Predicting façade deterioration using machine learning approach with drone imagery and microclimate data

Accurate prediction of façade deterioration due to microclimate effects is crucial for sustainable building management and preservation. This paper introduces a methodology for full façade risk assessment using image-based empirical evidence from real buildings. The PAIR methodology—Prepare, Analyze, Integrate, Relate—combines drone imagery, computer vision, weather data, and three-dimensional neighborhood models to create a database that organizes façade deterioration data into sections. This database supports a machine learning model to predict façade deterioration risks. A case study of 16 brick façades in Gothenburg, Sweden, demonstrated the model's strong performance (R 2 = 0.978, MSE = 0.0003) on the test sample. However, performance declined on an excluded validation façade (R 2 = −0.467, MSE = 0.024) due to limited training data and inaccuracies from prior maintenance. Despite these limitations, the methodology provides a computationally efficient alternative to full-scale hygrothermal modeling for assessing deterioration risk across entire façades. • Image and ML-based methodology for moisture risk assessments across facades. • Uses YOLOv4 on UAV images to quantify degradation in facade sections. • Links degradation data with 3D spatial and weather features. • Trains an ML model to predict the extent of degradation across other facades. • Demonstrates the methodology in a case study of 16 brick facades.