De-identified Multi-Sensor Tunnel Monitoring Data and Weak-Supervision Outputs
收藏资源简介:
This record contains the de-identified dataset and reproducibility materials supporting the manuscript “Auditable Weak Supervision for Anomaly Analysis of Sparse Multi-Sensor Tunnel Monitoring Data”. The package includes observations from 44 displacement, rock-bolt-stress, and pore-pressure channels; daily aligned records; 8,713 audited analytical windows; numerical weak-label provenance; structured non-sensitive engineering-evidence annotations; baseline and sequence-model outputs; model configurations; reproducibility scripts; a data dictionary; a file manifest; and manuscript figures. Of the analytical windows, 4,728 meet the observation and split requirements used for modelling. To protect project confidentiality, direct project and organization identifiers, exact locations, source workbook and worksheet labels, and report prose were removed. Sensors, zones, and sections use stable public identifiers. All research dates and date-times were shifted by one constant undisclosed offset, preserving temporal order and intervals. Private source records, engineering reports, and the re-identification mapping are not included. The reported scores evaluate agreement with weak proxy labels and do not represent field-validated diagnostic performance. The data and documentation are licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0). The software under code/ is licensed under the MIT License.



