorce rate also depends on it.

Down to Business!

We, just like Tyler Vigen, love Big Data very much, and process a lot of them. We know by our experience that spurious correlations can reveal concealed relations. Here is an example. 

In one of the turbines connected to PRANA, the level of axial vibration in generator bearing No. 4 increased as the active power decreased. Not only did PRANA predict when the vibration would reach the alarm threshold, but it also helped us to identify the root cause. In fact, the cause was rather unusual.

The vibration varied with the commutator cooling air temperature. The air duct passed through the bearing support. It turned out that as the unit operating conditions changed, both the cooling air and the entire bearing support temperature also changed. 

Cool air volume in the cooling duct vs. vibration

Since the duct was poorly located, any ambient air changes resulted in linear expansion of the studs which fasten [the chair to] the foundation, and decreased rigidity in the bearing support. That is why the vibration intensified!

Orange curve: vibration level. Red curve: cool air. Magenta curve: hot air

The standard monitoring tools and the plant personnel failed to notice the correlations among the thousands of unrelated process variables. However, the PRANA prognostics system was able to find this concealed relation. With early warning, the turbine operator was able to plan the repair and remedy the defect during a scheduled outage. The vibration was reduced more than by a half, and it did not increase again.