Zhaolin Yuan

Zhaolin Yuan

Industrial AI researcher

Sitemap

A list of all the posts and pages found on the site. For you robots out there, there is an XML version available for digesting as well.

Pages

Posts

Future Blog Post

less than 1 minute read

Published:

This post will show up by default. To disable scheduling of future posts, edit config.yml and set future: false.

portfolio

publications

Improving Word Representation Quality Trained by word2vec via a More Efficient Hierarchical Clustering Method

Published in In the proceedings of Cooperative Design, Visualization, and Engineering - 15th International Conference, CDVE 2018, 2018

Access paper here

Recommended citation: Zhaolin Yuan, Xiaojuan Ban, Jinlong Hu, "Improving Word Representation Quality Trained by word2vec via a More Efficient Hierarchical Clustering Method." In the proceedings of Cooperative Design, Visualization, and Engineering - 15th International Conference, CDVE 2018, 2018.
Download Paper

An Improved Reinforcement Learning Based Heuristic Dynamic Programming Algorithm for Model-Free Optimal Control

Published in In the proceedings of International Conference on Artificial Neural Networks, ICANN2020, 2020 (CCF-C)

Access paper here

Recommended citation: Jia Li, Zhaolin Yuan, Xiaojuan Ban, "An Improved Reinforcement Learning Based Heuristic Dynamic Programming Algorithm for Model-Free Optimal Control." In the proceedings of International Conference on Artificial Neural Networks, ICANN2020, 2020.
Download Paper

Continuous-Time Prediction of Industrial Paste Thickener System With Differential ODE-Net

Published in IEEE/CAA Journal of Automatica Sinica, 2022 (SCI-1区, IF=19.2)

This study develops a deep-learning, continuous-time predictive framework for thickening systems, aiming to forecast critical outputs such as underflow concentration and mud pressure for optimal process control. The proposed model addresses challenges from non-linearities, long delays, partially observed data, and continuous-time dynamics by integrating a sequential encoder, a state decoder, and a derivative module to learn a deterministic state-space representation. Evaluated on a tailing thickener from FLSmidth equipped with extensive sensors, the model outperforms existing discrete-time approaches, effectively mitigating the impact of long delays and accurately predicting both short- and long-term system behavior.

Recommended citation: Zhaolin Yuan, Xiaorui Li, Di Wu, Xiaojuan Ban, Nai-Qi Wu, Hong-Ning Dai, Hao Wang, "Continuous-Time Prediction of Industrial Paste Thickener System With Differential ODE-Net." IEEE/CAA Journal of Automatica Sinica, 2022.
Download Paper

Integrated three-dimensional visualization and soft-sensing system for underground paste backfilling

Published in Tunnelling and Underground Space Technology, 2022 (SCI-1区)

This study addresses the challenge of accurately monitoring cemented paste backfilling (CPB) processes in harsh underground environments by proposing a 3D visualization and soft-sensing system. Using a GeoSLAM ZEB-HORIZON laser scanner, the system captures detailed 3D models of underground goafs and estimates backfilling height in real time based on accumulated paste volume. The framework further enables web-based simulation and visualization of the filling process, improving operational awareness and decision-making. Field validation in a real copper mine demonstrates that the proposed soft-sensing technique achieves a relative error below 10%, confirming its practical reliability and value for production management in CPB operations.

Recommended citation: Zhaolin Yuan, Xiaojuan Ban, Fangyuan Han, Xingquan Zhang, Shenghua Yin, Yiming Wang, "Integrated three-dimensional visualization and soft-sensing system for underground paste backfilling." Tunnelling and Underground Space Technology, 2022.
Download Paper

ODE-RSSM: Learning Stochastic Recurrent State Space Model from Irregularly Sampled Data

Published in In the proceedings of Conference on Artificial Intelligence, AAAI2023, 2023 (CCF-A)

This work proposes the Ordinary Differential Equation Recurrent State Space Model (ODE-RSSM), a continuous-time deep state space framework designed for nonlinear and stochastic input-output systems with irregularly sampled observations. By integrating an ODE-based network to model latent state evolution between time points, the approach overcomes the limitations of conventional discrete-time SSMs. An efficient reparameterization method is introduced to enable parallel training on batched ODEs with non-uniform time spans, significantly improving computational efficiency. Extensive experiments on multiple datasets, including a private industrial system exhibiting long-term delay and stochasticity, demonstrate superior open-loop prediction performance, highlighting ODE-RSSM’s effectiveness for representation, forecasting, and decision-making in real-world dynamic environments.

Recommended citation: Zhaolin Yuan, Xiaojuan Ban, Zixuan Zhang, Xiaorui Li, Hong{-}Ning Dai, "ODE-RSSM: Learning Stochastic Recurrent State Space Model from Irregularly Sampled Data." In the proceedings of Conference on Artificial Intelligence, AAAI2023, 2023.
Download Paper

Autonomous-Jump-ODENet: Identifying Continuous-Time Jump Systems for Cooling-System Prediction

Published in IEEE Transactions on Industrial Informatics, 2023 (SCI-1区, IF=9.9)

This study presents the Autonomous Jump Ordinary Differential Equation Net (AJ-ODENet), a continuous-time modeling framework designed for complex industrial systems exhibiting periodic jump dynamics. AJ-ODENet combines multiple Hierarchical ODENets to capture stage-specific behaviors from irregularly sampled data, while a stage transition predictor enables autonomous transitions during open-loop simulation. An encoder–decoder implementation of the model is applied to a real data center cooling system, using multivariate inputs such as server power and environmental temperature to simulate runtime behavior. Results show accurate prediction of energy consumption within a 5% relative error, and the model further informs optimal cooling temperature settings, enabling 6%–25% reduction in energy usage.

Recommended citation: Zhaolin Yuan, Yewan Wang, Xiaojuan Ban, Chunyu Ning, Hong-Ning Dai, Hao Wang, "Autonomous-Jump-ODENet: Identifying Continuous-Time Jump Systems for Cooling-System Prediction." IEEE Transactions on Industrial Informatics, 2023.
Download Paper

Chat with MES: LLM-driven user interface for manipulating garment manufacturing system through natural language

Published in Journal of Manufacturing Systems, 2025 (SCI-1区, IF=12.3)

This study introduces Chat with MES (CWM), a conversational interface that replaces traditional manufacturing execution system (MES) graphical user interfaces with natural language interaction. By enabling operators to issue linguistic commands, CWM redefines human–computer interaction in industrial environments. To improve reliability, a multi-step Dynamic Operations Planning and Execution mechanism is proposed, decomposing complex tasks into controllable atomic operations. The framework is evaluated using a newly developed benchmark consisting of 55 crafted requests within a simulated garment manufacturing system. Experimental results show that CWM achieves an 80% task success rate, substantially outperforming GPT-4, which attains a 60% success rate under the same conditions.

Recommended citation: Zhaolin Yuan, Ming Li, Chang Liu, Fangyuan Han, Haolun Huang, Hong-Ning Dai, "Chat with MES: LLM-driven user interface for manipulating garment manufacturing system through natural language." Journal of Manufacturing Systems, 2025.
Download Paper

Controlling Partially Observed Industrial System Based on Offline Reinforcement Learning—A Case Study of Paste Thickener

Published in IEEE Transactions on Industrial Informatics, 2025 (SCI-1区, IF=9.9)

This article proposes an offline-data-driven control strategy for paste thickening systems, addressing the challenges of high process complexity, partial observability, and environmental noise in mineral processing operations. By leveraging offline reinforcement learning, the approach eliminates the risks associated with unsafe online exploration while enabling data-efficient policy training. Operational trajectories are collected using a proportional–integral–derivative controller informed by domain knowledge, forming a reliable offline dataset. To handle constrained observation spaces, a novel algorithm—temporal batch-constrained Q-learning (TBCQ)—is developed for partially observed decision processes. Evaluations in both simulation and a real industrial paste thickener in a copper mine demonstrate improved control performance and significant tracking error reduction.

Recommended citation: Zhaolin Yuan, ZiXuan Zhang, Xiaorui Li, Yunduan Cui, Ming Li, Xiaojuan Ban, "Controlling Partially Observed Industrial System Based on Offline Reinforcement Learning—A Case Study of Paste Thickener." IEEE Transactions on Industrial Informatics, 2025.

NetEventCause: Event-Driven Root Cause Analysis for Large Network System Without Topology

Published in IEEE Transactions on Neural Networks and Learning Systems, 2025 (SCI-1区, IF=10.2)

This article introduces NetEventCause (NEC), an event-driven, unsupervised, and nonintrusive root cause analysis algorithm designed for large-scale private cloud network systems with unknown or incomplete topologies. Addressing the limitations of existing topology-free RCA approaches, NEC learns from historical alarm events to model the occurrence patterns of diverse alarm types using a multivariate neural temporal point process. Leveraging conditional intensity predictions and attribution methods, the algorithm identifies root alarms and reconstructs anomaly transmission chains from cascading alarm events. Extensive evaluations on synthetic data and real-world datasets from the Huawei Shennong Intelligent Maintenance and Operation Center demonstrate NEC’s superior performance in alarm modeling and root cause identification.

Recommended citation: Zhaolin Yuan, Long Ma, Wenjia Wei, Xia Zhu, Mingjie Sun, Duxin Chen, Xiaojuan Ban, "NetEventCause: Event-Driven Root Cause Analysis for Large Network System Without Topology." IEEE Transactions on Neural Networks and Learning Systems, 2025.
Download Paper

Quantification of paste homogeneity by vision-based identification method: Case study for an industrial mixer

Published in Developments in the Built Environment, 2025 (SCI-2区, IF=6.2)

This study develops a vision-based device for real-time monitoring of paste homogeneity during mixing processes. An image segmentation method is proposed to identify non-paste regions and detect areas of insufficient uniformity. To quantitatively assess mixing quality, a non-homogeneity factor is defined, providing an objective measure of paste uniformity. The relationship between blade position and paste homogeneity is further modeled using Gaussian process regression, offering insights into mixing dynamics. The effectiveness of the proposed metric is validated in a paste backfill station, where the results demonstrate strong consistency with engineers’ practical assessments.

Recommended citation: Xiaorui Li, Zhaolin Yuan, Hezheng Wang, Yong Wang, Xiaojuan Ban, "Quantification of paste homogeneity by vision-based identification method: Case study for an industrial mixer." Developments in the Built Environment, 2025.
Download Paper

Shipment Scheduling and Routing Protocols in Cyber-Physical Internet for Prefabricated Construction Modules Logistics

Published in IEEE Transactions on Intelligent Transportation Systems, 2025 (SCI-1区, IF=8.4)

This paper investigates how Cyber-physical Internet (CPI) concepts can transform prefabricated construction logistics by introducing a protocol-based scheduling and routing framework designed to improve module shipment efficiency. Inspired by the logic of digital data transmission, the study draws parallels with internet communication principles to diagnose persistent challenges such as fragmented information sharing, limited coordination, and suboptimal routing decisions. To address these issues, the authors develop a suite of CPI protocols and propose a hierarchical, decentralized decision architecture governing logistics nodes. Numerical experiments and a real-world case study in the Greater Bay Area demonstrate enhanced efficiency, resilience, and collaboration, highlighting CPI’s potential to modernize complex construction logistics networks.

Recommended citation: Zhiyuan Ouyang, Zhaolin Yuan, Qiqi Chen, Ming Li, Zhiheng Zhao, George Huang, "Shipment Scheduling and Routing Protocols in Cyber-Physical Internet for Prefabricated Construction Modules Logistics." IEEE Transactions on Intelligent Transportation Systems, 2025.

talks

teaching