- A digital twin is a virtual representation of an object or system that spans its lifecycle, using real-time sensor data to monitor its performance and health.
- An overload of alerts from digital twins on assets throughout industrial settings creates alert fatigue that can lead to overlooked maintenance issues and asset failure.
- AI models from SparkCognition learn how to detect valid and relevant alerts, preventing costly machine downtime and burnout from O&M staff.
- The models also incorporate historical text and visual data to help monitor plant health and identify opportunities for operational improvements.
It’s a scene that should spark alarm and concern for anyone involved in the health and safety of a manufacturing site: a control room lit up by alerts and alarms generated by sensors from production machines on the plant floor that are operating outside of their normal parameters.
In theory, the digital twin technology connected to individual assets to monitor their effectiveness should do the job of spurring maintenance and operations teams into action to address the issues and return each system to its normal operating condition. But the reality is that plant staff typically see dozens of low-importance alerts per day that don’t translate into any meaningful impact before abnormal conditions tend to return to normal on their own.
What sets in is the alert fatigue that makes every new signal a judgment call on the part of experienced, but imperfect maintenance and operations managers, creating the opportunity for valid alarms pointing to legitimate issues to get overlooked.
This is what leads to breakdowns and asset failures that bring with them millions of dollars lost in downtime.
In a recent webinar for industrial operators, Curt Richtermeyer, SparkCognition’s executive vice president of global sales, said that the binary of “normal” and “abnormal” favored by digital twin capabilities on individual assets has created a forest-for-the-trees situation where technicians get overwhelmed by individual issues and lack clarity on the overall state of plants and processes.
“You’ll walk into a control room and you see lights flashing on the board, and you ask the operator what it means and they say, ‘It’s nothing, so ignore it.’ Lights flashing everywhere is a dilemma in most environments,” he said.
AI models offer a wide-angle view of plant health
Digital twins can provide value, which is why their use is required as part of warranty coverage for durable assets. But more sophisticated data solutions using artificial intelligence technology offer a next-level look at how all of the pieces in a plant setting work together.
That holistic view can reveal hidden problems, show opportunities to improve maintenance programs and find process efficiencies that reduce overhead costs or improve output and profitability.
The AI models at the foundation SparkCognition Maintenance Advisor use billions of data points from sensors on multiple assets to build a “normal behavior modeling” scenario that’s more sophisticated and technician-friendly than the binary good/bad outputs of digital twin technology.
The models offer a range of normal behavior with alerts made for critical situations and the capability to let the people involved in monitoring plant and asset health provide feedback on the validity of those alerts. That feedback loop helps the models learn and become even better at detecting problems beyond basic temperature and vibration abnormalities.
Customer companies have seen a quick change in their alerts behavior and the value of sensor data, with the number of alerts created per month dropping from thousands with little value down to only a handful that pointed to legitimate issues that needed attention to prevent a failure and costly downtime.
“You can calculate the risk score and see the trend lines of sensors involved, and a heat map of what is contributing to that situation to determine if it needs immediate action or just monitoring,” Richtermeyer said. “You can then begin to trend that and watch because you’re able to predict what’s happening far in advance and trend to see over time if this alert is risky or just a blip that will then return to normal.”
Using SparkCognition’s AI technology to augment both digital twins and condition monitoring systems will extract the most value out of data from sensors, while also pointing to where additional sensors could provide more value. By incorporating historical hard copy written and visual data, the models can reduce organizations’ reliance on collective hive knowledge that is at risk of becoming scarce amid the wave of retirements and job switching taking place amid the Great Resignation.
“Rather than monitoring by looking through a straw, it’s looking through a wide-angle lens. You can pair that information with the records of what steps have helped remedy problems in the past, and collate all the information to build ‘corporate memory’ to stay ahead of equipment,” Richtermeyer said. “You improve the reliability of entire systems… and have a better handle on what’s happening in your plant and how to keep it running at its highest level.”
Learn more about how SparkCognition can provide the best view of plant and asset health.