Big Data in Manufacturing: Is the Emperor wearing clothes?

Today I attended LNS Research's "Global State of MOM" webinar; quite a bit of good information that I'm still digesting (kudos to Matt Littlefield & company).  There is a bit I find hard to swallow though – Big Data in manufacturing.  Perhaps I'm just so far out of the loop on this that I'm just not comprehending the obvious, but I can't see a real business case for Big Data in the manufacturing environment, nor can I see plants investing in the infrastructure required to support Big Data solutions.

Big Data is (of course) making a big splash in the business press.  In June of 2014, Forbes magazine quoted several sources stating that Big Data analytics, services, and infrastructure will grow at a 30% rate over the next five years – what software, hardware, or integration vendor wouldn't want part of that pie?  But beyond the buzz, I question if there is real business value for Big Data at the plant floor level.

Manufacturing has always produced volumes of data; SPC, batch records, lot traceability, maintenance records, machine down time, material flow, root cause analysis, design of experiments – the list is huge.  But "a great deal of data" isn't the same thing as "Big Data"; you don't need Hadoop and MapReduce with petabytes of storage on multiple servers and ultra-high speed networking to deal with manufacturing data analytics.

Sorting through some of the reasons given for interest in Big Data during the LNS webcast:

• Better forecasts – This has some potential, but the data isn't generated on the plant floor.  This information comes from the marketplace, and the need for better forecasting isn't unique to manufacturing.  Better visibility into customer demand in near real-time should result in better capabilities to collaborate with suppliers, warehouses, and logistics.  But is "Big Data" really the answer for improving forecasting at the plant level?  Maybe I just need to see a good example of this in practice, but until then I remain skeptical.
• Better understand multiple metrics – I suspect this is more related to Enterprise Manufacturing Intelligence (EMI) tools than Big Data, and that there is confusion in the user community distinguishing between analytics and Big Data.  I could be wrong, but I don't think software vendors are doing much to clear up this confusion.
• Service and support customers faster – I would examine the existing business processes first before implementing a Big Data solution here.  I don't believe the lack of actionable information is what's causing service/support issues.
• Real time manufacturing analytics – Again, I think there's confusion between Big Data and analytics.  MES/MOM, Data warehouses, and historians are sufficient for this; does a plant really need a Big Data infrastructure to provide analytical insights?
• Correlate manufacturing and business performance – Honestly, I don't know why this is different from "Better understand multiple metrics" and "Real time manufacturing analytics".  Aren't these things done to ensure correlation of manufacturing and business performance?

At this point, I remain a solid Big Data curmudgeon, hoping someone more enlightened will share their insights.  The folks at A.T. Kearney have stated "While this massive wave of [Big] data promises to transform both top and bottom lines, few organizations have been able to operationalize and monetize this promise for their enterprise."  I believe efforts to integrate Big Data into manufacturing will prove this true.

Business Value and the HMI

This blog post may be sailing into stormy waters, but today I'm putting on my bean-counter hat and challenging conventional wisdom with regards to human-machine interfaces, or HMI for short.  This post is somewhat in response to the LinkedIn article "Look at all them pretty pictures" by Gerhard Greeff of MESA International. (As a side note, if you are in the manufacturing automation industry and are not following Gerhard on LinkedIn, you are the poorer for it!)  But my hope is that we can start a conversation about the true business value of graphical user interfaces in today's manufacturing environment.

Lifecycle Cost

How much are you paying for your HMI?  As an example, there was a new machine build for a discrete unit tray loader; the company had many older versions of this machine, but this would be built with a modern PLC and touch-screen graphical user interface (GUI).  Implementation of the HMI took over 300 man-hours, and training operators, electricians, and mechanics to use the new interface took even more time.  Then came maintenance costs associated with making the complex GUI "more user-friendly".  Using this example, lifecycle cost = initial cost + integration cost + maintenance cost.  (NOTE: Writing the actual control software for the PLC took significantly less time than creating the HMI!)

Business Benefit

What value are you getting for your HMI investment?  One way to approach this is to look at the incremental value provided by the GUI over non-GUI equivalents.  This was fairly easy to do in the unit tray loader example, because there were several implementations of the non-GUI machines to with years of run-time.  The examination proved disappointing from a business perspective; there was no measurable benefit from the GUI-based UI.  In fact, the opposite was true; operators, mechanics, and electricians found the user interface confusing and non-intuitive.  There were even requests to replace the GUI with the older push-button panels found on the other machines.  The expected benefits of the graphical HMI included faster diagnostics when downtime occurred, better insight into machine performance, and easier ways to change machine settings.  These proved to be no better than the non-GUI approaches.  Business benefit: less than zero!

I realize there is a danger of over-generalizing based on a single example, but I do wonder if this truly is an over-generalization.  Manufacturing organizations may have hundreds or thousands of assets on their plant floors, and if no one questions the value of the de facto approach then the drive to implement graphical human-machine interfaces will provide no return on investment and actually decrease the competitiveness of the organization!

A Smarter Approach

The problem isn't inherently one of GUI versus non-GUI; it is the custom nature of the prevailing approach to HMI implementation.  One of the enabling forces of the industrial revolution was standardized, interchangeable parts.  Prior to this concept, every part needed to be custom-made for its application.  Replacing parts meant getting a skilled craftsman involved.  This offers insight into how the custom user interface problem might be addressed.  What if, instead of spending hundreds of man-hours developing a new UI for each machine, there were a standard UI that could be customized in less than a day?  Or even require no customization whatever?  It would provide the same basic operations, independent of machine implementation, so training personnel would be very easy – once they've learned to run one machine, they have the basic skills to run any machine based on the standard UI.  It would not have all the intricacy of a custom UI, but those features are seldom used anyway and could be provided via alternative approaches.  And those alternative approaches may ultimately prove more valuable anyway, because they could involve the use of mobile technology which would lessen the requirements for fixed HMI.  This is exactly the approach ultimately taken by the company in my tray loader example.  While visualization standards are important (and again I'll reference Gerhard's wonderful article), true business value will only be realized when an organization-wide standard UI is implemented.