OperationsFramework / modelModelAccessible

Machine data capture

How can machine data capture support strategic choice or positioning?

AccessibleOperationalOrganisation2 min read

Contents

Machine data capture is basically sensors that are embedded into machines.

Machine data capture collects measurements generated by equipment, products and connected sensors. It can support condition monitoring, quality, safety, energy management and new services when the data has a defined purpose and trustworthy operational context.

When to use it

Use sensor data when a measurable physical condition can improve a decision: detecting degradation, scheduling maintenance, verifying performance or understanding product use. Begin with the decision and failure mode, not with the availability of a data stream.

Embedded sensing can reduce unnecessary maintenance and reveal problems before failure. In consumer products it may also create sensitive health, location or behavioural data, so usefulness must be balanced with consent, security and purpose limitation.

Origins

Machine capture developed from industrial instrumentation, automatic control, telemetry and supervisory systems. Falling sensor, storage and connectivity costs later expanded it into machine-to-machine communication and the Internet of Things. Predictive-maintenance practice combines this engineering lineage with reliability analysis and machine learning; no single inventor owns the field.

What it is

Sensors measure conditions such as temperature, vibration, pressure, current, sound, position and flow. Data may be processed at the machine, edge or central platform. A useful record includes timestamp, asset identity, operating state, units, calibration and quality—not only a numeric reading.

Why it matters

Calendar-based maintenance can replace healthy parts or miss failure that develops between inspections. Condition monitoring can target intervention, but a warning model must be validated against actual failure and maintenance outcomes. Connectivity also creates cyber and operational risk; a sensor does not make equipment safe unless it belongs to an engineered control and assurance system.

How to use it

Define the asset, failure mode, required lead time and action. Identify a signal linked to that condition, choose sampling and placement, validate calibration and establish secure identity and transmission. Record operating context so normal variation is not mistaken for degradation.

Integrate alerts with maintenance workflow, competence and spare-parts planning. Set thresholds from engineering evidence, estimate false alarms and missed detections, and provide human review. Monitor drift, sensor failure and model performance. Safety-critical changes require the relevant engineering, certification and regulatory process.

Possible data sources

Sources include installed control systems, retrofit sensors, equipment logs and connected products. Common variables include temperature, light, pressure, humidity, level, movement and proximity.

How difficult or costly is it to collect?

Modern assets may expose data through standard interfaces; older equipment can require retrofit, networking and validation. Total cost includes cybersecurity, integration, labelling, storage, maintenance, calibration, analysis and change management—not only the sensor.

Practical example

A widely cited Rolls-Royce example describes engines measuring about 40 parameters 40 times per second. Data supports condition-based service for more than 3,700 engines, and service activity was reported as roughly 70 per cent of civil-engine revenue. Footnotes 1 and 2 attribute the example to media and published big-data accounts.

This illustrates a shift from selling equipment toward availability or usage-based service. It does not mean raw telemetry is streamed without filtering or that analytics replaces certified maintenance and flight-safety systems. Commercial incentives, data rights and responsibility for outages must be explicit.

BBC 2 Bang Goes the Theory, Series 8, “Big Data” (March 2014); Mayer-Schönberger, V. and Cukier, K. (2013), Big Data, London: John Murray Publishers

Top practical tip

Start with one failure mode and one operational decision. Capture only the signals and context needed to improve that decision, then validate the closed-loop result.

Top pitfall

A modern aircraft may contain over 10,000 sensors, but volume is not value. Unsecured, uncalibrated or context-free telemetry creates noise and risk rather than insight.