Multi-Scale Spatiotemporal Feature Fusion and Deep Generative Modeling for Industrial Process Monitoring (基于多尺度时空特征融合与深度生成模型的工业过程监测)
Chair: 
Jiusun Zeng, Hangzhou Normal University, China
Co-chairs: 
Le Yao, Hangzhou Normal University, China 
Zheren Zhu, Hangzhou Normal University, China
Keywords: | Topics: |
· Spatiotemporal Feature Fusion (时空特征融合) · Distributed Process Monitoring (分布式过程监测) · Deep Generative Models (深度生成模型) · Safety-Critical Control (安全关键控制) · Edge Machine Learning (边缘机器学习) · Causal Digraph Analysis (因果图分析) | · Spatiotemporal Modeling for Industrial Process Monitoring (基于时空建模工业过程监测) · Multiscale Spatiotemporal Learning for Predictive Maintenance (基于多尺度时空学习的预测性维护) · Edge AI-based models for Real-Time Fault Detection and Classification (基于边缘AI模型的实时故障检测与分类) · Deep Generative Modeling for Imbalanced Industrial Process Data (面向不平衡工业过程数据的深度生成模型) · Safety and Robustness Enhancement for Industrial Modeling (工业模型的安全与鲁棒性提升) |
Summary:
· This special session focuses on cutting-edge methodologies for enhancing the reliability, safety, and efficiency of modern industrial processes. With the increasing complexity of industrial systems, the integration of multi-scale spatiotemporal feature fusion, distributed modeling, and deep probabilistic generative models has become critical for fault diagnosis, predictive maintenance, and safety control. Key topics include: 1) Leveraging edge machine learning for real-time fault detection and classification, achieving high accuracy. 2) Integrating spatial-temporal dependencies from heterogeneous data to improve feature extraction in dynamic industrial scenarios. 3) Collaborative frameworks decompose complex processes into sub-blocks, enabling localized monitoring while preserving global interactions. 4) Data augmentation and incomplete data recovery by deep generative models. 5) Adaptive control strategies and anomaly detection systems, combined with process performance analysis.
· 本次专题研讨会聚焦于提升现代工业过程可靠性、安全性及效率的前沿方法。随着工业系统复杂性的日益增加，多尺度时空特征融合、分布式建模与深度概率生成模型的集成技术，已成为实现故障诊断、预测性维护及安全控制的关键。核心议题包括：1）基于分布式边缘机器学习的实时故障检测与分类；2）基于异构数据的时空依赖性建模，提升动态工业场景下的特征提取能力；3）基于协同学习框架分解复杂流程为子模块，实现局部监测与全局交互的平衡；4）利用深度生成模型进行数据增强与缺失数据恢复；5）基于过程性能分析的自适应安全控制策略与异常检测系统。