Measure, reduce, advise—and evolve. In a recent SparkCognition webinar, “From Analysis to Action: AI’s Impact on Triage Processes in Oil & Gas,” Jim Eskew and Andrea Schmidt detail how our Industrial AI enables predictive maintenance for industrial equipment, allowing operators to improve reliability by:
- Measuring and tracking asset health over time
- Reducing unplanned downtime
- Providing insights into the next best action(s) to remedy emerging issues
- Using system feedback to support continuous learning and improve the operational experience
Developed over a decade-plus in close collaboration with some of the world’s largest industrial organizations, SparkCognition Industrial AI Suite helps companies learn from diverse data types to efficiently triage alerts that matter and minimize false positives. Eskew, SparkCognition’s VP of Product Management, and Schmidt, Director of Solution Architecture, reveal how our technology unlocks three categories of benefits—alert calibration, explainability, and continuous learning—that separate our solution from other approaches to predictive maintenance for industrial equipment.
Alert Calibration: Well-timed and well-tuned alerts delivered with all-important lead time to respond proactively come from multivariate AI models aligned with input from subject matter experts—a signature advantage of our Industrial AI platform. Our semi-supervised models are well-proven to catch known harmful events by virtue of threshold-based deviations, but also the ‘unknown unknowns’ that can lead to unplanned downtime: costly disruptions that knock production offline for hours or days without any clear forewarning. They are also adept at mitigating noisy alerting mechanisms that contribute to alert fatigue. Smart alerts and auto-calibration of thresholds are tunable via real-time SME feedback to automatically categorize, benchmark, and compare alerts, reducing false positives.
Explainability: It’s one thing—a very important thing—to be able to apply AI models to configurable data pipelines connected to structured or unstructured datasets to call attention as early as possible to an emerging issue. But what then? What is contributing to those anomalies? How do we isolate progression from the point in time where that anomaly was noticed through to the current point in time by looking at not just trend plots but also histograms? Industrial AI Suite provides an intuitive experience to investigate trends and categorize a timely, appropriate response. It brings contextual intelligence and generative AI tools to compare to previous incidents and decipher what is going wrong and how to address it.
Learning: Analyzing the status of assets as they move forward in time is just as crucial as analyzing how they have performed in the past. Driven by user knowledge and alert scoring, SparkCognition’s predictive maintenance for industrial equipment solution makes it easy to adjust to the ‘new normal’ of operational states as assets age and maintenance practices change. This is also a critical backstop to overcome the challenge of maintaining technical expertise as the workforce ages, making operations knowledge readily available to a new generation of workers.
Replying to an audience question (How does the solution handle asset aging and older assets’ different maintenance needs?) Eskew answered this way:
“What’s working under the hood with these models is they are retraining. So as there is this feedback loop from labeling [… for example, an incident that …] requires immediate action … or deferred maintenance … or is a false positive requiring no action whatsoever; all of these labels are then captured by the machine learning model that is continuously feeding in information and retraining over time.”
“If, for instance, there’s an alert that’s raised a false positive, then that regime or behavior will be sort of walled off as ‘do not alert under this condition going forward.’ All of that is to say that as these assets are performing differently over time—and operators have an opportunity to look at [whether their performance] is a byproduct of how we’re running it currently today or is it just how the asset is naturally changing over time—all of that information is captured within the sensor regime and the model itself. The model is learning it over time so that the noise factor, the number of alerts, doesn’t become exaggerated as the assets start to change. Really, the models change as the assets change over time.“
Adding here a key point about AI models vs. physics-based models, Schmidt said:
“The models that are developed by OEMs are these physics-based models that initially are designed around ideal conditions. The reality is that your equipment doesn’t always operate in ideal conditions. Throughout its lifetime it may be exposed to multiple repairs that have happened, multiple environments, extreme cold, extreme heat, and other environmental events that change the assumption of ideal conditions. The AI-based model is all based on: ‘What are the conditions that are operating right now, considered as normal?”
Providing a comprehensive product demo with illustrative examples, Schmidt and Eskew show how our innovative technology helps your data tell the real-time story of your operational environment so you can get ahead of downtime and optimize production—high ROI outcomes all industrial customers seek but SparkCognition’s customers find. If you’re curious to learn how our Industrial AI Suite’s operator apps determine and contextualize risks to critical equipment, analyze maintenance actions, monitor operational KPIs, interpret asset health over its lifecycle, and help your organization evolve with its changes to create a competitive advantage using predictive maintenance—watch this free webinar.
