We have spent the last twenty years making industrial software considerably better at understanding machines. We collect more telemetry, retain more history, build increasingly sophisticated models and apply analytics to identify patterns that would have been almost impossible for an operator to detect manually. More recently, AI has given us another layer through which all of that information can be interpreted. 

Yet the way we present this information to the people actually running industrial operations has changed surprisingly little. 

Walk into a modern control room and you will probably find an impressive collection of screens displaying asset health, alarms, trends, process diagrams, work orders, predictive analytics and perhaps a digital twin. The graphics have improved enormously, but the basic interaction model remains familiar. The system presents information and expects the operator to determine what matters. 

Perhaps the next significant evolution in industrial software should therefore be less concerned with understanding the machine and more concerned with understanding the person operating it. 

From dashboards to cognitive interfaces 

Today, two operators looking at the same asset will generally see roughly the same application. They may have different permissions or personalised dashboards, but the underlying assumption is that information is organised around the system rather than the individual decision being made. 

A Cognitive Operations Interface would reverse that relationship. Instead of presenting a predetermined collection of screens, it would assemble the experience around the operator, their environment and their immediate objective. 

Such a system could understand which asset an operator is looking at and which equipment is physically nearby. It could recognise whether they are following a planned procedure or investigating an unexpected event. It could understand their qualifications, previous experience and decision-making authority, while retaining context about information they have already reviewed or acknowledged. 

The interface could also incorporate what other members of the team have discovered. If an engineer on the previous shift has already investigated a possible cause and ruled it out, there is little value in making the next engineer rediscover exactly the same information. 

More experimentally, the interface could consider cognitive workload. Research into adaptive human-machine interfaces is already exploring eye tracking, behavioural signals and physiological measurements as ways of estimating attention and mental workload. The interesting industrial opportunity is not simply measuring those things, but connecting them with asset condition, operational risk and the decision that needs to be made. 

The design question then changes. Rather than asking how much information we can make available to an operator, the system begins asking what the minimum information is that this particular person needs to make the next safe decision. 

The same machine does not require the same interface 

Consider a relatively inexperienced technician investigating abnormal vibration on a pump. The system knows the asset, understands the symptoms and recognises that the technician has limited experience with this particular failure mode. In that situation, the most useful interface might identify components visually, provide a guided diagnostic sequence, surface the relevant isolation requirements and incorporate safety checks directly into the procedure. 

An experienced rotating-equipment engineer investigating exactly the same pump has a very different requirement. Walking them through a basic diagnostic process is unlikely to improve either speed or decision quality. Instead, the system could compare the current event with previous failures and explain that the vibration signature looks similar to earlier seal problems, but that suction pressure has remained stable this time. Based on that difference, it could suggest examining motor alignment before opening the pump. 

The underlying telemetry has not changed. Neither has the asset. What has changed is the information required by the person making the decision. 

This goes considerably further than personalising a dashboard, this is not about widget placement, customising which windows or apps are available on a per user basis. This is about the UI innately understanding the context of the situation the user finds themself in, stress/complications/first time addressing the situation etc, and dynamically adjusting the content onscreen to suit, adding content where more is required, but similarly removing it where basic simple procedural steps are all that is required and remove the noise of the other information onscreen. 

There is also an organisational benefit here. One of the persistent challenges in industrial operations is transferring experience from highly skilled people to those with less time in the role. Quite the majority of the difference is down to understanding what information matters in any given situation and filtering out everything else, that’s experience that comes with time. A cognitive interface potentially creates a mechanism through which some of that judgement can be captured and made available without attempting to turn every investigation into a rigid procedure. 

Designing for pressure rather than adding to it 

The concept becomes more valuable during abnormal operations. Industrial systems have traditionally responded to serious events by providing operators with more information. Alarms accumulate, indicators change state, notifications arrive and additional people become involved. Exactly when human cognitive capacity is becoming constrained, the technology increases the volume of information competing for attention. 

A cognitive interface could deliberately behave differently. 

Imagine that process conditions are deteriorating, several alarms have appeared in quick succession, and the system knows that a shutdown will become necessary if the condition is not corrected within six minutes. Rather than continuing to display maintenance history, long-term trends, efficiency information and every related alarm, the interface could temporarily reduce itself to the immediate action required: reduce load to 70 per cent, confirm completion and show the remaining safe operating window. 

Once the condition stabilises, the deeper analysis can return. The operator can then investigate why the event occurred, review the preceding telemetry and determine what maintenance intervention may be necessary. 

This is important because good interface design in critical environments is not necessarily about providing more capability. Sometimes the most intelligent thing a system can do is remove information. 

How much of this is possible today? 

Quite a lot of the underlying technology already exists. 

Industrial platforms can identify users and roles through existing identity systems. Competency management can establish qualifications and permissions. Asset-management and maintenance platforms contain equipment history, procedures and previous interventions. Operational systems provide real-time telemetry, while computer vision and spatial technologies can increasingly identify equipment and understand an operator’s physical environment. 

AI adds another important capability because it can interpret information distributed across these systems. Instead of requiring an engineer to separately inspect an alarm history, maintenance record, operating procedure and telemetry trend, an AI system can potentially combine them and explain why the current event differs from previous ones. 

Eye tracking and behavioural analysis are also relatively mature technologies, although their application in operational environments remains less established. Research continues into the use of fixation patterns, pupil behaviour, EEG, heart-rate variability and other signals as indicators of cognitive workload and stress. Combining multiple signals appears particularly promising because any individual measurement can be misleading. 

The individual ingredients, therefore, are not especially futuristic. The challenge is bringing them together reliably enough to influence an operational interface. 

That is where considerably more R&D is required. 

Context is the difficult part 

Take a situation where an operator comes across a new problem, one perhaps they do not have the experience of before. Naturally the pulse races a little as the stress mounts and the pressure is on to resolve the situation as quickly as possible. Being able to detect that someone’s heart rate has increased is easy compared with understanding why it has increased. The operator might be experiencing cognitive overload, or they might simply have climbed several flights of stairs to reach the equipment. An eye-tracking system can identify that someone has repeatedly looked at a particular gauge, but it cannot automatically conclude that the gauge is confusing them. They may simply consider it important. 

An industrial system cannot afford to make casual assumptions from weak signals, particularly when those assumptions influence which information an operator sees. 

The more interesting research problem is therefore context fusion. A useful cognitive interface would need to combine asset state, environmental conditions, task history, operator behaviour, expertise, team activity and potentially physiological indicators before deciding whether the presentation of information should change. 

Even then, adaptation needs boundaries. An interface that constantly rearranges itself could quickly become more frustrating, and potentially more dangerous, than the dashboards it replaces. Operators develop muscle memory and familiarity with systems. During an emergency, knowing where critical information will appear matters. 

The objective should therefore not be an interface that endlessly reinvents itself. A better model would be a stable interaction framework within which information density, guidance and contextual assistance can adapt in predictable ways. 

There are some uncomfortable questions 

The potential benefits are significant. Cognitive overload could be reduced, fault diagnosis could become faster and experienced engineers could receive concise information without forcing less experienced operators to work without appropriate guidance. Knowledge could flow more effectively between shifts and teams, while critical information could be prioritised according to operational context rather than simply alarm severity. 

However, the same technology does introduce a new problem: when an employer can determine where an employee is looking, how quickly they are responding, what they have read and potentially whether they appear stressed or cognitively overloaded, an interface designed to help an operator can very quickly become an extremely sophisticated employee-monitoring system. 

Wherever possible therefore, physiological information should remain on the operator’s device rather than becoming another stream of enterprise data. The wider platform may not need access to heart-rate variability, pupil response or other raw measurements. It may only need to know that the local interface has detected conditions under which reducing information density could be beneficial. 

Transparency matters as well. If an interface changes because it believes an operator is overloaded, the operator should be able to understand that assumption and challenge it. Otherwise, we risk replacing information overload with a system that quietly decides what people are capable of understanding. 

The safety case may be harder than the technology 

There is another problem that deserves considerably more attention than it usually receives in conversations about AI interfaces. If the system decides that a piece of information is not relevant and suppresses it, what happens when that information later turns out to have mattered? 

That question moves the concept from software design into safety engineering. Adaptive interfaces used in critical infrastructure would need clear constraints governing what can be hidden, summarised, prioritised or changed. Certain classes of safety-critical information may always need to remain visible regardless of what an AI model believes about the operator’s workload or intent. 

The distinction between assistance and authority becomes extremely important. AI may be allowed to change how information is presented without being allowed to change the underlying operating envelope. It may recommend what deserves attention without having the authority to decide that a safety-critical condition can be ignored. 

Getting those boundaries right will probably require as much research into human factors, governance and operational assurance as into AI itself. 

Perhaps the interface eventually disappears 

For decades, we have assumed that industrial software requires an application. Applications became dashboards, dashboards became mobile apps and mobile interfaces are increasingly being supplemented by spatial computing, augmented reality and conversational AI. 

The interface could eventually become a temporary layer between the operator and the physical world. Looking at a compressor could reveal its relevant condition. Asking why it is running hot could produce an explanation based on live and historical evidence. Approaching a failed pump could automatically bring its maintenance context into view. Beginning a procedure could change the interface into an instructional tool, while speaking with another engineer could bring their observations into the same operational context. 

During a serious event, much of that information could disappear again, leaving only the immediate actions required to maintain a safe operating state. When assistance is no longer required, the interface could simply get out of the way. 

None of this requires a single breakthrough technology. Most of the individual components already exist in some form. What we do not yet have is sufficient confidence in the orchestration, human factors, privacy boundaries and safety models required to combine them into something we would trust inside a refinery, power station, factory or other critical environment. 

That is why I suspect the next major change in industrial operations software may not come from another analytics platform, another digital twin or an AI assistant attached to an existing dashboard. Those technologies will undoubtedly contribute, but they still largely assume that the human should navigate the information architecture we have created for them. 

A Cognitive Operations Interface starts from a different premise. It assumes that the system should understand enough about the machine, the situation and the person operating it to determine what information is actually useful at that moment. 

If we can make that reliable, safe and appropriately private, we may eventually stop measuring the quality of an industrial interface by how much information it can display. 

We may start measuring it by how much unnecessary information it knows not to show. 

Explore what Industrial AI can do today

  • See how IFS.ai Operational Intelligence is helping organizations turn operational data into deeper intelligence across assets and operations. 
  • Explore IFS Intelligence to see how Industrial AI can bring intelligence and context into enterprise processes, decisions and workflows.