What is Predictive Maintenance?

Predictive maintenance is a condition-based maintenance strategy that uses real-time data from sensors, combined with analysis of that data, to predict when an asset is likely to fail – and schedule maintenance before it does. The goal is to intervene at exactly the right moment: late enough that you are not replacing components with plenty of life left, early enough that you avoid an unplanned failure and the costs that come with it.

It is the most sophisticated point on the maintenance strategy spectrum, and the one that IoT connectivity makes practically achievable at scale for the first time.

The three maintenance strategies compared

Reactive maintenance is cheap to set up and expensive to operate. A compressor failure on a production line does not just cost the repair bill – it costs every hour of halted production while parts are sourced and engineers are dispatched. In a manufacturing context, that cost can reach tens of thousands of pounds per hour. In a utility or offshore context, it can be higher still.

Preventive maintenance solves the unplanned failure problem but creates a different inefficiency: servicing assets on a calendar schedule regardless of their actual condition. A bearing that has three months of life left gets replaced because the maintenance schedule says it is time. A bearing that is failing ahead of schedule gets missed because the next service date is six weeks away. The schedule is a proxy for condition, not a measurement of it.

Predictive maintenance replaces the proxy with direct measurement. If you know the actual condition of every asset on your site in real time, you can make evidence-based decisions about when to intervene – and the evidence comes from IoT sensors transmitting continuous data to an analysis platform.

How IoT Enables Predictive Maintenance

The IoT stack for predictive maintenance has four layers: the sensors that capture condition data, the connectivity that moves it, the edge or cloud processing that analyses it, and the platform that presents actionable insights and triggers maintenance workflows.

Layer 1: Condition monitoring sensors

Sensors are attached to, embedded in, or positioned near the assets being monitored. They capture the physical signals that indicate asset health – vibration, temperature, acoustic emissions, electrical current draw, oil quality, humidity, and pressure, depending on the asset type. Modern IoT condition monitoring sensors are wireless, battery-powered or energy-harvesting, and transmit data continuously or at configurable intervals over a local wireless protocol – typically LoRaWAN, Zigbee, or Bluetooth LE – to a gateway device.

Layer 2: Connectivity and gateways

The gateway aggregates data from local sensors and transmits it upstream over a wide-area connection – most commonly cellular (4G LTE or 5G), but also fixed broadband or satellite where cellular coverage is absent. The gateway is where the connectivity architecture decisions matter most, and where the reliability of the entire predictive maintenance system is ultimately determined. A misconfigured gateway, a dropped SIM connection, or a cellular link with inadequate uptime SLA can silently create gaps in the data that make the analysis above it unreliable.

Layer 3: Edge and cloud processing

Raw sensor data from a vibration sensor measuring at 10kHz produces enormous volumes of data. Transmitting all of it to the cloud is impractical and expensive. Edge processing – running on the gateway itself or on a dedicated edge compute node – handles the first stage of analysis locally: signal processing, feature extraction, anomaly detection against a baseline. Only meaningful events and processed metrics are transmitted upstream, reducing bandwidth requirements significantly and enabling real-time response to critical anomalies without cloud round-trip latency.

Layer 4: Platform and action

The platform receives processed data, applies trend analysis and machine learning models, generates alerts when asset health crosses defined thresholds, and integrates with CMMS (Computerised Maintenance Management Systems) or ERP platforms to create work orders automatically. The maintenance engineer does not look at raw sensor data – they receive a notification that bearing 3B on pump 7 is showing vibration signatures consistent with race wear and needs inspection within the next 14 days.

Condition Monitoring Sensor Types

Different assets fail in different ways, and the right sensor type depends entirely on the failure modes you are trying to detect. A blanket approach – fitting every asset with the same sensor – is less effective than matching the sensor type to the asset’s known failure signatures.

Why Cellular Connectivity is the Critical Link

Predictive maintenance deployments span a wide range of physical environments – factory floors, substations, pumping stations, offshore platforms, remote agricultural sites, roadside infrastructure. The one thing many of these environments have in common is that running fixed network infrastructure to every asset location is impractical, expensive, or impossible.

Cellular connectivity – 4G LTE in the majority of current deployments, with 5G adoption increasing in high-density industrial environments – provides the wide-area data path that makes IoT predictive maintenance viable at remote and distributed sites without civil engineering.

What cellular connectivity provides that wired infrastructure cannot

• Rapid deployment – a cellular-connected IoT gateway can be installed and transmitting data in hours, not weeks. No cable routing, no network infrastructure build, no IT dependency for connectivity provisioning.
• Coverage at remote sites – pumping stations, substations, water treatment works, wind turbines, and agricultural buildings are frequently beyond the reach of fixed broadband but within cellular coverage. 4G LTE coverage in the UK now reaches the vast majority of industrial sites.
• Resilience – a dual-SIM cellular gateway with SIMs from two different mobile network operators maintains connectivity through single-MNO outages. Fixed broadband provides no equivalent built-in resilience without additional hardware.
• Scalability – adding a new monitoring point to an existing predictive maintenance deployment means installing a sensor and a cellular gateway, not running network cable. The connectivity architecture scales with the deployment without infrastructure changes.

Private APN for predictive maintenance deployments

For organisations running predictive maintenance across critical infrastructure – utilities, energy, manufacturing – a private APN provides a dedicated, isolated network path for IoT traffic. This has two specific benefits in a predictive maintenance context.

First, security. Sensor data from a manufacturing process or utility substation should not be traversing the public internet. A private APN keeps IoT device traffic on a private network path from the SIM card to the organisation’s own infrastructure, significantly reducing the attack surface compared to devices communicating over the public internet.

Second, fixed IP addressing. With a private APN and fixed IP SIMs, every gateway in a predictive maintenance fleet is directly addressable. Remote configuration, firmware updates, and health checks can be performed without VPN workarounds or dynamic DNS complexity. For a fleet of gateways spread across dozens of remote sites, this operational simplicity has real value.

Predictive Maintenance Use Cases by Sector

Predictive maintenance with IoT connectivity is deployed across a wide range of industrial sectors. The sensor types and analysis methods vary, but the underlying architecture – wireless sensors, cellular-connected gateway, edge processing, cloud platform – is consistent.

Manufacturing

Manufacturing has the longest history of condition monitoring, and the most mature IoT predictive maintenance deployments. Rotating machinery – motors, pumps, fans, compressors, conveyors – is the primary target. Vibration and temperature sensors on motor bearings provide continuous health data. A fault developing on a critical pump in a continuous process line can be identified weeks before failure, scheduled into a planned maintenance window, and repaired without production impact. The alternative – an unplanned failure on a running line – can cost more in lost production in a single hour than the entire cost of the monitoring infrastructure.

Utilities and water

Water utilities operate thousands of pumping stations, many of them unmanned and in remote locations. Traditional maintenance relies on periodic inspection rounds – an engineer visiting each site on a schedule, regardless of asset condition. IoT predictive maintenance replaces the inspection round with continuous remote monitoring. Vibration analysis on submersible pumps detects impeller wear and bearing degradation. Flow and pressure monitoring identifies blockages and pipe degradation. Cellular connectivity provides the data path from remote pumping stations where fixed network infrastructure is absent.

Energy and utilities

Transformer health monitoring uses temperature sensors, acoustic emission sensors, and dissolved gas analysis to detect insulation degradation and incipient faults in high-voltage transformers before they fail. A transformer failure in an electricity distribution network triggers a large-scale outage and an emergency replacement programme. Predictive monitoring of transformer condition enables planned replacement during scheduled outages – a fraction of the cost and disruption of an unplanned failure.

Offshore and oil and gas

Offshore environments present the most demanding case for predictive maintenance: extreme environmental conditions, high asset criticality, and enormous cost of failure or unplanned maintenance. Access to offshore assets for manual inspection is expensive and weather-dependent. IoT condition monitoring with satellite or cellular backhaul provides continuous asset health data without the logistics of a site visit. Rotating equipment on offshore platforms – compressors, gas turbines, pumps – is a primary target, with vibration analysis providing early warning of mechanical faults.

Fleet and vehicles

Vehicle telematics has long provided basic predictive maintenance data – engine fault codes, mileage-based service triggers, tyre pressure monitoring. IoT predictive maintenance takes this further: continuous monitoring of engine parameters, transmission health, brake wear, and auxiliary systems, with cellular connectivity providing real-time data from vehicles in operation. Fleet operators can prioritise maintenance interventions based on actual vehicle condition rather than mileage alone, reducing both unplanned breakdowns and unnecessary scheduled maintenance.

IoT Platforms for Predictive Maintenance Data

Sensors and gateways capture and transmit data. The platform receives it, stores it, analyses it, and presents actionable insights. The choice of platform determines what you can do with the data once you have it, and how easily the predictive maintenance system integrates with existing operational workflows.

For organisations building their own predictive maintenance infrastructure, open-source platforms provide a cost-effective foundation. ThingsBoard is widely used in industrial IoT deployments for exactly this use case – it handles MQTT and HTTP ingestion from edge gateways, supports custom dashboards for condition monitoring visualisation, provides a rule engine for threshold-based alerting, and integrates with CMMS and ERP systems via REST API. Both self-hosted and cloud-managed options are available, giving organisations control over where their operational data resides.

The key capabilities to evaluate in a predictive maintenance platform are:

• Time-series data storage – predictive maintenance generates continuous time-series data. The platform needs to handle high-frequency ingestion efficiently and support trend analysis over extended periods – weeks, months, or years of historical data are needed to build reliable failure prediction models.
• Threshold and anomaly alerting – configurable alert rules that fire when sensor values exceed defined limits or when anomaly detection identifies deviation from normal operating patterns.
• Asset hierarchy management – the ability to organise monitored assets in a logical hierarchy (site, area, asset, component) for clear navigation and roll-up reporting across a large fleet.
• Integration with CMMS – automatic work order creation in the CMMS when the platform identifies a maintenance requirement, closing the loop between condition data and the maintenance workflow.
• Remote access to dashboards – maintenance engineers and operations managers need access to asset health data from wherever they are. Mobile-accessible dashboards and configurable alerting via SMS or email are baseline requirements.

What Goes Wrong in Predictive Maintenance Deployments

Predictive maintenance is well-proven technology. Deployments that underperform typically do so for one of four reasons, none of which are the AI or the sensors.

Connectivity gaps treated as acceptable

The most common infrastructure failure. A cellular gateway with intermittent connectivity creates silent gaps in the data. The analysis platform continues running on the data it receives, with no indication that data is missing. A fault developing during a gap is invisible. The fix is reliable dual-SIM connectivity with active monitoring of connection health, not just data transmission health.

Wrong sensor placement

A vibration sensor measuring bearing condition needs to be mounted on the bearing housing, not on the motor casing a foot away. Sensor placement guidance for each asset type is well-documented in condition monitoring standards (ISO 10816, ISO 13373). Skipping this step produces data that is technically valid but diagnostically useless.

No baseline established

Anomaly detection works by comparing current condition to a known-good baseline. If a deployment goes live without establishing a baseline during a period of normal operation, the model has nothing to compare against. Early deployments should include a baseline capture phase before live monitoring begins.

Alerts with no defined response process

A predictive maintenance platform generating alerts is not useful if no one has defined what to do when an alert fires. Alert thresholds, escalation paths, and the process for converting a platform alert into a maintenance work order need to be defined and tested before deployment. A system that generates alerts nobody acts on quickly loses credibility and adoption.