Most maintenance teams have seen both of these. A bearing is replaced while it still had months of life left. Another bearing fails two days after its inspection and stops a line.
Predictive maintenance and preventive maintenance differ in what triggers the work. This guide explains the difference, shows where each one fits on a factory floor, and covers what AI maintenance intelligence adds: finding out why an asset is misbehaving and what to do about it.
Preventive maintenance services equipment on a fixed schedule, by time or usage. Predictive maintenance services it when measured condition shows a problem developing. Most plants need both: preventive work for wear that follows a pattern, and predictive work for critical assets whose signals can be measured.
Predictive vs preventive maintenance: the core difference
Maintenance strategies differ in one thing above all: what triggers the work.
| Approach | Work is triggered by | Works best when | Watch out for |
|---|---|---|---|
| Reactive | A failure | Equipment is cheap, non-critical and easy to replace | Unplanned stops and the most disruptive repairs |
| Preventive | A fixed time or usage interval | Wear follows a known pattern | Healthy parts get replaced, and failures between visits are missed |
| Predictive | Measured equipment condition and its trend | Critical assets produce signals you can read | Needs connected, good-quality data and enough operating history |
| AI maintenance intelligence | Condition, plus history, manuals, SOPs and similar past failures | Repeat failures keep costing time to diagnose | Needs governance, and a person must approve the action |
Most plants use all of these at once and choose for each asset. Condition-based maintenance sits between preventive and predictive: it starts work when a measured value crosses a set limit, instead of forecasting when a failure will happen.
Why calendar-based maintenance wastes money
A fixed schedule gets two things wrong. It replaces parts that are still good, which costs money and technician time. It also misses parts that start to fail between inspections, and that is where unplanned downtime, scrap and rushed repairs come from.
The second one costs more. When one machine stops, the work queued behind it waits too. On high-value process equipment such as semiconductor and solar tools, a single unplanned stop can disturb lots, recipes and schedules.
Predictive maintenance uses data instead of the calendar. Vibration, temperature, pressure, cycle counts and alarms show when equipment moves away from its normal behavior, and the team acts at that point.
Monitoring, diagnostics and prediction are three different jobs
The word “predictive” is used loosely. It helps to split it into three jobs:
- Monitoring shows what the equipment is doing right now: its state, alarms and KPIs.
- Diagnostics explains why something went wrong, using fault logs, manuals, SOPs and similar past failures.
- Prediction flags unusual behavior against each asset's own baseline before it becomes a failure, and suggests what to do about it.
A tool that only monitors does not predict. A model that predicts without context gives technicians an alert and no next step. Good predictive maintenance software does all three jobs and keeps them connected.
Predictive maintenance starts with connected data, even on legacy equipment
A prediction is only as good as the signals behind it. Most plants run new and old machines side by side, and they speak different languages: OPC UA, SECS/GEM, Modbus, Profinet, MQTT, and in older equipment serial links such as RS-232 and RS-485.
OPC UA, maintained by the OPC Foundation, is a widely used interoperability standard for industrial data. In semiconductor fabs, tools typically talk to factory systems through SEMI's SECS/GEM communication standards. An industrial IoT gateway translates these protocols into one data model at the edge, so you can read signals from working legacy machines without replacing them.
If you run semiconductor tools, see our page on SECS/GEM integration, and our deeper post on bridging OPC UA and SECS/GEM.
Two limits to plan for. Some older machines have no standard interface and need a custom adapter. And models need operating history to learn what “normal” looks like for each asset. Results depend on the quality of the data you can capture.
What AI maintenance intelligence adds
A predictive alert says something looks wrong. It usually does not say why, whether it has happened before, or what the manual says. That information is spread across the alarm log, the ticket history, the SOP folder and the memory of whoever fixed it last time.
AI maintenance intelligence brings those together. A typical workflow:
- Detect the issue from an alarm, an anomaly or a technician's observation, and tie it to the exact equipment and component.
- Find similar past failures and related root cause analysis records.
- Search SOPs and manuals for the relevant procedures.
- Draft a root cause analysis the engineer can review, with the evidence linked.
- Recommend an action with priority and risk, assign an owner, and track the ticket to closure.
TCAIGV's AI Maintenance Command Center is built around this path from alarm to action. It works alongside Equipment Intelligence, which keeps asset structure, component condition and maintenance history in one connected equipment record, so every investigation starts with context.
Evidence is what makes the output usable. If a technician can see which SOP or past ticket a suggestion came from, they can check it quickly. A suggestion with no source is one more alert to doubt.
Who approves the action
In a factory, a wrong automated action can stop a line, scrap a batch or put a technician at risk. So the system should show data and recommendations, and operators and technicians should confirm every action on the equipment. Role-based access and an audit trail record who approved what.
For guardrails on AI in production, see our post on practical AI in operations.
Preventive vs predictive maintenance: which fits which asset?
| Asset profile | Best-fit strategy | Why |
|---|---|---|
| Critical equipment with readable signals (vibration, temperature, pressure) | Predictive | Early warning is worth the data effort |
| Parts that wear on a known cycle (filters, belts, seals) | Preventive | The schedule is accurate enough and cheap to run |
| Low-cost, non-critical, easy-to-swap items | Run to failure | Monitoring costs more than the failure |
| High-value process equipment with repeat failures | Predictive plus AI root cause analysis | Finding why it keeps failing is worth more than another alert |
Review the mix once a year. As data quality improves, more assets can move to predictive.
How to start: a five-step roadmap
You do not need to instrument the whole plant. Start with one asset, see what the data shows, then scale.
- Pick the asset that hurts most. Choose one tool or line with repeated unplanned stops or costly repairs. The ISO 17359 guidelines for condition monitoring and diagnostics of machines are a useful reference for deciding what to monitor.
- Connect its signals and alarms. Bring equipment data into one place through a gateway, legacy machines included.
- Gather what your team already knows. Manuals, SOPs, ticket history and technician notes are the evidence the AI will reason over.
- Run in advisory mode first. Compare the recommendations with what technicians actually decide, and note where they differ.
- Expand once the advice holds up. Add more assets, then add approval workflows for the actions that carry the most risk.
Frequently asked questions
What is the main difference between predictive and preventive maintenance?
The trigger. Preventive maintenance is done on a fixed schedule, by time or usage. Predictive maintenance is done when the measured condition of the equipment shows a problem developing.
Is predictive maintenance better than preventive maintenance?
Not always. For critical assets with signals you can read, predictive maintenance usually pays off. For parts that wear on a known cycle, a schedule is simpler and cheaper. Most plants combine both and decide asset by asset.
Can you do predictive maintenance on legacy equipment?
Often yes. An industrial IoT gateway can read protocols such as OPC UA, Modbus, Profinet, MQTT and SECS/GEM, plus serial interfaces such as RS-232 and RS-485. Some older machines have no standard interface and need a custom adapter, so connectivity should be validated asset by asset.
What data do you need to start predictive maintenance?
Equipment signals such as vibration, temperature and pressure, alarms and fault logs, and maintenance history. Manuals and SOPs add the context needed for diagnosis. The more clean history you have for each asset, the better the baseline.
How is AI predictive maintenance different from condition monitoring?
Condition monitoring measures and reports. AI predictive maintenance learns each asset's normal baseline and flags drift, and when it is paired with maintenance intelligence it links the alert to likely causes and a recommended action for a person to approve.
See predictive maintenance on your own equipment
Ask for a demo and we will show how TCAIGV connects your equipment, links alarms to evidence, and leaves approvals with your technicians.
Book a demo