ETH Zürich × EPFL x SBB

AI-Powered Condition Monitoring for Critical Infrastructure

Designed, deployed and operated an end-to-end monitoring system for four 132 kV circuit breakers in the Swiss Federal Railways (SBB) network. The project combines multimodal sensing, embedded systems, IoT connectivity, machine learning and anomaly detection to support maintenance decision-making in critical infrastructure.

132 kV Breakers Monitored

12+

Months Continuous Operation

30k+

Laboratory Operations

200+

Field Operations

Project Overview

This project was conducted during my PhD at ETH Zürich in collaboration with Swiss Federal Railways (SBB), Hitachi Energy, BKW Energie and EPFL. The objective was to develop a scalable condition monitoring framework capable of detecting abnormal behaviour in high-voltage circuit breakers before failures occur.

Project Gallery

Click to enlarge

Field test setup on high-voltage circuit breaker — Field test setup on a circuit breaker

IoT condition monitoring system architecture — End-to-end condition monitoring system architecture

IoT condition monitoring box — Custom embedded IoT data acquisition and transmission box

My Contributions

Project leadership and stakeholder coordination across ETH Zürich and SBB
Requirements gathering, deployment planning and project management
Design of sensing architecture and monitoring strategy
Identify suitable sensors and installation locations
Machine learning model development and validation
Development of data engineering and preprocessing pipelines
Dashboard development and monitoring software
Design and execution of run-to-failure experiments
Field installation coordination and long-term operation support

System Architecture

Multimodal sensor integration
Embedded data acquisition platform
Remote IoT data transmission infrastructure with WiFi
Automated preprocessing and feature extraction pipelines
Anomaly detection algorithms trained on the data collected in the first few months, assuming healthy
Online monitoring dashboards and visualization tools

Field Deployment

Installed and commissioned monitoring systems on four 132 kV circuit breakers
Deployment completed within a limited maintenance window of only a few hours
Maintained stable operation for more than one year
Collected operational data under real railway infrastructure conditions

Dataset Development

Designed and executed extensive run-to-failure experiments
Collected a multimodal dataset comprising more than 30,000 laboratory switching operations
Collected over 200 real-world switching events from circuit breakers in the substation
Included current interruption events for model validation

Impact

Demonstrated successful deployment of AI-driven condition monitoring in a live substation environment
Bridged research prototypes and operational infrastructure deployment
Enabled long-term condition monitoring of critical infrastructure assets
Created one of the largest multimodal datasets for circuit breaker condition monitoring research
Established a scalable framework for future condition-based maintenance systems

Project Team

Chi-Ching Hsu — Project Lead, Machine Learning, Dashboard Development
Tino Gfrörer — Hardware System Design
Zoltán Marcsek — Firmware & Embedded Systems
Prof. Christian Franck — Academic Supervision
Prof. Olga Fink — Academic Supervision

Publication

An IoT Sensor Platform for Predictive Maintenance of High Voltage Circuit Breakers

Zoltán Marcsek, Tino Gfrörer, Tommaso Polonelli, Chi-Ching Hsu, Michele Magno, Christian M. Franck

2025 10th International Workshop on Advances in Sensors and Interfaces (IWASI), 2025

Paper Link

Back to homepage