Unsupervised Cross-Domain Soft Sensor Modelling via Deep Physics-Inspired Particle Flow Bayes
Data-driven soft sensors are essential for achieving accurate perception through reliable state inference. However, developing representative soft sensor models is challenged by issues such as missing labels, domain adaptability, and temporal coherence in data. To address these challenges, we propose a deep Particle Flow Bayes (DPFB) framework for cross-domain soft sensor modeling in the absence of target state labels. In particular, a sequential Bayes objective is first formulated to perform the maximum likelihood estimation underlying the cross-domain soft sensing problem. At the core of the framework, we incorporate a physics-inspired particle flow that optimizes the sequential Bayes objective to perform an exact Bayes update of the model extracted latent and hidden features. As a result, these contributions enable the proposed framework to learn a rich approximate posterior feature representation capable of characterizing complex cross-domain system dynamics and performing effective time series unsupervised domain adaptation (UDA). Finally, we validate the framework on a complex industrial multiphase flow process system with complex dynamics and multiple operating conditions. The results demonstrate that the DPFB framework achieves superior cross-domain soft sensing performance, outperforming state-of-the-art deep UDA and normalizing flow approaches.
Code (0)
등록된 구현이 없습니다.
Tasks
Domain AdaptationMissing LabelsSensor ModelingTime SeriesUnsupervised Domain AdaptationSimilar Papers 제목 키워드 기반
KANS: Knowledge Discovery Graph Attention Network for Soft Sensing in Multivariate Industrial Processes
Soft sensing of hard-to-measure variables is often crucial in industrial processes. Current practices rely heavily on conventional modeling techniques that show success in improving accuracy. However, they overlook the n…
Graph AttentionGraph structure learningRepresentation LearningSEA++: Multi-Graph-based High-Order Sensor Alignment for Multivariate Time-Series Unsupervised Domain Adaptation
Unsupervised Domain Adaptation (UDA) methods have been successful in reducing label dependency by minimizing the domain discrepancy between a labeled source domain and an unlabeled target domain. However, these methods f…
Domain AdaptationTime SeriesUnsupervised Domain AdaptationSensorimotor learning for artificial body perception
Artificial self-perception is the machine ability to perceive its own body, i.e., the mastery of modal and intermodal contingencies of performing an action with a specific sensors/actuators body configuration. In other w…
Cross-domain Transfer Learning and State Inference for Soft Robots via a Semi-supervised Sequential Variational Bayes Framework
Recently, data-driven models such as deep neural networks have shown to be promising tools for modelling and state inference in soft robots. However, voluminous amounts of data are necessary for deep models to perform ef…
Transfer LearningUnsupervised Person Re-identification by Soft Multilabel Learning
Although unsupervised person re-identification (RE-ID) has drawn increasing research attentions due to its potential to address the scalability problem of supervised RE-ID models, it is very challenging to learn discrimi…
Person Re-IdentificationUnsupervised Person Re-Identification