The conversation around intelligent manufacturing often centres on artificial intelligence, automation, robotics, IoT and connected machines. But beneath these technologies sits a less visible layer that determines whether a smart factory can actually function: the network.
For manufacturers moving towards increasingly automated and AI-enabled operations, network availability is no longer simply an IT concern. Connectivity now directly influences production continuity, machine monitoring, maintenance, logistics, commercial transactions and the ability to make decisions in real time.
According to Santosh Rai, Global CDIO, PGP Glass, the evolution towards smarter factories is also raising the expectations placed on manufacturing infrastructure. The ultimate ambition is moving beyond Industry 4.0 towards increasingly autonomous or “dark” factories, where machines, data, analytics and AI work together with minimal human intervention.
“Only implementing the Industry 4.0 framework or certain automation is not the end point. The end point is now people are focusing on dark factories.”
That ambition, he says, requires infrastructure capable of supporting the continuous flow of data between machines, systems and AI platforms.
A smart factory needs a smart network
For Rai, network reliability is fundamental to the intelligent factory. “For a dark factory, definitely, you require a reliable and high-speed network because all the devices are running at 5G speeds.”
Manufacturing environments generate enormous volumes of data, but Rai says this does not necessarily mean sending everything directly to the cloud. At PGP Glass, data is processed closer to where it is generated before relevant information is pushed towards cloud platforms for analytics and AI consumption.
“We prefer to make a stepping server or hopping server at a location, and then subject-wise data needs to be pushed to the cloud for analytics, for AI, for further consumption.”
The physical network itself also requires different considerations from a conventional office environment. Wi-Fi can face signal drops and obstacles on the shop floor, while cabling needs to account for electrical interference, physical damage and ongoing civil work.
“There is a cabling standard also. Nobody knows that or is aware, other than the IT guys, that we need to do the cabling very differently on the shop floors.”
The objective is ultimately resilience. Rai says manufacturing networks should be designed with redundancy across connectivity, switches and service providers. “We are not expecting anything less than 100% uptime in the network, whether it’s LAN, WAN or internet.”
At PGP Glass, this includes designing network topologies with alternate routes so traffic can continue if a cable fails. Redundancy also extends to last-mile connectivity and internet providers, reducing the risk of a single point of failure disrupting shop-floor operations.
Data only creates value when it changes a decision
Connectivity alone, however, does not make a factory intelligent.
Rai draws a distinction between collecting data and actually consuming it. “If we just collect the data, it has nothing to do with anything. Because, if you start consuming the data, then only you say whether it’s right or wrong.”
The value of analytics, he argues, is ultimately measured through the speed of the decision cycle. He uses financial reporting as an analogy: a company that closes its accounts on the first of the month has more time to act on the resulting insights than one that closes them on the tenth.
“The insight is coming from the consumption of the data, not from implementation or the collection of the data.”
The same principle applies to manufacturing. Data from machines, sensors and production systems becomes valuable only when it helps teams act faster.
When the network goes down, production can stop
The dependence on connectivity becomes particularly clear in condition-based monitoring.
Rai explains that sensors capture machine information, which is converted into digital records and analytical reports before being consumed by people or AI.
“If the network is down, the whole cycle will collapse and the factory will go into a steady state. The moment your network is down, you are unable to work.”
The impact extends beyond the production line. A network outage can prevent invoices, GST documentation and e-way bills from being processed, potentially stopping material from leaving the facility.
In just-in-time manufacturing, the dependency becomes even more pronounced.
“For the automobile industry, if they are running in JIT, their turnaround time is minimum and everything depends upon the network. If you receive a JIT call, on the basis of this call only, you will discharge the material. Network is down. How are you going to receive the call?”
The network now connects the entire manufacturing stack
Rai describes the manufacturing architecture through the ISA-95 framework, beginning with sensors and moving through SCADA and MES towards ERP, analytics and now AI.
“The first is the sensor. The sensor is connected to the SCADAs. SCADA is connected to the MES.”
Data can then move upwards towards ERP systems for commercial consumption or towards analytical platforms. With AI increasingly embedded into manufacturing, the AI layer sits above these systems and consumes information from across the stack. “Since AI is in boom, all this information passes on to the AI layer for the agent to do all the work.”
The data architecture also determines how insights reach decision-makers. Raw machine data is processed at the edge, organised into data lakes and pipelines, and presented differently to operational teams, plant management and CXOs. The objective is to ensure that each level receives the information it needs to act, rather than exposing everyone to the same volume of data.
This creates a dependency chain in which a network failure at any point can have consequences across the business.
IT/OT convergence creates a new security challenge
The growing connectivity between IT and operational technology is also creating new risks.
Rai points to a common problem in manufacturing: enterprises upgrade computers and networks while legacy PLCs and industrial equipment remain unchanged.
“If the IT network is connected with a device, the whole threat is coming to your mainline commercial network and going to impact your servers, going to impact your commercial transactions. And it will impact your brand.”
His approach is to segregate IT and OT environments to reduce the blast radius of a compromise. “If the infected device is not connected with an IT device, at least our commercial systems, mailing servers, everything will go up. In case of some threat, we only compromise on the production.”
Beyond segregation, Rai says OT environments require dedicated security tools, controls and governance. Where machines cannot be upgraded, additional safeguards can isolate them from the wider network. External devices such as USB drives are also controlled: files intended for machines are first transferred to a PC, scanned and then moved to the machine.
However, he argues that organisations seeking genuinely AI-enabled or “dark” factories also need to address the legacy equipment itself. “If you really want to make it AI-enabled, smart-enabled, you need to upgrade all your devices.”
Security and resilience are not the same
Rai makes a distinction between cybersecurity and network resilience. “If I take outage as a bigger set, security is a subset of the network.”
A security incident may affect a device or production line, whereas a network outage can affect an entire shop floor. “Security is going to impact your brand, but the network is going to impact your efficiency, your physical work.”
That difference also changes response requirements. Network requires instant response.
PGP Glass therefore maintains redundancy across multiple layers, including switches, routers, networks, edge infrastructure and servers. The organisation also avoids depending on a single hyperscaler for critical workloads.
“We have dual hyperscalers. So, at that level, we have everywhere redundancy: switch level, router level, network level, edge level, server level. Everywhere, wherever the device is there, it is redundant.”
At a global level, PGP Glass also uses an internal framework covering cabling, WAN, LAN, security and network architecture, supported by standardised devices and bills of material. This reduces complexity, establishes common skills and standard operating procedures, and enables remote teams to support plants across geographies.
AI is increasing the cost of network failure
As AI becomes embedded into production processes, network performance becomes even more critical.
For Rai, this is why the network cannot be treated as an underlying technical utility anymore. It is a core component of manufacturing performance.
The biggest challenge may be change management
Despite the technology involved, Rai believes one of the hardest parts of modernising manufacturing is not connecting systems but changing how people work.
He describes the challenge as changing people’s mindsets and getting teams to follow standard operating procedures and frameworks consistently. “The major mental block is working in a very secure environment on an SOP or on framework-based working.”
For Rai, discipline is essential to making IT/OT convergence work. “Everybody wants freedom, but in a free environment, the system will not work. The system requires standard procedures and processes, and which require discipline. So, change management is one that definitely is a top priority.”
Ultimately, he believes the measure of an intelligent factory is not how much technology it has deployed, but how quickly that technology helps the organisation make decisions. “If my decision cycle is very fast, everything is digitally implemented and digitally consumed also. If everything is implemented and we are not able to make a decision, this is a waste of money.”
For manufacturers, therefore, the journey towards intelligent operations begins beneath the visible layer of AI and automation. Without a resilient, secure and high-performance network connecting machines, data, systems and people, the intelligence promised by the smart factory cannot translate into operational outcomes.