The Case for Industrial Edge Computing: Why Processing Data Next to the Machine Changes Everything

Industrial edge computing is often presented as a story about artificial intelligence. That is only one part of it. The more important change is that industrial sites can now process, translate, store and act on data beside the equipment producing it, without sending every reading to a distant cloud platform or relying on a conventional industrial PC.

For decades, industrial automation followed a relatively clear hierarchy. Sensors and machines reported to a programmable logic controller. The PLC fed a supervisory control and data acquisition system. Operators watched the SCADA screens, while business systems remained several steps removed from the machinery itself.

That architecture has not disappeared. Nor should it. PLCs remain extremely good at deterministic control, safety-critical logic and operating machinery predictably for years.

What has changed is the amount of data surrounding the control process and the number of places that data is expected to go.

A modern industrial installation may need to extract registers from a legacy PLC, combine them with vibration sensors, translate the result into MQTT, store a local copy, send selected values to a cloud platform, show a local dashboard, alert an engineer, maintain a secure remote connection and perhaps run a machine-learning model over the data.

Trying to force every one of those jobs into the PLC is rarely sensible. Sending everything directly to the cloud can be equally problematic. The industrial edge computer has emerged to occupy the space between them.

What Is Industrial Edge Computing?

Edge computing means processing data close to where that data is generated. In an industrial setting, the edge might be a factory cabinet, pumping station, roadside enclosure, substation, renewable energy site, warehouse, vehicle or remote machine.

The important word is not merely computing. It is close.

Instead of every sensor value travelling through a wide-area network to a central cloud before anything useful can happen, an edge computer can handle part of the work on site. It might convert a protocol, remove duplicate readings, calculate averages, detect an abnormal condition, store data temporarily or trigger an immediate local response.

This arrangement can reduce latency, conserve bandwidth and allow the installation to continue operating when its external connection is unavailable. It can also simplify integration by providing one managed point at which old industrial protocols meet modern software.

Why the Industrial Edge Is Growing Now

The underlying idea is not entirely new. Industrial PCs, data loggers and protocol gateways have existed for years. What is new is the combination of affordable processing power, embedded Linux, containerised software, integrated cellular connectivity and compact hardware designed for unattended operation.

Several separate trends have converged.

Machines are producing more data

A temperature transmitter might produce a single reading every few minutes. A vibration sensor can produce a much richer stream. A camera produces vastly more data again. Sending every raw measurement to the cloud can consume unnecessary bandwidth and leave the application dependent on an uninterrupted external connection.

An edge computer can turn raw data into useful information before transmission. Rather than uploading an uninterrupted video stream, for example, it might send only an event, image crop, vehicle registration or defect classification.

Industrial systems need to talk to modern software

Much of industry still depends on Modbus RTU, Modbus TCP, serial communications, CAN bus and proprietary PLC interfaces. Cloud platforms and enterprise applications are more likely to expect MQTT, HTTPS, REST APIs, JSON or OPC UA.

The gap between those worlds is not disappearing. Edge computing gives integrators somewhere to bridge it without rewriting the machinery.

Remote installations are becoming normal

Water assets, solar farms, roadside systems, temporary infrastructure, environmental monitoring and distributed machinery may not have access to a corporate Ethernet network. They still need secure communications, software updates, diagnostics and data transfer.

This is where manufacturers with a cellular networking background become particularly relevant.

Software deployment has changed

Applications no longer need to be installed as one large, fragile software stack. Containers allow individual services to be packaged, deployed and updated separately. A gateway could run a Node-RED flow, an MQTT broker, a local database and a custom Python application as distinct workloads.

That makes an industrial edge computer less like a fixed-function appliance and more like a small managed application platform.

From PLC to Edge AI: Five Generations of Industrial Computing

Industrial edge computing is easier to understand when placed in the context of what came before it. The generations overlap, and newer systems do not automatically make older ones obsolete. Each solves a different part of the problem.

Generation one: the PLC

The PLC controls machinery. Its strengths are predictability, ruggedness and deterministic operation. It can process inputs and outputs repeatedly with very little uncertainty.

A PLC should not be dismissed simply because a newer edge platform has more processing power. A general-purpose Linux computer and a safety-rated control system are not interchangeable.

Generation two: the industrial PC

The industrial PC brought general-purpose computing into factories. It could run SCADA software, databases, Windows applications and complex visual interfaces.

It also brought operating-system maintenance, storage considerations, greater power consumption and a different set of reliability questions. Industrial PCs remain essential where existing Windows software, substantial local storage or x86 processing is required.

Generation three: the industrial router

The industrial cellular router connected remote systems to the outside world. It introduced VPN access, firewalling, 4G and 5G communications, Ethernet routing and centralised configuration.

Some routers can also run scripts, protocol services or containers, which creates an overlap between advanced routers and entry-level edge computers.

Generation four: the industrial edge computer

The edge computer combines rugged hardware with an open application environment. It can run Linux software, collect data through industrial interfaces, host containers and communicate through Ethernet, Wi-Fi or cellular networks.

It is not limited to one prescribed cloud or one fixed function. The integrator can decide what it does.

Generation five: the AI edge computer

Edge AI adds hardware acceleration for machine-learning inference. Instead of merely forwarding a camera stream or vibration dataset, the device can interpret it locally.

That might mean recognising a number plate, identifying a manufacturing defect, detecting missing safety equipment, classifying an abnormal machine sound or spotting a change in equipment behaviour before a fixed threshold is exceeded.

The Case for Processing Industrial Data Locally

Edge computing is sometimes justified with vague claims about speed and digital transformation. The practical case is more specific.

Lower and more predictable latency

A cloud application depends on every part of the communications path between the site and the data centre. Even when the average response time is acceptable, variations in mobile coverage, internet routing or platform load can make that response unpredictable.

Local processing removes the wide-area network from the immediate decision loop. That does not automatically make an application suitable for safety-critical control, but it can make monitoring, analysis and operational responses much faster and more consistent.

Reduced cellular and cloud traffic

Data is not free merely because it travels over the internet. Cellular data, cloud ingestion, storage and processing all have costs.

An edge device can filter repeated values, calculate summaries and transmit only meaningful changes. A remote site might collect readings every second but send a five-minute average, while retaining detailed data locally for diagnostics.

For camera applications, the difference can be more dramatic. Sending an event record or small image crop uses far less data than continuously uploading high-resolution video.

Operation during connectivity loss

A remote site should not become useless simply because its external communications fail.

The edge computer can keep gathering data, operating local dashboards, running applications and storing records until the connection returns. This is often described as store and forward, although a resilient system should also define storage limits, reconnection behaviour and what happens when the local store becomes full.

Privacy and data control

Some applications do not need to transmit raw data at all. A people-counting system may need to report a number rather than identifiable video. A quality inspection system may need to record that a defect occurred without uploading every image of the production line.

Processing at the edge can minimise the amount of sensitive or commercially valuable information leaving the site.

Integration with equipment that predates the cloud

Industrial equipment can remain in service for decades. Many machines were designed long before MQTT brokers, cloud dashboards and REST APIs became common.

An edge computer can preserve the existing control system while exposing selected data to modern services. That is generally less risky than replacing a working PLC merely to obtain newer communications capabilities.

Why InHand Networks Is Relevant to the Edge Computing Discussion

InHand Networks approaches edge computing from a different starting point to a traditional industrial PC manufacturer. Its background is closely associated with industrial networking, cellular routers, remote connectivity and distributed IoT deployments.

That matters because many edge applications are not installed inside a well-connected factory. They sit in roadside cabinets, vehicles, utility sites, energy installations, remote buildings and other places where cellular connectivity is part of the primary architecture rather than an optional add-on.

InHand now offers products spanning compact edge gateways, general-purpose industrial computers and increasingly powerful AI edge systems. Its range includes platforms based on processors from Texas Instruments, Rockchip, Qualcomm and NVIDIA, allowing buyers to move from relatively modest data-processing workloads to accelerated machine vision.

The attraction is not one headline processor or one TOPS figure. It is the possibility of combining industrial interfaces, local processing, cellular communications and remote management in a single deployable platform.

This is particularly compelling for organisations that would otherwise assemble an industrial PC, separate cellular router, protocol converter and remote-access appliance.

Linux: The Operating Foundation of the Industrial Edge

Many compact edge computers use Linux because it offers a mature software ecosystem without the footprint or licensing model of a conventional desktop operating system.

Linux can run on Arm or x86 hardware, supports extensive networking capabilities and is familiar to developers building Python, Node.js, database and container applications.

For industrial users, however, the word Linux is not enough. Two devices can both use Linux while offering very different long-term experiences.

Buyers should ask:

• Which Linux distribution and kernel version are supplied?
• How long will security updates be available?
• Can the customer install standard packages?
• Is root or administrative access available?
• Are applications preserved during firmware upgrades?
• Is there a secure boot process and hardware trust anchor?
• Can a failed update be rolled back automatically?
• Does the vendor publish vulnerability notices and remediation guidance?

Moxa, for example, promotes its own Moxa Industrial Linux and publishes support periods for eligible device families. That is valuable for long-lived infrastructure because the operating system can become a greater lifecycle risk than the processor.

InHand’s Linux-based platforms can be attractive where the customer wants an open application environment alongside integrated communications. Advantech offers both Linux and Windows-oriented platforms, which may make more sense where a customer must preserve an existing x86 or Windows application.

The correct choice depends less on which logo is on the enclosure and more on what software must still be supported in five, seven or ten years.

What Node-RED Actually Does at the Edge

Node-RED is frequently listed as a feature without explaining why it matters. It is a visual, flow-based development environment that allows data sources, processing functions and outputs to be connected as a series of nodes.

For an industrial integrator, that can make Node-RED the glue between systems that were never designed to work together.

A Node-RED flow might:

1. Read holding registers from a Modbus PLC.
2. Convert an integer value into engineering units.
3. Add the machine identity, timestamp and site location.
4. Reject values outside a plausible range.
5. Write the complete dataset to a local database.
6. Publish a smaller set of values to an MQTT broker.
7. Send an alert when a condition persists for a defined period.
8. Resume queued transmissions when cellular connectivity returns.

That same process could be written as a conventional software application. Node-RED’s advantage is speed of integration and visibility. An engineer can see the path taken by the data and modify individual stages without rebuilding an entire application.

It is well suited to protocol integration, proof-of-concept systems, data routing, dashboarding and many production workloads. It should still be engineered carefully. Authentication, encrypted communications, credential handling, dependency management, change control and backups remain essential.

Node-RED should not be treated as a magical replacement for all PLC logic or professional software engineering. It is a powerful tool, not an exemption from architecture.

Why Docker Matters on an Industrial Edge Computer

Docker packages an application and its dependencies into a container. Each container can run as an isolated service while sharing the host device’s operating-system kernel.

In real terms, this means one edge computer might run:

• a Node-RED container for data flows;
• an MQTT broker for local messaging;
• an InfluxDB or other time-series database;
• a Grafana or local web dashboard;
• a Python protocol converter;
• a VPN or secure connector;
• a machine-vision inference application.

Without containers, these applications can still coexist, but their dependencies and update processes may interfere with one another. A library update required by one service might break another. Containers create clearer boundaries.

They can also make deployment repeatable. Rather than manually configuring every remote device, an organisation can deploy a tested container image across the estate.

Compute and memory headroom matter. A device that can technically start Docker may not have enough RAM, storage endurance or CPU capacity to run several services reliably under peak load.

This is one reason buyers should define the actual software stack before choosing the hardware.

What Do TOPS Numbers Mean in Real Applications?

AI edge products are commonly advertised using TOPS, or trillions of operations per second. It is a useful indication of accelerator capability, but it is not a universal measure of real application performance.

Two systems with the same headline TOPS figure may produce different results because of memory bandwidth, data type, accelerator architecture, software optimisation, model format, input resolution and thermal constraints.

TOPS should therefore be treated as an initial sizing indicator rather than a purchasing answer.

A basic pumping-station application reading Modbus values, calculating flow totals and publishing alarms does not need 100 TOPS. It may not need an AI accelerator at all.

A production line inspecting several high-resolution camera feeds at speed may need substantially more than a compact gateway can provide. The model must be benchmarked on the intended hardware using the intended camera resolution and frame rate.

The real questions are:

• Which model will run?
• At what resolution?
• How many streams or sensor sources?
• At what frame or sample rate?
• What maximum latency is acceptable?
• What happens when the device becomes hot?
• How much power is available?
• What other applications must run simultaneously?

Where Industrial Edge Computing Is Used

Manufacturing and machine integration

An edge computer can collect data from several generations of PLCs and machines, normalise the information and publish it into a unified data model.

Node-RED or a custom application might turn unidentified register values into named measurements with units, timestamps, asset identities and quality indicators. This information can then feed production reporting, predictive maintenance or an industrial data platform.

Machine-vision models can inspect components for defects, check labels, verify assembly steps or confirm that protective equipment is being worn.

Water and wastewater

Water infrastructure is geographically distributed and often cellular dependent. An edge computer at a pumping station or treatment asset can gather PLC and sensor data, maintain a local record, generate alarms and forward selected information to a central platform.

Local logic can identify abnormal pump cycling, persistent high levels, communication failure or a sensor that appears to be stuck.

The architecture must distinguish monitoring from control. Essential control should not depend on a public mobile connection or cloud service.

Renewable energy and distributed generation

Solar farms, battery systems and wind assets can contain equipment from several suppliers using different protocols. Edge processing can aggregate inverter, meter, weather and protection-system data before passing it upstream.

Local buffering is valuable where coverage is variable. Cellular connectivity can provide a primary or secondary path without extending fixed communications infrastructure to every site.

Transport and roadside systems

Roadside cameras, signs, ticketing equipment, passenger systems and vehicle gateways create both computing and communications requirements.

Edge AI can recognise vehicles, assess occupancy or classify events locally. Sending structured results rather than continuous raw video can reduce bandwidth considerably.

Rail and vehicle applications also introduce more demanding requirements for vibration, temperature, power input, connectors and sector-specific certification.

Retail and unattended sites

Remote retail, vending, kiosks and branch equipment benefit from local applications that continue working when the WAN is unavailable.

An edge device can integrate local sensors, cameras, serial equipment and payment-adjacent systems while maintaining a cellular connection to central services. Security boundaries must be designed carefully, particularly where multiple networks or regulated systems share a site.

Predictive maintenance

Predictive maintenance is not simply a sensor sending a threshold alarm. More advanced systems examine patterns over time, compare multiple variables and identify behaviour that differs from the normal operating signature.

Processing locally allows high-frequency data to be analysed without continuously uploading the entire raw dataset. The cloud can receive health indicators, exceptions and selected evidence while the edge retains detailed short-term history.

Agriculture and environmental monitoring

Remote sites may combine LoRaWAN or local sensor networks with a cellular backhaul. The edge gateway can operate the local network, validate data, run irrigation or environmental rules and synchronise with the cloud when connectivity is available.

InHand’s development of LoRaWAN-capable edge platforms is particularly relevant to this architecture because it joins local low-power devices to wide-area communications and computing.

Cellular Connectivity Changes the Edge Computing Model

Many descriptions of edge computing assume that a reliable wired network already exists. That assumption excludes a large part of the real industrial market.

Cellular connectivity allows the same application architecture to be deployed at sites where Ethernet, fibre or managed corporate WAN services are unavailable or uneconomic.

An integrated 4G or 5G modem can reduce the need for a separate router, although integration alone does not guarantee a good deployment. Antenna selection, installation position, signal quality, network behaviour, SIM design and remote diagnostics remain critical.

4G versus 5G at the industrial edge

Many telemetry and protocol-conversion applications require little bandwidth. A stable 4G connection may be entirely adequate.

5G becomes more relevant for applications involving large files, multiple cameras, higher uplink demand or low-latency private network environments. Buyers should resist using 5G as a substitute for workload analysis.

The edge computer may substantially reduce the amount of traffic transmitted, which can make a modest cellular connection more useful than expected.

Dual SIM is not the same as network resilience

A dual-SIM device provides two SIM positions or subscriptions. It does not automatically provide two independent mobile networks.

Both SIMs could ultimately depend on the same underlying operator, roaming arrangement, core platform or connectivity provider. A resilient design needs to understand the complete path, not merely count plastic cards.

The failover logic must also be tested. The device needs a reliable method of deciding that the primary path has failed, moving to the alternative and returning without becoming trapped in a switching loop.

eSIM and remote profile management

eSIM can allow connectivity profiles to be installed or changed without physically visiting the device. This can be useful for inaccessible or internationally deployed equipment.

However, the word eSIM covers different architectures and provisioning standards. Buyers should establish which eUICC standard is supported, who controls profile changes, what happens when provisioning fails and whether a bootstrap connection remains available.

Public IP, private APN and VPN design

The edge computer’s outbound cloud connection is only one part of the network design. Engineers may also need to reach a PLC, HMI, camera or local web interface behind the edge device.

That can be achieved through vendor remote-access services, private APNs, fixed or public IP services, site-to-site VPNs or overlay networks. The correct approach depends on security policy, scale and whether inbound access must survive a change of mobile network.

InHand, Moxa and Advantech: Three Different Starting Points

It is tempting to declare one manufacturer the winner. That would be misleading because the three companies approach edge computing from different histories and serve overlapping but non-identical requirements.

When InHand stands out

InHand is particularly persuasive where cellular connectivity is central to the deployment rather than a backup added later.

A remote site that needs industrial interfaces, Linux applications, dual-SIM connectivity and cloud management is close to InHand’s natural territory. Its broad range also gives customers a path from compact gateways to accelerated AI systems without abandoning the wider vendor ecosystem.

When Moxa stands out

Moxa has a strong reputation in industrial communications and operational technology. Its published emphasis on Moxa Industrial Linux, long-term maintenance and IEC 62443-aligned product security will appeal to organisations where lifecycle assurance carries more weight than the lowest initial hardware cost.

Moxa is a particularly credible choice where the edge computer becomes part of a wider Moxa networking and serial communications estate.

When Advantech stands out

Advantech’s industrial PC heritage becomes valuable where the application requires Windows, x86 software compatibility, expansion cards, substantial local storage or higher-end GPU processing.

An organisation replacing a conventional industrial PC may find Advantech’s architecture and support ecosystem more familiar than a compact Arm gateway.

Edge Computer, Industrial Router, PLC or Raspberry Pi?

These products overlap, but they are not the same.

A Raspberry Pi can demonstrate an application extremely effectively. That does not automatically make it the right production platform for an unattended pumping station expected to operate for ten years.

Conversely, buying an expensive industrial AI computer for a simple proof of concept can be equally irrational. The platform should match the stage and consequence of the application.

Resilience: What Happens When the Edge Computer Fails?

Edge computing can improve resilience when it allows a site to operate independently of the cloud. It can also introduce a new single point of failure if several previously separate functions are consolidated into one box.

Mission-critical architecture therefore needs more than dual SIMs and a watchdog.

Application recovery

Containers and services should be monitored. A crashed application may need to restart without rebooting the whole computer. The system should distinguish between a failed application and a failed internet connection.

Operating-system recovery

Remote updates need a rollback path. A failed package or firmware installation should not leave an inaccessible device in a roadside cabinet.

Read-only system partitions, dual-image firmware and atomic update mechanisms can reduce risk, but capabilities differ between products.

Hardware watchdog

A hardware watchdog can restart a device that has become unresponsive. It is useful, but repeated automatic rebooting can conceal an underlying problem. Remote management should record why the restart occurred and how often it is happening.

Storage endurance

Continuous high-frequency logging can wear flash storage. Designers should calculate write volume, use industrial-grade storage where required and buffer frequent writes in memory before committing them to disk.

Storage health should be monitored, and important data should not exist only on one embedded eMMC device.

Power resilience

Power failure, brownout, surge and poor grounding can cause more trouble than the mobile network. The installation may need a UPS, battery, redundant feed, surge protection or a controlled shutdown process.

AI systems also draw more power under load than entry-level gateways. Power supplies must be sized for the complete system, including cameras, USB devices, storage and cellular transmission peaks.

Network diversity

Cellular failover should use genuinely diverse underlying paths where the consequence of disconnection justifies it. A wired WAN plus cellular backup may provide more independence than two SIMs sharing the same operator infrastructure.

Operational replacement

Eventually, hardware fails. The important question is what happens next.

Can a replacement device be delivered quickly? Is the configuration backed up? Can containers and certificates be restored automatically? Is there a preconfigured spare? Who is authorised to replace it? How long can the site operate without it?

The industry standard is not a magic duplicate box waiting invisibly in every cabinet. It is a documented combination of fault detection, graceful degradation, recoverability, redundancy where justified and a tested human response.

Security and Certification

An edge computer bridges operational and external networks while running custom software. That makes it useful and potentially sensitive.

Relevant features may include secure boot, signed firmware, TPM 2.0, encrypted storage, role-based access, firewalling, VPN support, certificate management and a documented vulnerability disclosure process.

IEC 62443 is increasingly important in industrial cybersecurity procurement. However, buyers should identify exactly what has been certified. A secure-development lifecycle certification, an operating system claim and a product-level IEC 62443-4-2 Security Level certification are not identical.

Other certifications depend on the sector:

• ATEX, IECEx or Class I Division 2 for hazardous locations;
• EN 50155 and related rail requirements for onboard installations;
• vehicle and transport standards for mobile environments;
• EMC, shock, vibration and temperature testing;
• UKCA or CE conformity for products placed on the relevant market;
• sector-specific requirements imposed by utilities, transport authorities or asset owners.

A product brochure stating “industrial grade” should never replace checking the exact certificate, model number and intended installation environment.

How to Choose an Industrial Edge Computer

The selection process should start with the application rather than a preferred manufacturer.

Questions to Ask an InHand or Competing Edge Supplier

• What operating system and kernel does this exact model use?
• Until what date will security updates be provided?
• Can the device run standard Docker images without vendor modification?
• How much RAM and storage remain after the standard platform services are running?
• What is the storage technology and write-endurance rating?
• Does the firmware support automatic rollback after a failed update?
• Can containers be deployed and monitored centrally?
• Which industrial protocols are native, licensed or dependent on third-party software?
• Which cellular bands and 5G modes are supported by the precise regional variant?
• Can two SIMs use different operators and how is failover health checked?
• Which eSIM standard is supported, if any?
• What happens to local applications during a modem or network restart?
• Which IEC 62443 or sector certifications apply to the exact product?
• What is the maximum power draw with the modem, accelerator and peripherals active?
• What happens when the unit exceeds its thermal design limit?
• What UK support, advance replacement or RMA service is available?
• Can a replacement unit automatically recover its configuration and applications?
• What does the remote management platform cost at the expected fleet size?

Is InHand the Clear Winner?

InHand does not win every comparison. No serious industrial supplier does.

An Advantech platform may be the obvious answer when the existing application is built for Windows or requires x86 expansion. A Moxa platform may be easier to defend when long-term industrial Linux support, OT security certification and integration with an existing Moxa estate dominate the decision.

InHand’s strongest case is where computing and communications must be designed together.

For a remote, cellular-first deployment, an InHand edge computer can offer a more coherent proposition than combining a general industrial PC with a separate router and management environment. The benefit grows when the site needs several of the following at once:

• 4G or 5G connectivity;
• dual-SIM or multi-path communications;
• industrial serial, Ethernet or I/O interfaces;
• Linux applications;
• Docker containers;
• local data processing and storage;
• machine-learning inference;
• remote device and application management;
• secure access to PLCs and other local equipment.

The value is not that InHand has invented edge computing. It is that its cellular and industrial networking experience is increasingly being combined with meaningful local computing power.

That gives it a credible position in a market where many projects need the edge computer to be both a computing platform and the communications boundary of the remote site.

The Edge Computer Is Becoming Industrial Middleware

Twenty years ago, industrial automation was largely concerned with controlling the machine. Today’s projects also need to understand, move, secure and interpret the machine’s data.

The edge computer sits between those two worlds.

It can speak Modbus to the PLC, MQTT to the cloud, SQL to a database, HTTPS to an API and cellular to the outside world. It can run a Node-RED flow designed by an integrator, a Docker container supplied by a software company and an AI model trained by a data team.

That does not make every edge project simple. In some respects, it introduces more responsibility. Someone must maintain the Linux operating system, secure the applications, manage credentials, monitor storage, test failover and decide what happens when the device eventually fails.

But it also gives industrial organisations something they have often lacked: a flexible, remotely manageable software layer close enough to the machinery to remain useful when the cloud cannot be reached.

Products from InHand Networks, Moxa, Advantech and other industrial computing manufacturers are not merely smaller industrial PCs or more powerful routers. They represent the emergence of a distinct layer of industrial infrastructure.

For InHand, the opportunity is particularly clear. Its edge systems can combine the local intelligence of an industrial computer with the wide-area reach of a cellular router. In remote and distributed industrial IoT, those two capabilities increasingly belong in the same conversation.

Frequently Asked Questions