AI-Driven Predictive Maintenance for Public Lighting Systems

Use Case:

AI-powered analysis of public lighting telemetry data to anticipate coverage battery and hardware failures in distributed luminaires.

Outcome:

• Accurate forecasting of 3 failure types in public lighting (radio, battery, hardware);
• Full technology transfer and AI upskilling of IHMAN’s engineering team;
• Scalable and generalizable solution across multiple Spanish municipalities;
• Reusable architecture and codebase built in Python with ElasticNet, SVR, LGBMRegressor models.

Ecosystem Support:

• StairwAI support program;
• Technical mentorship;
• Collaboration with CONNECTHINK AI experts;
• Use of historical data from HELIOS-managed cities (Canyelles, Illora, Mejorada del Campo)

AI Relevance:

This pilot showcases AI accessibility and transfer in a traditionally non-AI domain via:

1. Low-barrier integration of ML models with legacy systems;
2. End-to-end knowledge transfer from AI experts to SME staff;
3. Reuse of long-term telemetry data;
4. Platform-independent deployment using MLflow, Python, and open-source tools.

Summary:

IHMAN, a Spanish SME specializing in human-machine interfaces for smart cities, launched the HELIOS-AI project to improve predictive maintenance in public lighting systems. With support from the StairwAI program, IHMAN collaborated with AI experts from CONNECTHINK to develop a robust AI system trained on real-world lighting telemetry data collected since 2019 across three municipalities. The AI model focused on predicting three key failure types—radio signal, battery, and luminaire hardware issues—linked to individual distribution panels managing groups of streetlights. After an in-depth exploratory data analysis and model training phase, the AI system achieved an 8.5/10 reliability score in failure prediction. In addition to building the AI solution, a core goal was knowledge transfer. CONNECTHINK engineers conducted workshops and collaborative programming sessions to upskill IHMAN staff in AI development, model integration, and future maintenance. MLflow was deployed for experiment tracking, with all code developed in Python and hosted on a dedicated in-house server.
The project reached TRL 7 and delivered both a functional predictive tool and a go-to-market business plan that explored commercialization channels, partner networks, and monetization models. The solution now positions IHMAN to lead smart infrastructure maintenance with reduced downtime, improved service reliability, and data-informed planning.