With Process Data, What’s at the Edge Is the Unseen
The 4th of July is fast approaching which means summer, hot weather, beaches, and sharks. Ironically, most shark attacks in North America happen in the fall, but the important thing is sharks are what many people fear about the beach. They should instead worry about sunscreen, sandals, and a towel, but it’s the extreme edge case that captures their imagination courtesy of Jaws (and maybe Sharknado?). And if you are wondering, the odds are 1 in 3.7M (ish) of dying from a shark attack, but that likely isn’t going to calm the fears of many actually on a beach and staring out at the sea.
I realize shark attacks may be an odd place to start a Seeq blog post, but the focus on the extreme edge case came to mind when comparing what’s important – based on recent customer experiences – versus what lurks at the edge of our imagination.
With process data, what’s at the edge is the unseen, the unknown correlation or relationship impacting process outcomes. If there was only a magic box to find the unfindable, which is a desire expressed in the hype around cognitive computing – machine learning, artificial intelligence, deep learning, etc. The promise is to automatically detect these anomalies without requiring the expertise of process engineers and analysts.
Our reality, as experienced in customer engagements across many process manufacturing verticals, is that the edge case should be the least of anyone’s concerns. What we find is the most basic foundations of analytics should be the source of concern, and not the edges. Three recent customer discussions help illustrate the point.
- A pharmaceutical company wanted to predict when it would need to cleanse/process an asset in their production facility. The data was collected, but analysis with conventional tools managed to defy every effort to extract the desired insight. It turns out the data was downsampled as it was transferred from a plant historian to an enterprise historian, so the required high-fidelity data was simply not accessible. When the data is too sparse, too dirty, or simply missing—no amount of technology can fill the gap.
- An Oil & Gas company brought in a consulting organization to uncover insights using their big data platform, machine learning, and data scientists. They were confident they would be able to find what their existing tools and employees were missing with millions of dollars in benefit. Instead, at the end of a multi-month engagement, the deep insight from the project was that temperature varied. Not a correlation of temperature and outcomes, but that temperature varied over time. That result is as ridiculous as it sounds, but when experts aren’t experts in process manufacturing, they often don’t know how to correlate their findings to yield useful insights.
- A Power Generation company was discussing their asset analytics requirements with us and the causes of unplanned shutdowns. They explained that they understood the relationship between two variables, and how one issue correlated to an outcome in another process. But when pressed, they admitted they didn’t know why this was so, it was simply tribal knowledge based on history. A quick analysis in Seeq showed the relationship between the two variables was in fact random and not correlated, so there was no causation. Why didn’t they know this already? Because the pain of a multi-week analysis in Excel had always been put off because they thought they knew the answer.
Data. Expertise. Tools. These are the basic building blocks of any analytics effort, and before an organization invests in the edge of what they don’t know, it’s probably a good idea to start with the basics. Is the data right? Are the engineers involved and enabled with the right software application tools help them succeed?
This shouldn’t be taken as an attack on machine learning or other cognitive computing solutions that capture our imagination for finding the unfindable insight. Certainly Seeq wouldn’t exist without its machine learning-enabled features, with new capabilities in this area planned for our next release (R18). In fact, next month CEO Steve Sliwa will be writing a blog post about Seeq’s strategy and implementation of machine learning, so we’re as engaged and invested in cognitive computing as any vendor. The difference is that Seeq is using these features to enable what matters formost in process manufacturing: data, engineers, and a modern solution for insight and analytics. The core of the issue, not the edge.
Or put it in the terms of the start of the blog: Seeq’s the sunscreen, not the shark repellant, for your beach visit. Now go enjoy your summer!