Predictive-SHM is a lightweight, open-source toolkit for multi-sensor structural health monitoring and multi-step time-series forecasting. It bridges the gap between advanced prediction methods and practical deployment, where integrating heterogeneous sensing streams, replacing models, and serving operator-facing workflows is often costly. The software delivers an end-to-end pipeline covering multi-source ingestion, unified preprocessing, pluggable prediction, configuration-driven visualization, and residual- and threshold-based alerting. Models are integrated via metadata-driven registration and adapters that map a unified logical time-series view (Universal Logical Data Model, ULDM) to model-specific tensors and return standardized, timestamped forecasts. Although adapter-based plugin designs and dashboard-style abstractions are well-established patterns, this work's emphasis is their coherent realization for SHM with a reference Transformer–CNN implementation. Stability and reproducibility are demonstrated through a case study on the Yuhuang Pavilion historic building. By open-sourcing Predictive-SHM, we aims to lower the barrier from research prototypes to field-ready SHM practice and to enable transparent model interchange within a single stack.
Yang et al. (Mon,) studied this question.