Wednesday, November 5, 2014

The emperor's clothes, not fully dressed yet, but not completely naked - Big Data in Manufacturing

By Gerhard Greeff, MES/MOM Proponent and Facilitator, MESA GEP Contributor and Trainer, Chairman - MESA Southern Africa.

This post is in response to a post from Patrick Weber, MBA PMP regarding Big Data. I suggest you read it here http://www.linkedin.com/pulse/article/20141028204312-11146365-big-data-in-manufacturing-is-the-emperor-wearing-clothes? Patrick attempts to look through the hype at the real issues and try to see if business benefit can in fact be derived from Big Data in Manufacturing.


The post asks a number of questions and I had to think hard about how to (or even if I could) answer. I wanted to comment but when I finished it turned out to be too much and I decided to rather submit this as a post.

I agree that there may be some confusion regarding what big data is and the difference between big data and analytics, and the vendors are not really helping. However I see that manufacturing can benefit from "Big Data" tools (such as Hadoop and others) to pull together structured, unstructured and time-stamped data. Manufacturing is also not only plant/s or factories, but the complete value-stream and even the supply chain for larger corporates.

So how do I address the questions/concerns raised by Patrick? I will respond on each of the points he makes:

Better Forecasts - Having actual proven (not theoretical) takt-time information and throughput figures per product per line, with planned maintenance schedules and accurate breakdown reasons and historic data of the time it takes to fix, will enable planners to forecast Available to Promise dates far more accurately. To do this information will be required from the MOM system, the ERP system, the Maintenance system, the plant historians, the quality system and even from the warehouse system. This will also require the analysis of user-entered text/comments normally associated with maintenance systems and difficult to trap and structure for analysis with normal analytical tools. Planners or customer service representatives will also be able to inform the customer if these dates change as result of a breakdown in advance and not only on the day when delivery was supposed to happen. I believe better plant CAPABILITY Forecasting will be possible when using big data tools. In the bigger supply-chain, big data will improve demand forecasting on the plant. Combining the capability and demand forecasting information, planning and scheduling will also be improved.

Better understand multiple metrics - I believe big data IS about analytics. What big data tools do better than normal EMI tools is identify patterns over time. For instance, if metric 1 increase by 10% then metric 2 typically reduce by 2.5%. Unless you build your EMI solution to specifically look for this, you will not see the pattern. Big data tools can also be used to identify relationships between "silo" metrics for instance the relationships between stock-turns, throughput, yield, final quality, maintenance disciplines, product, customer, breakdowns, shifts, personnel and time of year or seasons. EMI tools are good in providing real-time operational intelligence of defined relationships. Big data tools are good at finding relationships between data sets.

Service and support customers faster – In my first point above I talked to this point. Having proven and validated capability information will make it a lot easier to provide accurate available to promise delivery dates and times. Having proven cause/effect information available at all levels will improve the ability of sales/customer service agents to inform their customers (in advance) when things go wrong at plant level.

Real time manufacturing analytics – EMI provides this, but once again only looks at defined data sets. Big data tools for instance will enable "machine learning" within the factory. For instance if a factory complex have 200 pumps of say five different sizes, it will be able to analyse the state of the pumps over time and develop patterns for "healthy" and "imminent breakdown" states. If a breakdown happens it will be able to identify the state prior to breakdown and will be able to associate this state to all the other pumps of the same size in the factory. This will enable the solution to warn of an imminent equipment failure for all pumps well in advance of the actual event. Now combine this warning with accurate capability information and the customer service agent will be able to predict planned maintenance and delayed deliveries for a certain period. This example is a maintenance example but as per the multiple metrics example above, it is about finding patterns and relationships.

Correlate manufacturing and business performance – Patrick is correct in that to get to the correlation, you have to do a number of other things first (or as well). But think about this, we all know that an improved OEE should translate into better business (financial) performance. We know that or it is generally accepted in industry otherwise people would not use that metric. But what is the actual financial impact of a one percent drop in OEE for a specific company? In the silo'ed way we structure, store and analyse manufacturing data, with the volatility of the market-place, stock cycle times etc normal BI and EMI tools will find it more difficult to provide an accurate answer to that question than big data tools.

In saying all of the above, I am not proposing that everyone should run and get a Big Data solution, far from it. What I propose is that as the manufacturing fraternity, we investigate what big data does for companies and evaluate how we can apply the concepts in manufacturing, with or without actual Big Data tools.



No comments: