Summer time is all about R&R; for most that means rest and relaxation. At LNS we like to add a third, being Research. The past week was busy with briefings and strategy sessions; with companies like iBASEt, Schneider Electric, OSIsoft, Bentley, and Parsec among many others.
Across all of these interactions a common theme emerged, Industrial Big Data Analytics. Vendors and end users alike are interested in the promise of Big Data and Predictive Analytics, but there are a number of common themes and questions underpinning this interest:
- What is Big Data Analytics?
- What are the different types of analytics?
- What’s new in the industrial space, we have had metrics, KPIs, and optimization for decades?
- What are the top use cases?
- What are the top challenges?
Over the past several years LNS has been steeped in trying to answer these questions. We have advocated for industrial companies to take a simple definition of Big Data: Volume, Velocity, and Variety. We have also advocated for industrial companies to think of analytics in simple terms: descriptive, diagnostic, predictive, and prescriptive.
Our research in new frameworks has shown how the Industrial Internet of Things (IIoT) intimately connects to Big Data Analytics as part of the IIoT Platform. We have shown with our survey data how industrial companies are looking for Industrial Big Data Analytics to solve quality, productivity, energy, and reliability issues, and the biggest challenges are with creating a business case not using technology. Unfortunately, all of this work may have missed one of the most important and quickly maturing technologies in analytics, Machine Learning (ML).
In previous posts I have tried to get to the heart of the matter by talking about closing the “Data Science Divide.” I've also spoken about how industrial companies are filled with engineers, not data scientists and how engineers like to work with models not algorithms. This means those companies bringing these two roles together most quickly will be able to capture the promise of Industrial Big Data Analytics first.
I think my colleague, Andrew Hughes, gets to the heart of the matter even more quickly. He asks just one simple question: “Can you answer questions you didn’t know to ask?”
The answer almost every company gives to this questions is of course “No.” The problem is, most companies don’t even think to ask this question, never mind figure out a way to answer “Yes."
New Answers to New Questions
In answering this question, we see the real promise of ML. Predictive Analytics is great, but Predictive Analytics alone won’t let your organization answer yes; to do so requires ML learning based analytics.
To help clarify these subtleties and build an approach that will work for your organization, we have analyzed the Big Data Analytics market by simply segmenting the market by Data vs. Big Data and Analytics vs. Machine Learning Analytics.Data – Analytics: Is the starting point for most companies and includes what they have been doing for decades with traditional data sources (time series, transactional, etc.) and analysis tools (visualization, statistical, optimization, etc.). Examples would include scenarios like:
o Measuring OEE based on production equipment state
o Optimizing supply chain performance based on inventory levelsBig Data – Analytics: Is when companies use traditional analytical techniques with Big Data that meets the volume, velocity, variety criteria. Examples would include scenarios like:
o Correlating OEE with customer, supplier, and pricing data to understand the impact of an asset on profitability.
o Correlating IIoT data from smart connected products with manufacturing data to test the accuracy of product reliability models.Data – ML Analytics: Is when companies apply new ML algorithms to old data sets. Examples would include scenarios like:
o Predicting asset failures based on sensor data measuring vibration, corrosion, flow, temperature, etc.
o Predicting product failures from manufacturing process data, even when products meet quality and reliability specifications in processBig Data – ML Analytics: Is when companies apply new ML algorithms to Big Data. This is when companies finally begin answering questions they didn’t know to ask in the first place:
o Companies discover previously unknown relationships between certain materials, assets, production schedules, plants, and workers.
o Companies being to predict asset and product failures based on previously unanticipated relationships.