Data-driven probabilistic reconstruction of incomplete series extremes for tunnel health monitoring

Health monitoring of tunnels is essential for long-term operational safety. However, continuous signal loss during critical environmental conditions, especially when extremes exceed thresholds, can overlook catastrophic failures. Current data reconstruction methods require large training datasets and struggle with out-of-distribution extremes. Thus, this study proposes a Probabilistic Correlation Mapping approach to reconstruct missing extremes using only a single complete and similar pattern from a neighbouring monitoring point. Pseudo series are generated through correlation mapping between the complete and incomplete series, inferring trends and local variations in the missing segment. Then, Monte Carlo simulation samples parameters for a logistic function to model the temporally non-linear weights for pseudo-series. Effective simulations are screened via a K-means approach considering continuity errors, after which the final reconstruction is obtained by aggregating weighted pseudo series. The approach is first validated using two synthetic cases, demonstrating its effectiveness in reconstructing trend-dominated extremes. Its applicability is further verified through real-world events recorded in the immersed tunnel of the Hong Kong – Zhuhai – Macao Bridge: one under temperature cycles and the other during extreme conditions caused by a super typhoon. Finally, the superiority of the method is comprehensively demonstrated across all cases.