A Statistical Feature and Multiscale Dilated Inception Enhanced Multitask Learning Method for Detecting Multipattern Anomalies of Bridge Strain Monitoring Data
Structural Control and Health Monitoring · 2026-01 · 2026 (1)
DOI: 10.1155/stc/7962455 ↗Kurz gefasst
Fachartikel in der Zeitschrift Structural Control and Health Monitoring, online erschienen am 13.11.2026. Verfasst von Jianfu Lin und 6 weiteren Autor:innen. Die Arbeit nennt 48 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Guangdong Special Support Plan, Guangdong Provincial Pearl River Talents Program, National Natural Science Foundation of China.
Abstract (Originalsprache)
Diverse data anomaly patterns prevalent in large‐scale structural health monitoring datasets frequently induce misinterpretation and generate false alarms during condition assessment. The identification of data anomalies is resource‐intensive and challenging, particularly when multiple anomaly types coexist within a single measurement record. To address this issue, a multipattern anomaly detection‐oriented improved customized gate control (MPAD‐ICGC) method is synergistically designed to detect missing, trend, drift, and their concurrent anomaly patterns. The model’s novelty lies in its statistical feature‐guided task‐specific experts and multiscale dilated inception enhanced shared experts, which are connected through a dynamic gating mechanism with an improved batch‐level Focal loss for optimization. This cohesive architecture enables adaptive knowledge sharing with explicit control, thereby mitigating negative task interference. Specifically, the task‐specific expert network augmented with statistically derived features and an adaptive attention mechanism is constructed to accentuate subtle pattern differences. Meanwhile, the shared multiscale dilated inception expert is incorporated to capture hierarchical features across varied temporal scales, thereby accommodating both abrupt and gradual anomalies. These feature streams are subsequently fused via a customized gate control network, facilitating dynamic and task‐specific representation sharing while suppressing cross‐task interference. Finally, a stability‐driven expert configuration optimization and an improved Focal loss function formulated with batch‐level confidence are introduced to balance model capacity across tasks, ensuring robust training against data imbalance. Extensive experiments conducted on a field‐collected bridge strain dataset demonstrate that the proposed model achieves superior accuracy in detecting both single and concurrent missing, trend, and drift anomalies, outperforming multiple competitive single‐task and multitask models.
Förderung
- Guangdong Special Support Plan
- Guangdong Provincial Pearl River Talents Program
- National Natural Science Foundation of China
Zitieren
APA
Lin, J., Liu, X., Ni, Y., Huang, J., Chen, S., Wang, J., & Wu, W. (2026). A Statistical Feature and Multiscale Dilated Inception Enhanced Multitask Learning Method for Detecting Multipattern Anomalies of Bridge Strain Monitoring Data. Structural Control and Health Monitoring, 2026(1). https://doi.org/10.1155/stc/7962455
Vancouver
Lin J, Liu X, Ni Y, Huang J, Chen S, Wang J, et al. A Statistical Feature and Multiscale Dilated Inception Enhanced Multitask Learning Method for Detecting Multipattern Anomalies of Bridge Strain Monitoring Data. Structural Control and Health Monitoring. 2026;2026(1). doi:10.1155/stc/7962455
BibTeX
@article{Lin2026,
author = {Lin, Jianfu and Liu, Xuanyu and Ni, Yiqing and Huang, Jianliang and Chen, Shihong and Wang, Junfang and Wu, Weilin},
title = {A Statistical Feature and Multiscale Dilated Inception Enhanced Multitask Learning Method for Detecting Multipattern Anomalies of Bridge Strain Monitoring Data},
journal = {Structural Control and Health Monitoring},
year = {2026},
volume = {2026},
number = {1},
doi = {10.1155/stc/7962455}
}
EUVIMED