Digital Twin for Photovoltaic System Maintenance Support

With the continued growth of industrial photovoltaic (PV) installations, the need to maximise plant uptime and operational efficiency has become increasingly important. Reducing downtime through monitoring and predictive maintenance enables earlier fault detection and improved decision-making for maintenance planning. This thesis presents a modular, holonic digital twin (DT) architecture for maintenance support of industrial PV plants. The proposed architecture integrates rule-based and machine-learning-based (ML) failure analysis within a unified structure, enabling both real-time fault detection and predictive maintenance functionalities. The architecture was evaluated through a proof-of-concept implementation using monitoring rules and a previously published machine learning method. The implementation consisted of multiple plant- and inverter-level DTs, combined with a service-oriented architecture. Experimental validation demonstrated that the system achieved real-time anomaly detection and predictive failure analysis. When scaled to larger PV plant configurations (150 plants, each with 15 inverters), the implementation maintained functional stability, with moderate performance degradation that remained within acceptable operational limits. The research contributes a modular and holonic DT architecture that combines rule-based and predictive maintenance within a scalable framework for industrial PV plants. A proof-of-concept implementation validates that the modular service-oriented architecture can support both real-time monitoring and predictive analysis while maintaining architectural flexibility and stability under increased loads. The work provides a generic architecture and methodological foundation that can be adopted by both academia and industry to develop future DT-based maintenance systems, probably beyond PV systems. With further refinement and deployment, such systems could enhance operational reliability and efficiency, supporting the broader transition toward sustainable and data-driven energy management.

Grantor dc:publisher

Stellenbosch : Stellenbosch University

Year dc:date.issued

2026

Author dc:creator

• Swanepoel, Heinrich Marc

Advisor dc:contributor.advisor

• Basson, Anton Herman

Language dc:language.iso

en

Repository record dc:identifier.uri

https://scholar.sun.ac.za/handle/10019.1/135793

OAI identifier oai:identifier

oai:scholar.sun.ac.za:10019.1/135793