Depending on available capabilities and varying goals, there have always been numerous ways to analyze data in the process industries. More recently, digitalization and emerging artificial intelligence (AI) is allowing them to multiply further and extend their reach.
“Data analytics come in a variety of configurations these days. All of the major historian platforms, such as Aveva PI, Aspen IP.21 and Velotic (formerly GE) Proficy, used to be strictly on-premise, but now support hybrid, cloud-computing capabilities of some sort,” says Heath Stephens, PE, digitalization leader at Hargrove Controls & Automation in Mobile, Ala., a division of Hargrove Engineers & Constructors and a certified member of the Control System Integrators Association (CSIA). “AWS and Azure are usually considered cloud-first or cloud centric, but they also offer many on-premise solutions. Even when installed onsite, most of Hargrove’s clients prefer virtual installations to keep hardware and maintenance costs lower and improve reliability. Many clients are backing away from VMware with its new, more expensive licensing model, and looking at Microsoft Hyper-V or even free solutions like Proxmox. Meanwhile, companies like Inductive Automation fully support containerization of its their Ignition SCADA solution that supports many third-party add-ons that provide analytics and other functionality extensions.”
Still need to fit the application
However, whatever the era, the trick is still evaluating and selecting the most appropriate, efficient, cost-effective and durable option. Unfortunately, all of today’s multiplying and digitalized options aren’t making the right choice any simpler, especially for users and organization that are conservative and reluctant to change in the first place.
“The process industries are a relatively slow technology adopters compared to others because of the often physically hazardous applications they run, their lean staffing, limited funding, and various regulations that must be followed,” explains Stephens. “So, when I’m asked what analytics can do now that they couldn’t before, my advice is usually—try installing the technology that’s already been around for 10-20 years in your competitors’ plants before asking me what’s brand new,” says Stephens. “That said, I think the two biggest changes are reduced costs and increasing ease of use. Software companies have realized they’ll only be able to sell so much software to the big petrochemical companies. If they want to increase market share, they have to sell useful software solutions to companies with smaller budgets and fewer resources to implement and maintain those solutions.”
For example, Hargrove recently installed a custom manufacturing execution system (MES) solution for a client, but Stephens reports that even off-the-shelf MES solutions require a lot of customization for particular installations. “Sometimes, custom solutions can be a better fit for them,” he adds. “We’ve also been doing a lot of simulation work with digital twins, mostly RAM modeling with Aspen Fidelis. However, we’ve also had more client interest in solutions from our PLC partners, such as Rockwell Arena and Siemens Tecnomatix.
AI on approach
Stephens agrees with most other observers that AI is emerging, though much of it’s simply rebranded machine learning (ML), while other AI functions require more development.
“I think the next big evolutionary step isn’t quite here yet, and that’s the introduction of AI into analytics software. I say that at the risk of offending some industrial AI software companies, but currently most of these efforts fall into two categories—they’re just classic ML algorithms being labeled AI, or AI tools that show a lot of promise, but aren’t quite there in terms of functionality,” explains Stephens. “I do think we’re really close though. I’ve seen some really interesting previews of soon-to-be-released software from a few vendors, or even software that’s available but has a limited install base so far. This is a space where the story could be very different in just a year or two, so I’m looking forward to these statements being completely dated when someone reads this article in a few years.
In any event, AI is going to have a big impact, and it’s coming. It will make existing tools more intuitive, enable easier on-the-fly data mining and analysis, and create whole new categories of analytics tools. It will also free engineers from some of the more mundane, time-consuming tasks they do now to better use analytics tools. It should be an exciting future.”
To tackle these or similar analytics projects, whether they’re AI-related or not, Stephens suggests users pick the right size and application for each analytics application.
“That means identifying a potentially solvable problem with a calculable revenue impact,” says Stephens. “You want a project that isn’t so big that failure would be unacceptable (these projects sometimes fail), but not so small that the revenue doesn’t justify the investment (or that the impact could be attributed to something other than the project effort). You want something with a high probability of success and impact that will feed into your next bigger bet, either rolling this solution out more widely or moving on to another tool and solution.”
