At the Emerson Exchange EMEA 2024 in Düsseldorf, a conference for chemical, life sciences, metals and mining, energy, hydrogen, biofuels, carbon capture and power companies, leading automation experts from around the world came together to discuss “boundless automation” to inform and exchange information. CHEManager editor Volker Oestreich sat together with Peter S. Zornio, Chief Technology Officer of Emerson, to discuss trends in automation, AI and cybersecurity.

CHEManager: What exactly do you mean by "boundless automation"?

Peter Zornio: ‘Boundless automation’ is our term for what we see as the future architecture of manufacturing operations technology (OT) – the intelligent field devices, automation software, optimization software and the compute and data architecture that will unify them together. And we’re looking at all the functional elements of operations; the automation of the processes and equipment of course for cost and production optimization, but also how sensing, data, and applications will come together in an architecture that serves other functional areas as well, such as reliability, safety, and sustainability. We picked the term ‘boundless’ as we wanted to convey the concept of ‘no boundaries’; that data and information across those domains can be easily accessed anywhere, or ‘democratized’ as we call it. Today’s OT systems have lots of boundaries; specific network architectures as defined by the Purdue model, silos of data and applications that are functionally specific — and accessing data in a consistent and usable context across those boundaries can be very difficult. This has proven to be one of the biggest barriers to companies’ digital transformation or Industry 4.0 programs. We have all kinds of new technology like cloud, Edge, AI, new sensors, APL, mobility, and so on available to us to help optimize all of the OT functions and empower manufacturing workers; but the problem of extracting and understanding data from the myriad of individual systems that most manufacturers have — securely — gets in the way of developing the software and applications that would achieve this optimization.

So, we need to envision a broader architecture that enables the democratization of that data – and will enable manufacturers to truly optimize their facilities across all operational goals, not just production but reliability, safety, and sustainability. We see the new edge technologies and the cloud playing a bigger role in that new architecture, and the field becoming more a system all on its own.

What impact will AI have on process automation?

P. Zornio: First, let me remind everyone that, as I like to say, "automation applications were AI before AI was cool". There has been a steadily growing outburst of interest in AI technologies for the last ten years, which has redoubled with the advent of large language models or LLMs and generative AI.  But we have been routinely using numerical forms of AI like neural nets and other models in automation since the mid-1980s, with applications such as model based control and planning and scheduling. Our DeltaV control system has had both APC and neural nets as standard features since its introduction. Machine learning algorithms have been in broad use for equipment diagnosis and prediction since the 1990s. This next generation of ‘foundational’ models that include LLMs is opening new doors, however. I broadly think of the application areas in three buckets – configuration and design of the automation system, that is, autoconfiguration; trouble shooting of automation system, that is, super product assistant; and operations assistant, that is, super operator.  Also, in general, LLMs become a natural way to interface with numerical AI models as well since you can formulate queries using language, which is the natural, native way for humans to interact. For the near future, it is expected that these enhancements will most greatly benefit individuals that are new to industry or in their plants. GenAI with LLM’s puts access to role-required information and capabilities at their fingertips without requiring that they have deep system knowledge. These resources are further stressed as organizations move to centralized operations of multiple plants, increasing the complexity that must be understood.

“Boundless Automation” is our term for what we see as the future architecture of manufacturing operations technology (OT) – the intelligent field devices, automation software, optimization software and the compute and data architecture that will unify them together.”

What specific use cases exist or will soon exist?

P. Zornio: There are many to mention, but likely more immediately:

• Enhanced digital twins with integrated AI capabilities to train models on normal conditions and better identify anomalies.
• More efficient migration using tools like Emerson’s Revamp
• Improved ability to perform historical analysis of plant operations to flag deviations from normal
• Copilots for all software products and any experience that benefits from additional context and guidance to users.  These copilots can combine all the knowledge in manuals, documented customer assistance & service calls, and training material into a ‘super user’ of a product
• Auto generation of graphics from combination of P&IDs and product data
• Auto generation of configuration from P&IDs and Emerson product data

Longer term I think everyone is intrigued by the idea of the ‘super operator’ – an AI system that combines existing models of the plant (first principle and data driven) with plant history – process data, operator logs, alarm logs, maintenance system data, etc. to become the ultimate operator advisory on how to operate the facility; or even become an autonomous operator itself in some areas. This is certainly an interesting possibility, and the core AI technology advancements are getting us closer.

How is cyber risk affected by AI?

P. Zornio: Like many topics related to genAI, there are opportunities and risks related to impacts on cybersecurity. 

One risk is the fact that genAI creates more legitimate threat actors quickly. GenAI lowers the bar an inexperienced hacker to create significant disruption. It also allows hackers to attack more of the exposed surface by automatically generating scripts using the corpus of training data that contains device and system information. Another is that many organizations have understandable concerns about intellectual property. AI providers are proactively providing mechanisms to ensure privacy of sensitive data to address these concerns. This includes private instances of AI cloud services and on-premises models to allow different approaches to AI depending on the use-case.

Can AI have a positive impact on cybersecurity?

P. Zornio: Yes, it already has. AI based pattern recognition is used extensively to identify abnormal activities in network traffic, system access, etc. This alerts the operators of those systems to attempted or underway penetrations and hacks. Another positive angle we see coming with GenAI is using it to create more complete test scenarios quickly with the test generation capabilities. This will result in fewer potential gaps in device and software testing and devices with even greater levels of security. Of course, the entire topic of ‘deep fakes’ and discerning what is real or AI generated is a challenge across the board in many uses of generative AI.