In chemical processes, missing data is a significant challenge that often impedes the development of reliable soft sensors. Existing imputation approaches generally lack supervision during data reconstruction and tend to overlook the critical relationships between process variables and key quality variables. To address these limitations, this paper proposes a quality-aware imputation method for incomplete time-series data. The proposed framework comprises two main components. First, a diffusion-based model guided by a conditional score criterion is designed to recover missing process data in a self-supervised manner. Second, a quality-aware learner is introduced to enable supervised imputation that jointly accounts for the interplay between process and quality variables, particularly under scenarios with sparse quality measurements. Furthermore, a first-order difference evaluation strategy is incorporated to quantitatively assess the temporal consistency and plausibility of the imputed data. Finally, the recovered data are utilized to train a soft sensor model for accurate online quality prediction. The proposed method demonstrates superior prediction performance in both a numerical example and a real chemical process.
- Quality-aware industrial data imputation with self-supervised recovery for process soft sensor development
- Yun Dai - Zhejiang University of TechnologyChao Yang - Northeastern Univ, State Key Lab Synthet Automat Proc Ind, Shenyang 110819, Peoples R ChinaKaixin Liu - North Univ China, Shanxi Key Lab Signal Capturing & Proc, Taiyuan 030051, Peoples R ChinaYi Liu - Zhejiang University of TechnologyYuan Yao - National Tsing Hua University, Department of Chemical Engineering
- Elsevier
- 14
- U23A20328 / National Natural Science Foundation of China; National Natural Science Foundation of China (NSFC) RFC2022002 / Fundamental Research Funds for the Provincial Universities of Zhejiang NSTC 113-2221-E-007-012-MY2 / National Science and Technology Council, Taiwan MS2024017 / Nantong Social Livelihood Science and Technology Project
- Journal article
- 01/12/2025
- Computers & chemical engineering, Vol.203, 109328
- English