The application of artificial intelligence to an array of industrial processes has been getting a lot of attention over the past few years. Much of that attention has focused on what could be possible with the technology or test applications run largely in simulations.
With its Asset Performance Management Suite, AVEVA is delivering artificial intelligence (AI) to industrial end users today.
Jim Chappell, vice president of Information Solutions at AVEVA, says there are four Ps in AVEVA’s approach to AI: Predictive, Performance, Prescriptive, and Prognostics. He explains the four Ps as follows:
- Predictive is a machine learning type of anomaly detection used to deliver early warning of problems using big data analytics.
- Performance is another machine learning application that baselines actual performance of equipment in the plant so that performance predictions will match the reality of day-to-day operations, thereby reducing the need for manual tunings.
- Prescriptive AI determines the root causes of problems and then directs users in how to address them. “Following AVEVA’s MaxGrip acquisition earlier this year, we now have more than 900 asset types and 10,000 faults with prescriptive actions,” Chappell says.
- Prognostics uses neural net and deep learning capabilities to predict the future based on what has happened previously. “This can be used to determine asset life, but also for early warnings and schedule optimization,” says Chappell. “For example, this can be used for ‘what-if’ analyses such as: if I lower pressure, can I continue operating without a problem, or can I make it to next planned outage.”
These capabilities help reduce unscheduled downtimes, improve operations and quality, prevent equipment failure, optimize maintenance strategies, reduce costs and risks, increase line and asset utilization, extend equipment life, identify underperforming assets, and improve safety.
Chappell explains that AVEVA is “in the process of infusing AI into many products within all four of the company’s business units—Engineering, Asset Performance Management (APM), Monitoring & Control, and Planning & Operations. In APM, we already have predictive, performance, prescriptive, and prognostics embedded. For example, within PRiSM, we provide automated predictive analytics, root cause analysis with prescriptive actions, and will soon have an early release of remaining useful asset life estimation delivered via prognostics. We also offer the optional integration of ROMeo, our simulation system, into PRiSM where we’re using AI to autotune the ROMeo simulation for use as an anomaly detection system based on first principles analytics. This can be fully integrated into PRiSM’s one-stop-shop alerting system so that people are automatically notified of early detected issues from both systems.”
AVEVA’s AI products work with Wonderware System Platform, InTouch HMI, Citect SCADA, Historian, and the AVEVA Insight cloud platform. They can also connect to any data historian, database, or data lake connected to any control system. AVEVA AI solutions are offered as on-premise applications, in the cloud via subscription, or as hybrid on-premise/cloud applications with “minimal impact to your existing software footprint,” Chappell says.
To begin applying AI to your processes, Chappell says you only the need the historical and real time data that most industrial companies already have. “You start with the historian data, your database, or your data lake,” he says. “Our AI then creates the digital signature of your processes from this data and compares it to incoming real time data to see if any anomalies exist.”
Chappell adds that, without the help of AI, it’s hard to find potential problems in your systems because “it’s difficult to see these kinds of patterns in the raw data. That’s the strength of AI—comparing current to historical and predicted data against alarms to see how things are trending to provide early warning. The system detects abnormal behaviors, prescribes solutions, and helps you understand when to take actions.”
Chappell says that enhanced operations and maintenance are two target applications for early applications of AI in industry that can offer quick returns. “Enhanced Operations provided by predictive technology changes the traditional way of responding to alarms, which has historically been after the fact—when you’ve already incurred the problem. But with predictive technology, you get alerts to fix the problem before it occurs. Likewise, with inspection, complex issues can be difficult to detect manually, but predictive analytics’ root cause analysis highlights exactly what needs to be assessed, minimizing the amount of time spent on inspection. This capability is key to addressing the random failures that comprise 82% of equipment failures. Only 18% of equipment failures are age related. Analytics can provide insight into these random failures to avoid the problems that affect production.”
To help users better understand how AVEVA customers are applying AI today with the APM Suite, Chappell offers several examples, such as:
- Pump & Valve Maintenance — with APM’s predictive analytics, a customer saw that pressure data was deviating from normal and was able to determine that water was getting into oil reservoir and creating too much pressure on pump’s seals. “This helped change maintenance practices on this equipment and also educated the company on how to look for this problem company wide,” Chappell says.
- Another APM user was notified by the software that the current on a conveyor was too high for conditions, indicating that tension on the belt had changed, causing it to pull excessively. Chappell notes that, “left undetected, this could have damaged the conveyor’s bearings or the motor—or the belt itself could have separated. Any of these problems would have resulted in an unplanned outage and likely damage to the equipment; but it was caught early on by APM’s predictive system.”
A system notification that a motor’s amps had increased from 14 to 18 highlighted a potential issue for one user. This increase, viewed on its own, was not necessarily a problem and would not have set off an alarm. However, after looking into the cause based on APM’s predictive alarms, the user found a leak in the floor above the equipment had saturated the insulation and caused expansion in the motor shroud. “This issue had nothing to do with a problem in the motor,” says Chappell, “but provides a good example of the type of insights into your operations these analytics can provide.”
