There are 10 kinds of people in this world — those who understand binary and those who don’t. Let that sink in for a moment.
And there you go.
The binary system, used in the logic circuitry of computers and smartphones, is a base-2 number system. A value is either 0 or 1, on or off, running or not running. We humans have 10 fingers, which presumably is why we traditionally have used a base-10 system of counting. We can attribute a condition or a value to each finger.
[pullquote]If it helps, think of it like the difference between run-to-failure and predictive maintenance. In run-to-failure or reactive maintenance, the system is operating or it’s failed and needs to be fixed. That’s binary. For predictive maintenance, systems can be in a variety of conditions, and the many components within that system must be monitored to predict successfully whether those conditions indicate continued operation or looming failure. Obviously, failure of critical equipment leads to unplanned maintenance, which can be very costly.
I recently spoke with Bob Rice, vice president of engineering for Control Station (www.controlstation.com), at Rockwell Automation Fair in Philadelphia. “Eighteen percent of maintenance is unplanned,” he said. “That’s a $100 billion total maintenance spend in the United States. Unplanned maintenance costs four times more than planned.”
Rice’s company has developed a monitoring and analytical system he calls “clustering” that determines machine health based on historical data.
“Often times, with a single sensor,” he explained, “the setpoint is when the damage is done. Many variables define the condition of a piece of equipment. With predictive analytics or clustering, you have a reference database to compare current operating conditions with the history of multiple variable operating conditions.”
To mitigate future risk and improve reliability, the clustering concept uses historical process data, such as pressures, feed rates, and vibration metrics, to determine what caused a failure. “Clustering is assigning data channels into groups,” explained Rice. “Patterns are generated based on proximity. Data clusters become fingerprints.”
So, we’re back to tracking with our fingers again. That system has seemed to work pretty well so far. About the only thing clustering has in common with a binary system is its open- and closed-book modes. The system is either learning what imminent failure looks like or determining probability of failure based on what it’s already learned.
Open-book mode is learning and can form clusters from data points, explained Rice. In closed book, all conditions or neighborhoods of data points are known, and new data are compared to what is in the book. “In closed-book mode, you see the probability of failure based on data clustered from previous failure states,” he said.
In the end, single-point alarms aren’t enough to avoid unplanned maintenance. “Condition monitoring requires a multi-dimensional solution,” said Rice. We have to think bigger than just the piece of rotating equipment. Rice believes we must look at the conditions of components within the entire system for an accurate assessment of plant knowledge and performance. And I agree, so that makes 10 of us.