Do Chemical Engineers care about Big Data…?

You bet!  The American Institute of Chemical Engineers hosted their first ever track on big data in Austin, TX this week.  I had the pleasure of delivering the kick-off session.  One of the opening statements from my presentation ("Big Data:  What it is, and why you should care") really seemed to hit home with many people that I spoke with afterwards:  Big data isn't a thing, it's a journey.  Lloyd Colegrove confessed later that at that point he almost stood up and said "Amen"!

Conferences like this often bring an update on technology trends and directions.  For me though, the customer use cases and stories are almost always the most interesting part because it's where the rubber really hits the road.  Without that, all you have is cool technology.  So I had a few case studies in my presentation, including:

• Americas Styrenics, which eliminated much of the latency in BI and analytics by implementing SAP on HANA in the cloud.  This eliminated the need to create, operate and maintain a separate data warehouse infrastructure for reporting and analytics.  Without the batch updates normally used to feed a data warehouse, Americas Styrenics was able to close the books faster and consolidate faster at month end.
• The GE business unit that provides a remote monitoring service for the gas turbines it sells for power generation.  A single service center aggregates data from over 1,500 machines in 58 countries to feed predictive models that provide advanced notification of equipment failure.  The result is a 25% reduction in downtime for customers.
• Schwering & Hasse use complex event processing from Software AG to ensure continuous production of high quality copper magnet wire.  With 400 production lines, the company monitors approximately 20,000 data points a second, plant-wide, to help manufacture 87,000 miles of wire each day.

Many other presenters had case studies too.  Keith Holdaway, Advisory Industry Consultant at SAS, included many from the oil and gas industry.  In fact, Keith reinforced something that I've written about before - that improving maintenance is going to be the low hanging fruit for industrial IoT.  Although Keith included examples of using predictive models to enhance exploration and production, reducing unscheduled downtime was a recurring theme.  For instance, SAS helped cut unplanned downtime for electrical submersible pumps (ESPs) in the gulf of Mexico.  These are expensive pumps, with a replacement cost of $20m - and potentially $200m in lost revenue should they fail.  And, the operator had 10,000 ESP's across the Americas.  Unfortunately though, following the manufacturers preventative maintenance schedule didn't work well - there was too much unplanned downtime.  By building a predictive model, SAS was able to provide three months advance notice of an impending failure, allowing timely invention during routine maintenance operations.

I was completely blindsided by my favorite case study though.  Erika McBride and Justin Kauhl presented on the use of text analytics and sentiment analysis at Dow.  Frankly, I was surprised to see this application in the chemical industry as they are usually consumer oriented.  And, as Justin and I agreed later, text analytics seems to be the perennial next big thing.  However, Dow have a couple of applications, both hand-built by Justin in Python.  The first is a fairly classic analysis of consumer sentiment, only with a twist.  Dow sells only to other manufacturers, not directly to consumers - B2B2C if you like.  However, Dow still found it valuable to mine social media to gauge consumer sentiment on its customers products.  With the insights gained, Dow was able to suggest alternate approaches and improvements to its customers that would benefit the ultimate consumer. 

The second application involved mining financial reports and publications for sentiment that could be used to improve internal forecasting and planning.  For this, the team worked with subject matter experts to learn what publications and data sources were trusted sources of knowledge on, for example, the plastics industry.  By mining economic sentiment from a number of sources, Dow was able to reduce the error in forecasting models by about 5%, as well as cut latency so that models could be updated more frequently.

There are many other insights from the AIChE conference to share in later blogs.  But in the meantime, if you'd like a copy of my slides, just ask and I'll send them over.