Peter Zornio is CTO for Emerson, a Fortune 200 global technology and engineering company headquartered in St. Louis, Mo.
In the manufacturing world, analytics is a hot topic right now — and for good reason. Manufacturing generates more data than any other sector, and according to McKinsey, analytics has the potential to deliver more than $4 trillion of growth in industrial manufacturing alone. At the same time, in a proprietary survey my company conducted last year, digital leaders in the industrial sector cited new technology such as artificial intelligence (AI) as the top driver for digital transformation.
But with growing buzz can come confusion. It’s estimated there are hundreds of companies supplying products and services, such as analytics, that are driving the Fourth Industrial Revolution — promising performance improvements utilizing everything from machine learning and computer vision to video content recognition and smart robots.
Recent press and attention surrounding analytics has come from its use in customer buying preferences and marketing. While there are many analytics solutions that can help functions like customer management, human resources or finance, the highest-value opportunity for industrial manufacturers is the analytics surrounding the manufacturing production itself.
This is what we call operational analytics — analytics with the potential to impact and improve the performance of simple equipment, complex assets and process units and entire plants. When effectively applied, operational analytics can improve key performance areas such as reliability, safety, production optimization and energy management. These represent some of the biggest profit levers for a manufacturer’s operations.
Still, it can be difficult for digital leaders looking at analytics across a business to determine how best to deploy analytics to achieve real and measurable results in their operations. The first step to successfully navigate this landscape is to review and prioritize manufacturing pain points and identify those problems that can be tackled with analytics — and what type of analytics are best suited for those situations.
Using models (digital twins) or mathematical algorithms to diagnose problems in manufacturing is not a new concept. In many cases in the industrial world, the physical laws governing behavior, or the relationships between failure cause and effect, are known. Principles-driven analytics are based on known rules, physical laws or principles — and most manufacturing plants and equipment were originally designed using this knowledge.
An example of this in the manufacturing space is a heat exchanger. We know the flow equations, heat transfer rules and other physical laws and principles around heat exchangers, so we can easily model heat exchanger online performance based on these laws. With the right sensors, we can determine if it’s underperforming due to fouling, for example. If we are trying to diagnose a failure condition, knowing how and why equipment fails can quickly diagnose a problem and determine how to repair it.
Coming up with such a failure model is known as failure mode and effects analysis (FMEA). With an FMEA model — and the right sensors — equipment can be continuously monitored, and impending signs of failure can be detected before they occur. And just as importantly, you know the root cause of the failure and how to fix it. For common equipment in an oil refinery, we can predict 80% of failures with FMEA models and the right sensors, according to our company research. This principle has been increasingly applied to vehicles, with a host of sensors that are able to “self-diagnose” a problem.
These models exist for many types of equipment, and we call deploying such models providing “known solutions to known problems.” Solving known problems with known solutions helps you gain a quick return on investment and early wins that can be duplicated across facilities. These small wins also help enhance organizational support from employees — which is critical for any analytics program — and management for continued investment.
Diagnosing and optimizing discrete pieces of equipment also builds a base to start doing the same for complete interconnected manufacturing areas, unit operations or an entire plant; however, new analytical techniques may now be required.
These more complex issues often require AI, machine learning and other advanced data-driven analytics. Data analytics can be as simple as regression models, but they more frequently rely on machine learning, neural networks and other statistical data modeling techniques to identify solutions to problems where the laws or principles are unknown or are too intertwined.
An example here would be using advanced pattern recognition techniques to uncover how a distillation column is interacting with a pump and compressor as an integrated “unit” to optimize performance across all three. Data analysis techniques can sometimes use existing sensor data to spot issues in other areas. One Emerson customer identified a stuck valve after looking at equipment temperature data analytics over a two-year period. If this problem had gone unnoticed, it would have resulted in long-term damage or a costly shutdown.
When trying to drive improvement across an entire plant, or even across multiple plants at the enterprise level, problems are more complex, usually involving systems of systems, with yet-to-be-discovered solutions that rely on data-based analytics tools. But I have noticed that these newer data-based methods are sometimes “oversold” in their capabilities; it’s certainly an attractive proposition to think insights can be gained without additional sensors or a principal understanding of relationships. My advice? If you know these fundamental relationships, put them in your analytics model. Don’t wait for the third time you run out of gas for your data-driven model to figure out your car needs gas to run!
The excitement about analytics is not likely to end any time soon, given it represents the largest big data segment in manufacturing. When adopting analytics technologies, it’s important to remember the value of time-proven methods and understand your goals and objectives before choosing an analytics solution. Starting with the fundamentals and targeting known solutions to known problems is a good first step. These solutions can then be combined with more advanced tools, like AI and machine learning, to help companies tackle more complex challenges and digitally transform their operations in a way that enables them to experience real performance improvements.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?