Saving money and man hours with machine learning
- Feb 1, 2016 7:00 am GMT
- 769 views
By Mike Reed
In the power industry, the ability to avoid equipment failure is invaluable. It can save lives, man hours and millions of dollars. Technologies available through the Industrial Internet of Things (IIoT) and big data analytics are enabling utilities to do just that. In particular, predictive analytics software with machine learning capabilities is providing utilities with early warning notification that a piece of equipment is experiencing problems or heading towards failure, days, weeks or months ahead of operational alarms.
Predictive analytics software relies on various modeling and data mining techniques to understand the behavior of an asset during all loading, ambient and operational process conditions. The software utilizes machine learning to create a unique operating profile for the asset. It then compares real-time data against the asset model to detect subtle variances that are often the early warning signs of equipment failure. Rather than spending resources and wasting time searching for issues, or worse, waiting for an important piece of equipment to fail, staff is notified in real time than an asset is not behaving as expected. This early warning leaves more time for analysis, maintenance planning and corrective action.
Take for example, a large power provider in North America. The company uses predictive analytics software for monitoring the health and performance of their transmission and distribution assets, including equipment at more than 3,000 substations. In one instance, the software was able to identify a slight equipment variation that would have otherwise gone unnoticed. Company personnel discovered that when a capacitor bank was energized, the neutral current was abnormally high and the condition was not tripping the real-time operational alarm. Through advanced pattern recognition, the software caught the issue at an early stage, before additional capacitors failed, enabling maintenance staff to take action before a major fault occurred and before the utility’s peak season.
Another power utility with plants in multiple states had estimated $8 million in avoided costs over one year through the use of predictive analytics software. In one significant event, or ‘catch’ as it is often called, plant engineers received an email notification from the predictive analytics software showing that a steam turbine experienced a vibration step change that did not trigger any alarm, nor was detected through standard monitoring practices. With further analysis, plant personnel verified that the condition they had been alerted to indicated a potential loss of mass in the turbine blade path. At the time, the utility’s customers were experiencing extremely cold temperatures due to a polar vortex and could not be without electricity under any circumstance. Under increased observation, the utility was able to continue running the unit, and safely shut it down after the critical peak period.
A subsequent borescope inspection verified missing shroud material and revealed several other segments that were close to liberating. Had this issue not been identified at such an early stage, it could have caused immediate unplanned downtime, loss of generation, possible catastrophic failure and danger to personnel. The early warning notification and staff action resulted in a potential estimated savings of more than $4 million in lost revenue and repair costs with this one catch alone, in addition to maintaining the safety of the operating engineers.
These types of finds or early warnings are not uncommon and span from simple asset malfunctions, to identification of errant sensors, to environmental effects and wide-scale efficiency loss on the grid. Other practical applications include asset health ranking and dissolved gas analysis tracking for transformer monitoring, among many others.
The investment in predictive analytics software can be justified through increased asset utilization and reduced downtime, in addition to being able to identify equipment problems before a major failure causes significant damage. Maintenance costs can be reduced due to better planning; parts can be ordered and shipped without rush and equipment can continue running until a more convenient time to bring it offline. Insights gained through the use of predictive analytics software also help plants assess the risk and potential consequences associated with each monitored asset and can be used to better prioritize capital and operational expenditures.
Not only do utilities control costs by extending equipment life and increasing asset availability, other savings are realized when considering the costs that “could have been,” such as replacement equipment, lost productivity, additional man hours, etc., when a major failure is avoided.
As these real-world examples clearly demonstrate, predictive analytics software transforms data into actionable information that empowers utility staff to prevent equipment failure while improving reliability and performance. These insights equip staff with a deep understanding of how their assets are performing, how they should be performing and the impact these factors have on reliability, safety, efficiency and profitability.