Real-time health assessment of timber buildings using continuous monitoring data by online unsupervised anomaly detection

The continuous monitoring of timber buildings provides valuable information for assessing their structural integrity under varying environmental and operational conditions. Offline unsupervised anomaly detection techniques can evaluate the health and integrity of these buildings at periodic intervals, but they face several limitations, including delayed detection, limited adaptability to newly arriving data, and the need for complete retraining whenever environmental or structural conditions change. Moreover, offline strategies often fail to cap-ture short-term transient anomalies or gradual drifts in dynamic behavior that occur between analysis intervals. To overcome these challenges, this study presents a real-time health as-sessment framework for timber buildings based on online unsupervised anomaly detection using continuously updated monitoring and environmental data. The central component of the proposed strategy is unsupervised incremental learning, in which an initially trained anomaly detector with limited data is progressively updated as new data becomes available. The approach performs streaming analysis of modal features derived from acceleration measurements, where adaptive statistical and machine learning models automatically identi-fy deviations from normal dynamic behavior. The proposed framework allows building own-ers to continuously monitor the structural condition of their buildings, receive early warn-ings of abnormal behavior, and make timely maintenance or safety decisions without inter-rupting ongoing monitoring. The framework is validated using long-term modal frequency data from a timber building instrumented with monitoring systems. Results demonstrate the ability of the proposed approach to continuously assess the condition of timber buildings us-ing only vibration-based features.