Jianshuang Li

Jianshuang Li

Auto-generated profile, based on publicly available information

Jianshuang Li


Research Focus & Metrics

Academic Career Length

1Year (2026-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
18 September 2026 in SSRN Electronic Journal
Verified
Preprint
Open access
YXYang Xiao
Yang Xiao
HWHuaqing Wang
Huaqing Wang
JLJianshuang Li
Jianshuang Li
DMDongjing Miao
Dongjing Miao
WCWeikang Chen
Weikang Chen
QWQingfeng Wang
Qingfeng Wang
Long baseline multiparameter measurement exhibits pronounced temporal dependence, operating condition variation, and coupling among heterogeneous parameters, making normal fluctuations difficult to distinguish from true anomalies and complicating anomaly localization and source tracing. To address these issues, a digital twin-assisted condition aware contrastive reconstruction method is proposed. First, a condition aware temporal encoding and reconstruction network explicitly incorporates nominal speed, motion direction, and motion stage into normal behavior learning, while contrastive regularization enhances same condition compactness and inter condition separability in the latent space. Then, reconstruction deviation, latent space deviation, and deviation in parameter relationships are fused into a joint anomaly score, and dominant anomalous parameters and corresponding measurement chains are identified through contribution analysis. In addition, physically verified normal samples continuously update the normal behavior baseline through physical-virtual feedback. Experiments were conducted on an 80 m long baseline platform with 63 synchronized measurement parameters and seven representative anomaly types. The proposed method achieves an F1-score of 95.77±1.02% and a false positive rate of 2.39±0.24%, outperforming five published anomaly detection baselines while maintaining stable performance across nine operating speeds and different motion conditions. The average event detection rate reaches 97.35%, with onset-time and onset-position errors of 0.030±0.018 s and 11.82±6.98 mm, respectively. Under gradual distribution drift, feedback updating improves the F1-score from 91.62% to 95.32% and reduces the false positive rate from 5.90% to 2.82%. The inference time is only 4.36 ms per window, demonstrating potential for long-term online anomaly analysis and providing a new approach for long baseline multiparameter measurement.