How DOCA classifies oil in high-purity compressed air

The DOCA Project’s Work Package 2 focuses on sensor algorithm development for oil classification. Its purpose is to turn optical measurements into useful information about contamination in high-purity compressed air, identifying oil in liquid, aerosol and vapour forms.

This software layer is essential for industries where a small amount of contamination can affect product quality, equipment reliability or patient safety. Pharmaceutical plants in Sydney and Melbourne, hospital services, electronics production, automotive workshops and clean-room operators all need monitoring that is dependable, rapid and practical to use.

Why oil classification matters

Oil contamination does not appear in a single, consistent form. Liquid droplets, fine aerosols and oil vapour interact with light differently, so a sensor must interpret more than a simple increase or decrease in optical intensity. The algorithm needs to distinguish a genuine oil event from changes caused by humidity, pressure, temperature, dust or sensor drift.

A useful classification system can help operators understand the seriousness and likely behaviour of contamination. A vapour issue may require a different investigation from a failed coalescing filter producing visible aerosol. Clear classification can also support preventative maintenance, reduce unnecessary shutdowns and provide evidence for quality assurance records.

For Australian facilities, this matters across very different operating conditions. A clean-room in Adelaide, a hospital in Melbourne and a remote Western Australian site may use different compressed-air systems, maintenance schedules and environmental controls. The algorithm must deliver consistent results rather than depend on ideal laboratory conditions.

From optical response to a usable signal

Work Package 2 links the physical sensor to a decision-making process. The optical detector produces raw data, which may include signal intensity, changes over time and responses at different operating points. Algorithm development converts these measurements into features that can be compared with known contamination states.

Possible features include signal amplitude, rate of change, baseline stability and the relationship between multiple optical channels. A sudden, persistent response may indicate an oil ingress event, while a short disturbance could be condensation, vibration or an installation effect. Filtering and compensation help prevent these disturbances from being reported as contamination.

The classification logic may combine calibrated thresholds with pattern recognition. Whichever approach is selected, it must remain explainable enough for engineers, quality teams and service technicians to understand why an alert was produced.

Building reliable training and test data

An oil classification algorithm is only as strong as the data used to develop it. DOCA’s testing therefore needs controlled samples representing liquid oil, aerosol and vapour, together with different concentrations and flow conditions. Data should also cover clean compressed air so the system learns what a normal baseline looks like.

Labelling is particularly important. Each measurement must be associated with a verified condition, test setup and environmental record. Temperature, humidity, pressure, flow rate and oil type can all influence the optical response. Repeated tests help identify natural variation and prevent the algorithm from learning a one-off laboratory artefact.

Testing in Australia may also need to account for hot, humid Queensland conditions and long periods of dry heat in inland or Western Australian locations. These environments can affect compressed-air treatment equipment and the behaviour of contaminants, making broader validation more valuable than a narrow bench test.

Distinguishing contamination from interference

A practical sensor must cope with the real world. Water droplets, particles, bubbles, vibration and fluctuations in compressed-air flow can produce optical signals that resemble oil. The algorithm must therefore assess the shape and persistence of a response, rather than classify every change as an oil event.

Signal conditioning can include baseline correction, noise reduction, drift monitoring and outlier handling. A confidence score may give operators a clearer indication of whether a result is definitive, probable or requires inspection. This is useful where the sensor is connected to an alarm, a plant control system or a maintenance platform.

The aim is to minimise both false alarms and missed events. Too many false alerts can lead technicians to ignore warnings; missed contamination can allow affected air to reach sensitive equipment or production processes.

Validation for industry and compliance

Algorithm validation should compare sensor classifications with reference methods and independently controlled contamination levels. The results can show whether the optical system correctly identifies oil phase, detects changing concentrations and maintains performance over time.

For Australian users, confidence in the technology may also depend on documentation that supports quality systems and audit processes. NATA-accredited testing, site validation and records aligned with relevant Australian Standards can help organisations assess the device alongside their existing compressed-air monitoring procedures. Pharmaceutical and healthcare operators may also need to consider expectations associated with the Therapeutic Goods Administration.

The project’s industrial testing and technical reporting are therefore important parts of commercial readiness. A strong algorithm is valuable, but customers also need to know its operating range, limitations, calibration requirements and response to abnormal conditions.

Applying the results across Australian sites

The classification output should be simple enough for a plant operator to act on quickly. A dashboard might display the contamination category, confidence level, trend and alarm status, while retaining detailed data for engineers and quality managers. This supports both immediate response and longer-term analysis.

In Australia, service access and distance can shape product requirements. A facility in regional New South Wales or a mining-related operation in the Pilbara may not have a specialist technician available on the same day. Stable calibration, remote diagnostics and clear fault messages can reduce the burden on local maintenance teams and tradies.

The same flexibility is useful in metropolitan hospitals and manufacturing clusters around Sydney, Melbourne and Brisbane, where compressed air may support critical operations around the clock. A compact online sensor with dependable classification can provide continuous oversight without relying solely on periodic laboratory sampling.

Priorities for dependable sensor algorithms

• Use reference measurements to label liquid, aerosol, vapour and clean-air conditions.
• Test across temperature, humidity, pressure and flow variations likely to occur in Australian facilities.
• Separate oil signals from water, dust, vibration and electronic noise.
• Report confidence and trends, rather than relying on a single alarm threshold.
• Document calibration, limitations and verification methods for quality teams and auditors.
• Design outputs that integrate with plant monitoring and remote maintenance systems.

Work Package 2 gives the DOCA sensor its interpretive capability. By combining optical measurement, signal processing and carefully validated classification, the project can help organisations identify oil contamination earlier and respond with greater certainty.

Explore the DOCA Project’s technical progress, testing activities and industrial applications to follow how the sensor algorithm develops from laboratory data into a practical tool for high-purity compressed-air monitoring.