• Unplanned Downtime: Mid-route failures causing network-wide delays.

• Maintenance Inefficiency: Unnecessary replacement of healthy components due to fixed schedules.

• Visibility Gaps: Lack of root-cause analysis for recurring fault patterns across different vendors.

• Inventory Strain: Inability to accurately forecast spare part consumption.

Exponentia.ai delivered an end-to-end AI solution designed for high-velocity IoT data:

• Data Ingestion: Real-time streaming of telemetry (temperature, pressure, vibration) via Kafka.

• Data Lakehouse: Integration of maintenance history and fault logs using Talend and Apache Druid.

• AI Engine: Catboost Classification models for predicting failures in critical systems like traction motors and axles.

• Command Center: Qlik Predict dashboards providing failure probabilities and recommended maintenance windows.

The architecture integrates real-time IoT ingestion from onboard sensors, stores and processes data in a high-performance analytics layer (Apache Druid), and enables predictive insights through AI models and interactive dashboards.

Technologies Used

• AI/ML: Catboost, Qlik Predict

• Data Ingestion: Kafka, Talend

• Storage/Analytics: Apache Druid

• IoT: Onboard locomotive sensors

• 30% Reduction in unplanned downtime.

• 15% Improvement in locomotive availability.

• 18% Savings in spare parts inventory costs.

• 8% Boost in network-wide punctuality.

‍