Author: Jagadish Gattu
Title: CEO, UptimeAI

This article first appeared on Jagadish Gattu’s Linkedin profile.

There are a lot of industrial software solutions on the market that will tell you that you have a problem. For decades, detecting that you have, or will soon have, a problem has been digital technologies’ gold standard to improve plant reliability metrics.

Having spent a decade of my career in this domain, I’ve witnessed first-hand the value that a successful “catch” (avoiding a failure) brings to an industrial plant. But is that enough? The north star in plant reliability has never been knowing you have a problem. Real improvement comes from identifying the root cause and preventing recurrence, not relying on short-term fixes that only postpone repeated failures. In business, you are either improving or you are dying.

You get a high temp reading on a bearing, so you go and grease the bearing… And the temperature comes down… And you probably don’t discover that the temp was high because the contents of the reactor is more viscous than normal… And it’s more viscous because it’s starting to polymerize… And it’s polymerizing because the feed chillers aren’t cooling as well as normal… And they’re not cooling effectively because the chiller tubes are fouled.

Root cause analysis (RCA) tactics like the “5 Whys” exist to address this investigation gap, but are typically applied retroactively, once a reliability event has occurred. While understanding why something has happened in the past is useful, a much greater financial impact will be felt when plant operators can detect, diagnose, and correct the root cause of operating issues in real-time.

Breaking Down Data-Driven Decision-Making

Data-driven industrial organizations have worked tirelessly over the previous decades to minimize the number of manual steps that take place on the journey from data to action. The process generally flows as follows:

  • Data to Information:
    Simple data cleansing, descriptive analytics, and statistical aggregations tell you what happened, what’s actively happening, and provide the mechanism for retroactive reporting. Automation of this step has been common for multiple decades with calculations taking place at the required reporting frequency, as often as near-real-time.

  • Information to Insight:
    Advanced analytics like predictive and diagnostic analytics turn information to insights, telling you what is going to happen, sometimes even why and when it will happen. Insights stop just short of telling you what you should do about it. Automation of this step has become more common in the past decade and occurs on the scale of minutes for many use cases.

  • Insight to Recommendation:
    Historically the rate-limiting step in the data-driven decision-making process, this has been the hardest for technology to replicate.That’s because making an optimal recommendation requires understanding the root cause of the underlying problem. It requires interpreting context from multiple sources, reasoning across data and functional silos and priorities, and identifying the optimal solution. For this reason, automation of this step has been challenging, and the manual task has is left to a small number of long-tenured plant experts.

  • Recommendation to Action:
    Someone (or some system) decides whether to accept or decline a recommendation, then initiates a change. A human operator makes the decision to accept a setpoint recommendation and enters it into the DCS. An Advanced Process Control (APC) or Real-Time Optimization (RTO) system follows rules-based logic and adjusts the set points automatically. This manual or automated step typically occurs on the frequency or minutes to hours.

  • Response Logged as Data. Repeat.

    The response to any type of operational action is automatically captured in the time series data, historized for all of time, and thus the cycle begins again.

The Multi-million Dollar Opportunity

When we started UptimeAI 5 years ago, Industry leaders were telling us, “I already know that I have a problem, and I want to know what the most optimal solution to my problem will be.” We saw an opportunity to leverage advances in analytics, deep learning, and AI to take on the rate-limiting step of getting from insight to recommendation. Today’s AI can reason across data sources, functional silos, and competing priorities to think and make recommendations in a capacity that was previously limited to plant experts.

Every company that we talk to is concerned about the rate at which expertise is leaving their facilities. We’re excited to offer a scalable technology alternative to keep industry running optimally as the workforce evolves.