Predictive Maintenance Sensor

A Predictive Maintenance Sensor is a sophisticated edge device that combines high-fidelity sensing, local analytics, and machine learning to predict equipment failures before they happen. Unlike traditional monitoring sensors that simply report current values, this device:

• Monitors multiple parameters – vibration, temperature, current, voltage, humidity, and acoustic emissions
• Performs edge-based AI inference – using pre-trained models to detect anomaly patterns
• Predicts remaining useful life (RUL) of critical components
• Provides actionable insights – recommending specific maintenance actions
• Learns and adapts – continuously improving prediction accuracy over time

This sensor represents the pinnacle of the Smart Airport paradigm, transforming maintenance from reactive "run-to-failure" to proactive "predict-and-prevent" strategies.

The predictive maintenance sensor's purpose is to shift AGL maintenance from reacting to failures after they happen to acting on early warning signs while a fault is still developing, at the specific point where the highest-consequence equipment — CCRs, isolating transformers, LED drivers — is most likely to fail unexpectedly. By tracking subtle changes in vibration, temperature, or current that precede a breakdown, it gives maintenance teams the lead time to schedule a repair on their own terms rather than responding to an unplanned outage that could interrupt a runway or taxiway circuit during operations. Its value is therefore measured less in the data it collects and more in the maintenance visits it prevents from becoming emergencies.

Predictive Maintenance Sensors are deployed on critical AGL equipment where early failure detection is essential:

• Constant Current Regulators (CCRs): Monitoring transformers, cooling fans, and power electronics for early signs of degradation.
• Isolating Transformers: Detecting winding insulation breakdown, core saturation, and thermal stress.
• LED Drivers and Fixtures: Predicting driver failures and LED lumen depreciation.
• Motors and Actuators: Monitoring bearing wear, misalignment, and imbalance in beacon drives and moving parts.
• Switchgear and Circuit Breakers: Detecting contact wear, arcing, and insulation degradation.
• UPS and Battery Systems: Predicting battery end-of-life and inverter failures.

They are deployed as part of comprehensive condition monitoring programs, typically integrated with IoT Gateways and cloud-based analytics platforms.

The sensor is fixed directly to the component it monitors, typically bolted, magnetically clamped, or adhesive-mounted onto the transformer casing, motor housing, or driver enclosure so vibration and thermal readings are taken from the source rather than through an intervening structure that would damp or delay the signal. Mounting position and orientation follow the manufacturer's guidance for the specific measurement (an accelerometer needs a rigid, flat mounting face aligned to the axis being monitored), and the unit is sited so its antenna or communication module has a clear path to the nearest gateway. Power is drawn from the host equipment's local supply or a self-contained battery, and cabling or wireless commissioning is completed and range-tested at installation before the sensor is added to the monitoring platform.

The Predictive Maintenance Sensor operates through a sophisticated pipeline of sensing, signal processing, and AI inference:

• Multi-Modal Sensing: Continuously captures vibration (accelerometer, 3-axis), temperature (thermistor/RTD), current (Hall effect), and acoustic emission (ultrasonic microphone) data at high sampling rates (up to 25 kHz).
• Signal Conditioning and Pre-processing: Filters noise, performs FFT for frequency domain analysis, and extracts key features (RMS, crest factor, spectral energy).
• Edge AI Inference: Runs a lightweight neural network (TinyML) or statistical model directly on the device to detect anomalies and predict failure probability.
• Self-Learning: Uses online learning techniques to adapt to changing equipment conditions and improve prediction accuracy over time.
• Reporting: Transmits health scores, anomaly alerts, and recommended actions to the ALCMS or cloud platform.

The sensor is typically powered by energy harvesting (vibration or thermal) or long-life batteries, with a wireless communication interface (LoRa, Wi-Fi, or NB-IoT) for data transmission.

Control logic here runs almost entirely at the edge rather than centrally: sampling, FFT and inference happen on the sensor itself, and only a health score or anomaly flag is escalated to the ALCMS or cloud platform when a threshold is crossed, rather than streaming raw waveform data continuously — this keeps bandwidth low over LoRa/NB-IoT links. The sensor's "control" is advisory, not command-and-control: it triggers a maintenance workflow or alert, it does not switch a circuit. Where it monitors a fixture on a series circuit already watched at loop level by the CCR, the sensor's per-unit diagnosis complements that loop-level fault detection, narrowing a generic "loop fault" alarm down to a specific fixture or transformer before a technician is dispatched.

The sensor is a compact, hermetically sealed unit combining a MEMS 3-axis accelerometer, an RTD or thermistor, and (where fitted) a Hall-effect current sensor on a single PCB, potted or conformal-coated against moisture and vibration to reach an IP67/IP68 submersible rating. A magnetic or bolt-on mounting base clamps it directly to the monitored equipment — a CCR contactor, a transformer, or a fixture housing — so the accelerometer picks up vibration through solid contact rather than through the enclosure alone, since any gap between sensor and equipment would damp the very signal the unit is trying to read. Onboard flash holds the quantized AI model that runs the edge prediction, and a small internal battery (or energy-harvesting module) powers the sensor through the -40°C to +85°C range for a rated 5–10 years without a wired power feed.