Manufacturers are being encouraged to embrace connected, intelligent and increasingly autonomous production. Yet much of the world’s manufacturing capacity was installed long before those concepts existed. The challenge is therefore not simply how to build the factory of the future, but how to create it from the factory that already exists.
The factory of the future may already have been built
Walk through almost any long-established manufacturing facility and the reason quickly becomes apparent. A recently installed robotic cell may sit alongside a twenty-year-old CNC machining centre. Further down the line may be equipment controlled by an older PLC, a standalone inspection machine and processes that still depend heavily on manual intervention.
These assets were installed at different times, supplied by different manufacturers and designed around different generations of control and communication technology. Yet they are expected to operate as parts of the same manufacturing system.
At the same time, expectations of that system have changed dramatically. Manufacturers increasingly want real-time production visibility, condition monitoring, automated quality assurance, energy optimisation and better integration between manufacturing and enterprise systems. More advanced applications promise predictive maintenance, adaptive processes and increasingly autonomous decision-making.
Consider a CNC machining centre installed around 2005. Mechanically, it remains sound. It produces conforming parts, performs a necessary operation and potentially has many years of productive life remaining. Digitally however, it belongs to another generation.
Should a productive machine be replaced simply because its digital capability has become obsolete before its mechanical capability? Often the answer will be no. But neither does it follow that every legacy machine should be connected, instrumented and automated simply because the technology exists.
This is the central challenge of brownfield automation. It is less about acquiring new technology than about systems integration: enabling equipment, controls, data and people from different technological generations to work coherently together.
Before automating anything, decide whether you should
The temptation in an automation project is to begin with the technology: sensors, controllers, gateways, networks, analytics or robotics. The better starting point is the asset. For existing equipment, there are fundamentally three choices: retain, retrofit or replace.
An asset should be assessed against its condition, estimated remaining useful life, process capability, criticality, supportability, integration difficulty, safety requirements and the economics of intervention. Retaining an asset unchanged can be entirely legitimate. If a machine performs a simple, stable and non-critical operation, additional connectivity may create little meaningful value.
Retrofitting becomes attractive when the underlying mechanical or process capability remains valuable but better visibility, connectivity or control could materially improve performance. This is where brownfield automation can be particularly powerful: digital capability can sometimes be added without discarding the productive capability already embedded in the asset.
Replacement is appropriate when poor condition, inadequate capability, obsolescence, safety limitations or integration complexity make further investment difficult to justify.
There is also a sustainability dimension. Extending the useful life of mechanically sound equipment can avoid premature asset replacement, while improved monitoring and control can reduce energy consumption, scrap and unplanned losses. But keeping inefficient or increasingly unreliable equipment alive indefinitely is not inherently sustainable either.
The decision remains an engineering and economic one. Our 2005 CNC machine passes that first test. It remains mechanically capable and commercially useful. Retrofitting deserves consideration. The next question is more difficult: how far should the retrofit go?
Automation is not one destination
Brownfield automation becomes easier to reason about when three related concepts are separated. Digitalisation improves the ability to capture, move and use information. Automation allows predetermined actions to occur without direct human intervention. Autonomy goes further by giving a system bounded authority to determine an appropriate action within defined objectives and constraints. They are related, but they are not interchangeable. For an existing manufacturing asset, a useful progression is:
- Observe
- Connect
- Understand
- Control
- Optimise
- Autonomise
Importantly, this should not be treated as a maturity ladder that every machine is expected to climb. For one asset, observation and connectivity may capture almost all the available value. Another may justify closed-loop control or optimisation. Only particular applications may benefit sufficiently from greater autonomy. The objective is not maximum automation. It is optimum automation.
From observation to autonomy
Observe – make the invisible visible
Before a system can make better decisions, it must reliably understand what is happening. The first step should be to use information that already exists. Machine states, alarms, cycle counts and process variables may already be available from the controller even if they have never been exposed to a wider manufacturing system.
Where important information is unavailable, additional sensing can be introduced. Temperature, vibration, electrical current, pressure, position or acoustic measurements can provide useful insight without necessarily requiring substantial modification to the underlying machine. Our CNC machining centre example might gain spindle vibration and temperature monitoring alongside information extracted from its existing controller. Motor current could provide additional insight into operating conditions.
Operators and maintenance technicians are important here. A technician who has maintained a machine for fifteen years may recognise a sound, vibration or operating pattern that reliably precedes failure. Brownfield automation provides an opportunity to convert some of that tacit knowledge into measurable information. But instrumentation should have a purpose. The falling cost of sensing does not mean that every measurable variable should be collected. The engineering question should be: what information would allow a better decision to be made?
Connect – get the information where it is needed
Observation creates data. Connectivity makes it usable elsewhere. This is where the reality of brownfield manufacturing becomes particularly apparent. Equipment of different generations may use different communications technologies, proprietary interfaces or, in some cases, provide almost no useful external interface at all.
Protocols and technologies such as Modbus, PROFIBUS/PROFINET, EtherNet/IP, OPC UA and MQTT may therefore coexist within the same manufacturing environment. Gateways and edge devices can provide bridges between equipment that was never designed to participate in a common architecture.
The engineering challenge is not to make every machine communicate in exactly the same way. It is to create reliable interfaces through which the information required by the wider manufacturing system can flow. Nor should connectivity mean sending every available data point to a central or cloud platform. Some decisions must remain close to the process because latency, reliability or safety demands it. Other information can appropriately be aggregated for longer-term analysis.
Connectivity is about getting the right information to the right decision point at the right time. It also changes the risk profile. A machine that operated for twenty years in relative isolation may now communicate across an operational technology network and potentially with higher-level systems. Cybersecurity therefore becomes part of the engineering design, not an IT consideration to be added later.
Understand – turn data into manufacturing information
Connectivity alone does not create understanding. A value of 73.4 has little meaning unless the system also knows what was measured, where it was measured, in what units, under what operating conditions and at what time. This contextualisation becomes increasingly important as information moves beyond the individual machine.
Standards such as ISA-95/IEC 62264 address this broader integration challenge by providing common models and terminology for information exchanged between manufacturing operations and enterprise systems. The principle is more important here than the particular standard: systems need a common understanding of the information being exchanged.
For our CNC machine, vibration data becomes much more useful when associated with spindle speed, machine state, cutting operation, tool and workpiece conditions. What appears abnormal under one operating condition may be entirely normal under another. Connected data is not necessarily understood data.
Control – cross the boundary from knowing to acting
There is an important engineering boundary between observing a machine and changing its behaviour. A condition-monitoring system might initially display an alert. The next level could automatically create a maintenance notification. A more advanced system might alter an operating parameter or initiate a controlled stop.
Each step gives the automation system greater authority, and increases the consequence of an incorrect decision. This introduces questions of validation, interlocks, communications failure, fail-safe behaviour, manual intervention and, where applicable, functional safety. A retrofit should not inadvertently compromise the protections already engineered into the machine.
A useful question at this stage is deceptively simple: What happens when the automation does not work? If a sensor fails, communications are lost or an external system becomes unavailable, the expected behaviour must be understood. As system authority increases, so must engineering assurance.
Optimise – move from reaction to improvement
Once reliable, contextualised information is available, manufacturers can move beyond simply observing events and begin using data to improve how assets operate. Condition-based and predictive maintenance are obvious examples. Process parameters can be analysed against quality outcomes. Energy consumption can be related to production conditions. Repeated minor stops may reveal constraints that conventional performance reporting misses.
For the CNC machine, a combination of spindle vibration, temperature, electrical current and operating context could reveal developing degradation before a conventional alarm threshold is reached.
This is also where machine learning can add value. Patterns across multiple variables may reveal relationships that fixed thresholds cannot easily capture. But sophistication should not be confused with value. Predicting the failure of an inexpensive component that can be replaced in minutes and fails without significant consequence may offer little benefit. A technically impressive predictive model can still be a poor engineering investment. In this instance, the question is not can it be done?, but is it worth doing?
Autonomise – decide how much authority to delegate
Automation and autonomy are not the same. At its simplest, conventional automation might specify: if X occurs, do Y. An autonomous system is given an objective, information and boundaries within which it can determine an appropriate response.
In a more advanced version of our CNC example, the manufacturing system might recognise deteriorating equipment condition and consider estimated remaining useful life alongside production demand, maintenance availability and alternative capacity. Rather than simply generating an alarm, it could recommend – or within tightly defined boundaries execute – a different operating strategy.
Artificial intelligence may support such decisions, but it belongs within the engineering architecture rather than above it. AI cannot compensate for unreliable sensing, poorly contextualised data, unstable processes or weak system integration. Those foundations must exist first.
For many brownfield assets, full autonomy will not be the appropriate destination. That is not a failure of technological ambition; it may simply be good engineering judgement.
The further you automate, the more you inherit
Every layer added to a legacy asset becomes another layer that must eventually be supported. Sensors fail. Networks change. Gateways become obsolete. Software requires updates. Cybersecurity vulnerabilities emerge. Suppliers withdraw support. Custom interfaces that made perfect sense to the original project team can become mysterious dependencies ten years later.
A retrofit intended to eliminate one form of obsolescence can therefore create another. This is why lifecycle supportability must be considered alongside initial functionality. A useful question is: Are we modernising the asset, or simply adding another technological layer that somebody will have to support for the next fifteen years?
Designing for failure matters too. If the new monitoring platform becomes unavailable, can production continue safely? If a network connection is lost, does the machine fail unnecessarily? Can operators revert to an appropriate mode? Is sufficient knowledge retained within the organisation to diagnose the system several years after commissioning? These considerations become more important as automation moves from observation towards control and autonomy.
Across the entire pathway (Observe, Connect, Understand, Control, Optimise and Autonomise) five disciplines therefore remain constant: safety, cybersecurity, reliability, economics and lifecycle supportability. Greater capability should not come at the expense of engineering resilience.
Technology does not eliminate the human system
Legacy factories contain more than legacy equipment. They also contain accumulated human knowledge. Operators and maintenance teams often understand behaviours, workarounds and failure signatures that are poorly documented, or not documented at all. Ignoring that knowledge during automation risks replacing practical understanding with technically elegant but incomplete models of how a process actually behaves.
The role of people may also change as automation increases. Routine intervention can decrease while diagnostics, exception handling, maintenance and process understanding become more important. Paradoxically, as automation becomes better at dealing with normal conditions, human expertise may become increasingly concentrated on the abnormal conditions the system could not anticipate. Effective brownfield automation therefore needs to integrate people as deliberately as it integrates machines.
Knowing when to stop
Return to the 2005 CNC machining centre. It may now have reliable condition monitoring, automated machine-state capture, contextualised production information and selected automated responses. Its maintenance team may have better warning of deterioration, production leaders may have better visibility of its performance, and its useful life may have been extended without wholesale replacement.
Does it also need autonomous decision-making? Maybe, or maybe not. Stopping at optimisation rather than autonomy does not make the retrofit incomplete. If additional automation adds more complexity, cost and risk than operational value, stopping is exactly the right engineering decision.
That principle applies beyond an individual machine. Rather than asking, “How can this factory become fully automated?”, manufacturing leaders should ask:
- What problem are we actually trying to solve?
- What useful capability already exists within the asset?
- Should we retain, retrofit or replace it?
- How far along the automation pathway does the operational and economic case genuinely justify going?
- What happens when the new technology fails?
The factory of the future may actually already have been built. The engineering challenge is not to make every machine as automated as technically possible, but to determine how much visibility, connectivity, intelligence and control each asset actually needs.
For brownfield manufacturing, the objective should not be maximum automation. It should be optimum automation.
Want to learn more about automation in manufacturing?
To explore these topics further and hear from industry experts, register for the upcoming IET Manufacturing Technical Network event focused on automation in manufacturing.
- Name: Automation & the Future of Manufacturing: A 25-Year Outlook
- Description: As automation technologies evolve at breakneck speed, the manufacturing sector stands on the brink of a transformative era. This forward-looking webinar explores how robotics, AI, machine learning, and smart systems will reshape manufacturing over the next quarter-century – from factory floors to global supply chains.
- Date: 30th September 2026
- Time: 2:30PM Local time (BST/CET/UTC +1 hours)
- Registration: https://localevents.theiet.org/baae7e
