Thermographic Digital Twin for Proactive Fire-Risk Prediction in Legacy Residential Electrical Installations: A Systematic Review and Transferability Framework for Ageing Housing and Buildings in Peru

Technical Summary: Thermographic Digital Twin for Proactive Fire-Risk Prediction in Legacy Residential Electrical Installations

Problem Context and Research Gap
Residential fires of electrical origin pose a significant public safety risk, particularly in dwellings with ageing infrastructure. In high-income contexts like the United Kingdom, many housing stocks predate modern overload safety regulations (e.g., BS 7671), leaving them vulnerable to undetected hot spots in distribution panels and splices. In middle-income contexts like Peru, the risk is compounded by the coexistence of formal, ageing dwellings and informal settlements with unregulated electrical connections. Current mitigation strategies are largely reactive, relying on periodic inspections or alarms that activate only after a fault manifests or a fire has started.

While infrared thermography combined with machine learning has demonstrated high classification accuracy (79.8%–94%) in industrial settings and power substations, and digital twins are increasingly used for urban-scale fire management, a critical gap remains. The literature lacks studies that explicitly integrate: (a) 3D thermographic modelling of legacy residential electrical installations, (b) multi-day-ahead hot-spot prediction via machine learning, and (c) a validated protocol for transferring these technologies to the regulatory, climatic, and material conditions of middle-income countries like Peru.

Methodology
This work employs a systematic literature review following PRISMA 2020 guidelines, adapted for a technological object of study. The search strategy queried databases including Scopus, ScienceDirect, MDPI, Nature/Springer, and specialized disaster-risk repositories using Boolean operators combining "digital twin," "infrared thermography," "electrical fault/fire," and "residential/housing."

• Eligibility: The review included peer-reviewed articles published between 2017 and 2026 with verified DOIs. Sources were excluded if they lacked scientific backing, focused exclusively on wildfires without electrical infrastructure components, or were promotional.
• Analysis: Thirteen verified sources were analyzed across three axes: thermography/ML for electrical diagnosis, digital twins in infrastructure, and fire risk in legacy/informal housing. Quantitative performance data (accuracy, sensitivity) were extracted where available.
• Conceptual Development: Based on the synthesis, the authors developed a conceptual architecture and a transferability framework, specifying recommended tools such as handheld thermographic cameras (320x240 pixels), BIM software for 3D modelling, cloud storage for time-series data, and gradient-boosting machine learning libraries (e.g., XGBoost).

Key Contributions and Proposed Architecture
The study makes three primary contributions:

1. Systematic Synthesis: A quantitative comparison matrix of thirteen studies, highlighting that while industrial thermography models achieve up to 94% accuracy, they do not address the specific lead-time or residential context required for proactive fire prevention.
2. Conceptual Architecture: A four-layer predictive thermographic digital-twin architecture designed specifically for legacy residential installations:
   • Physical Layer: The actual residential electrical installation and on-site thermographic inspection.
   • Data Layer: Integration of 3D geometric models (BIM) and thermal/electrical time series into a cloud database.
   • Predictive Layer: A continuously updated digital twin using supervised machine learning to classify risk against critical temperature thresholds.
   • Decision Layer: Translation of risk classification into proactive alerts and corrective maintenance recommendations.
3. Transferability Framework: An explicit protocol for adapting the architecture to Peru. This framework adjusts for four specific factors: higher ambient temperatures in Lima, increased domestic electrical loads (e.g., electric stoves), generally lower wiring material quality in informal/low-cost installations, and the specific regulatory framework (Código Nacional de Electricidad-Utilización) rather than BS 7671.

Results
The synthesis confirms that the individual components of the proposed system (thermography, 3D modelling, gradient-boosting ML) have evidence of favorable performance in related domains. However, the review identified no existing source that combines these elements specifically for proactive fire-risk prediction in legacy residential electrical installations with a transferability protocol for a middle-income country. The proposed architecture is positioned as an original conceptual contribution rather than a replication of an empirically validated system in this exact domain.

Significance and Limitations
The paper concludes that available evidence supports the technical feasibility of the proposed approach. The significance lies in integrating thermography, machine learning, and digital twin concepts into a single framework tailored for ageing housing, offering a methodological basis for future empirical validation in the UK and Metropolitan Lima.

The authors maintain a modest stance regarding the current state of the work:

• The proposed architecture is a conceptual synthesis and has not been empirically validated in the field by the authors.
• Specific quantitative performance metrics for legacy residential installations in the UK or Peru remain to be established through subsequent primary studies.
• The approach is primarily applicable to multi-family buildings with accessible distribution panels or single-family dwellings with documented access; extending it to informal dwellings without accessible panels is identified as a line for future research.

This work serves as a foundational step, providing the technical and methodological basis necessary to guide the next stage of research: empirical field validation.