“Our equipment is too old for AI.”

I hear this constantly from manufacturers running equipment from the 1990s, 1980s, or even older. And I understand the thinking—these machines were built before IoT, before machine learning, before most of what we call modern manufacturing technology.

But age doesn’t have to be a barrier. With the right approach, you can add AI capabilities to equipment that’s been running for decades.

Why retrofit rather than replace?

The obvious answer is cost. New equipment with built-in intelligence is expensive, often millions of dollars. Retrofit solutions typically cost a fraction of that.

But there are other reasons:

Proven performance: Old equipment that’s still running has proven itself. It works, it’s understood, spare parts are available. New equipment is unknown.

Disruption avoidance: Replacing major equipment means extended downtime. Retrofit can often happen during normal maintenance windows.

Skilled maintenance: Your team knows how to maintain existing equipment. New equipment requires new skills.

Capital constraints: Even if new equipment is justified, capital may not be available. Retrofit enables capability improvement within tighter budgets.

What can you actually retrofit?

The possibilities depend on what you’re trying to achieve and what the equipment allows.

Condition monitoring

Adding sensors to track equipment health—vibration, temperature, current draw, and other indicators.

What you need: Sensor mounting points, power for sensors, connectivity to collect data.

What you get: Early warning of failures, reduced unplanned downtime, data for predictive maintenance.

Typical cost: $5,000-50,000 per machine depending on complexity.

This is the most common retrofit application. Almost any equipment can be monitored by adding external sensors.

Process data collection

Capturing process parameters that the equipment already measures but doesn’t store or transmit.

What you need: Access to existing control signals, interface devices, connectivity infrastructure.

What you get: Production records, process visibility, data for quality analysis and optimisation.

Typical cost: $10,000-100,000 depending on control system complexity.

Many older machines have internal sensors and control signals. Getting that data out requires interface work but is usually feasible.

Enhanced visualisation and alerting

Displaying equipment status and alerts in modern dashboards rather than just local indicators.

What you need: Data collection (per above), visualisation platform, display infrastructure.

What you get: Real-time visibility, remote monitoring, automated alerts.

Typical cost: $20,000-80,000 for a facility depending on scale.

AI-assisted decision support

Adding analytical capabilities that interpret collected data and provide recommendations.

What you need: Data infrastructure (per above), analytical platform, AI models.

What you get: Anomaly detection, predictive alerts, optimisation recommendations.

Typical cost: $50,000-200,000 depending on scope and complexity.

This builds on the foundation of data collection. Without data, there’s nothing for AI to analyse.

Control augmentation

Adding AI-influenced control on top of existing control systems.

What you need: Understanding of existing control, interface capability, careful integration.

What you get: Optimised setpoints, adaptive control, automated adjustments.

Typical cost: $100,000+ due to integration complexity and risk management.

This is the most ambitious retrofit approach. It’s also the highest risk—interfering with working controls can cause problems.

Technical approaches

Several technical approaches enable retrofitting:

External sensors

Adding sensors externally, independent of the equipment’s original systems. Vibration sensors that clamp onto housings. Temperature sensors on surfaces. Current transformers on power feeds.

This approach doesn’t require any modification to the equipment itself. It’s low risk and widely applicable.

Control system interfaces

Connecting to existing control system signals through available interfaces—serial ports, network connections, PLC I/O points, or field devices.

Many older control systems have interfaces that were never used. A PLC from 1995 might have a serial port that can expose data with the right protocol.

Protocol converters

Translating older industrial protocols to modern formats. Modbus RTU to Ethernet. Proprietary protocols to standard formats.

A growing ecosystem of converters and gateways makes this increasingly practical.

Edge devices

Industrial PCs or gateways that collect, process, and transmit data. These sit between the equipment and the cloud/network infrastructure.

Edge devices are the hub of most retrofit solutions—collecting from multiple sources, doing initial processing, and forwarding to analytical platforms.

Common challenges and solutions

”There’s nowhere to mount sensors”

Often true at first glance, but creative mounting solutions usually exist. Magnetic mounts on ferrous surfaces. Adhesive mounts elsewhere. Custom brackets. Mounting on associated infrastructure rather than the equipment itself.

”The control system is proprietary”

Many older systems are proprietary, but often they communicate using standard or semi-standard protocols. Research the specific system—sometimes there’s more accessibility than expected.

If direct interface isn’t possible, indirect approaches work: monitor current draw instead of internal signals; use vibration as a proxy for operational state.

”We can’t touch the equipment”

For safety-critical systems, you might be restricted from modifying anything. But external sensors that don’t connect to the equipment—only measure it from outside—often don’t trigger the same restrictions.

Involve your safety and compliance team early to understand what’s allowed.

”The environment is harsh”

Industrial environments destroy consumer electronics quickly. But industrial-grade sensors and edge devices are designed for these conditions. IP ratings, temperature ranges, vibration resistance—specify appropriately.

”We don’t have network infrastructure”

Older facilities often lack the network infrastructure modern systems need. Options include:

• Wired network installation (expensive but reliable)
• Industrial wireless solutions
• Cellular connectivity (4G/5G modems on edge devices)
• Low-power wide-area networks for limited data needs

A practical retrofit project

Here’s how a typical retrofit project might proceed:

Assessment (2-4 weeks)

Understand the equipment: control systems, existing instrumentation, available interfaces, physical access, environmental conditions.

Define objectives: what data do you need, what decisions will it inform, what outcomes justify the investment.

Design (2-4 weeks)

Specify sensors, edge devices, connectivity, and software platform. Plan integration approach. Identify risks and mitigations.

Procurement (4-8 weeks)

Source equipment. Industrial components often have longer lead times than consumer equivalents.

Installation (1-4 weeks depending on scope)

Install sensors and devices. This often happens during planned maintenance windows to minimise production impact.

Integration (2-6 weeks)

Connect systems, configure data flows, validate data quality. This is where unexpected challenges typically emerge.

Analytics development (4-12 weeks)

Build dashboards, configure alerts, develop AI models. This can overlap with earlier phases.

Deployment and tuning (4-8 weeks)

Go live, monitor performance, tune models, train users.

Total timeline: 4-9 months for a meaningful retrofit project.

Making the case

Retrofit projects need business justification like any investment.

The strongest cases usually involve:

• Equipment critical enough that downtime is very costly
• Known reliability or quality problems that monitoring could address
• Skilled operator shortage (data enables less experienced operators to perform better)
• Regulatory or customer requirements for process documentation

Quantify the opportunity: downtime costs, quality costs, labour costs. Compare to retrofit investment. Most justified retrofits show payback under two years.

When replacement is actually better

Retrofit isn’t always the answer.

Near end-of-life equipment: If you’ll replace the equipment in 2-3 years anyway, retrofit investment may be wasted.

Major capability gaps: If the equipment fundamentally can’t do what you need, no amount of sensors will change that.

Safety-critical updates: If the equipment needs safety system upgrades that require major modification, replacement might make more sense.

Technology obsolescence: If control system parts are no longer available and you’re at risk of irreparable failure, replacement addresses that root issue.

Getting started

If you’re considering retrofitting older equipment:

1. Identify high-value targets: Which equipment would benefit most from monitoring and AI?

2. Assess feasibility: Can you physically access signals and mount sensors? What interfaces exist?

3. Define scope: Start focused. Monitor one critical machine before instrumenting the entire factory.

4. Find expertise: Retrofit projects often require skills across OT, IT, and AI domains. If you don’t have these internally, find partners who do.

5. Plan for scale: Design infrastructure that can expand as you prove value and add more equipment.

Your old equipment doesn’t disqualify you from manufacturing AI benefits. With thoughtful retrofitting, legacy machines can gain modern intelligence without the cost and disruption of replacement.