Industrial IoT in Transit: Predictive Maintenance for Metro Systems

Unplanned downtime in public transportation is not just an inconvenience; it is a massive financial and logistical burden. The MetroPT-3 Dataset offers a rare glimpse into real-world Industrial IoT (IIoT) data, collected from the Air Production Unit (APU) of an operational metro train in Porto, Portugal.

1. From Preventive to Predictive

Traditionally, transit agencies rely on Preventive Maintenance (scheduled repairs regardless of condition). This is inefficient. Predictive Maintenance (PdM) uses IoT sensors to monitor health in real-time, predicting failures before they happen.

This dataset is unique because it captures the degradation process of a specific component: the Air Production Unit (APU), which is critical for the train's braking system and pneumatic doors.

2. Analyzing the Sensor Data

The dataset contains multivariate time-series data sampled at 1Hz. Key signals include:

• Pressure Sensors (TP2, TP3): These measure the compression efficiency. A gradual drop in pressure relative to motor speed often indicates a leak or valve failure.
• Motor Current: High current consumption without corresponding pressure increase is a classic signature of mechanical friction or motor burnout.
• Oil Temperature: Spikes here often precede catastrophic mechanical failure.

3. Proposed Methodology: Anomaly Detection

Instead of simple classification, an Unsupervised Anomaly Detection approach is often superior for this dataset:

1. Train an LSTM Autoencoder on the "Normal" data only.
2. The model learns to reconstruct the normal sensor patterns.
3. When fed "Failure" data, the Reconstruction Error will spike.
4. Set a dynamic threshold for this error to trigger a maintenance alert.

Business Impact

Implementing an AI model trained on this dataset can reduce maintenance costs by up to 30% by eliminating unnecessary scheduled repairs and preventing costly service interruptions during peak hours.