Special Session-9

Special Session Organizers: 

Prof. Zhuyun Chen, School of Electromechanical Engineering, Guangdong University of Technology, China

Assoc. Prof. Guo Yang, School of Electromechanical Engineering, Guangdong University of Technology, China

Assoc. Prof. Chong Chen, School of Computer Science, Guangdong University of Technology, China

Assoc. Prof. Ji Tan, School of Automation, Guangdong University of Technology, China

Introduction and Topics:

Conventional intelligent maintenance methods, though effective in specific scenarios, struggle with limited generalization, poor interpretability, and heavy reliance on labeled fault data. As industrial systems become more complex and operate under variable conditions, a more adaptive and knowledge-driven paradigm is urgently needed.

Industrial large models pre-trained on massive, multi-source operational data, offer a transformative solution. By capturing deep temporal and cross-modal dependencies and integrating domain knowledge via prompting or fine-tuning, these models enable zero/few-shot fault recognition, explainable anomaly reasoning, and generative decision support. When combined with digital twins, they facilitate autonomous adaptation to emerging failure modes without manual re-engineering. Beyond equipment health monitoring, these models also excel in visual defect detection, e.g., surface cracks, casting flaws, weld imperfections, by leveraging vision-language understanding and few-shot adaptation, bridging the gap between physical degradation and product quality.

Nevertheless, key challenges remain: efficient pre-training with limited annotations, real-time edge deployment, physical knowledge embedding to avoid hallucinations, and trustworthy verification for safety-critical assets. This session invites cutting-edge research on industrial large models for intelligent O&M, aiming to turn industrial big data into actionable, autonomous asset management.

Topics of interest include, but are not limited to:

• Pre-training and continual learning of industrial foundation models
• Zero/few-shot fault diagnosis and anomaly detection
• Large model–driven visual defect detection (surface, structural, assembly defects)
• Generative reasoning for root cause analysis and maintenance decisions
• Digital twin–large model fusion for adaptive life prediction
• Lightweight edge-cloud deployment and real-time inference
• Trustworthy, explainable large models for critical assets and quality control