STFM: Time-Frequency Stockwell Transform and Flow Matching for Data-Augmented Soft Sensor Modeling
IEEE Sensors Journal · 2026-11 · 26 (21) · 32793-32801
DOI: 10.1109/jsen.2026.3732333 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Sensors Journal, 2026-11. Verfasst von Hwa Hui Tew und 6 weiteren Autor:innen. Die Arbeit nennt 36 Literaturangaben.
Abstract (Originalsprache)
Soft sensor modeling aims to estimate hard-to-measure process variables, yet its predictive performance is often challenged by spectral nonstationarity and the costly, time-consuming nature of labeled measurements in modern industrial settings. Generative diffusion modeling alleviates data scarcity, but reliance on iterative sampling steps restricts a streamlined data augmentation strategy. In light of this, we develop a soft sensor framework that enables data augmentation via the modeling of spectral–temporal dynamics of sensor time-series (TS) through the Stockwell transform (ST) and a variant of flow matching (FM) models. In particular, the Stockwell domain leverages adaptive windows that preserve smooth and complex spectral–temporal interactions in the time-frequency representation, while FM enables few-step sampling with high generative fidelity. The synthetic samples are then added on top of the limited original samples for evaluation of the three real-world soft sensor datasets. Experimental results validate that the proposed STFM surpasses existing state-of-the-art TS generative models, achieving superior fidelity and predictive accuracy for industrial soft sensing.
Zitieren
APA
Tew, H. H., Chang, X. C., Hoh, K., Loo, J. Y., Ding, Z. Y., Tan, C. P., & Ting, C. (2026). STFM: Time-Frequency Stockwell Transform and Flow Matching for Data-Augmented Soft Sensor Modeling. IEEE Sensors Journal, 26(21), 32793-32801. https://doi.org/10.1109/jsen.2026.3732333
Vancouver
Tew HH, Chang XC, Hoh K, Loo JY, Ding ZY, Tan CP, et al. STFM: Time-Frequency Stockwell Transform and Flow Matching for Data-Augmented Soft Sensor Modeling. IEEE Sensors Journal. 2026;26(21):32793-32801. doi:10.1109/jsen.2026.3732333
BibTeX
@article{Tew2026,
author = {Tew, Hwa Hui and Chang, Xian Cheng and Hoh, Kar-Ee and Loo, Junn Yong and Ding, Ze Yang and Tan, Chee Pin and Ting, Chee-Ming},
title = {STFM: Time-Frequency Stockwell Transform and Flow Matching for Data-Augmented Soft Sensor Modeling},
journal = {IEEE Sensors Journal},
year = {2026},
volume = {26},
number = {21},
pages = {32793--32801},
doi = {10.1109/jsen.2026.3732333}
}
EUVIMED