AI Agent-Driven Intelligent Catalog Framework: A Governance-Centered Approach for Cleaning and Normalization of...

Area of Science:

• Data Science
• Industrial Internet of Things (IIoT)

Background:

• Industrial Internet of Things (IIoT) generates massive heterogeneous sensor data, complicating cleaning and normalization.
• Existing algorithm-centric methods address data quality issues in isolation, lacking unified governance.

Purpose of the Study:

• Propose a novel governance-centered framework for multi-source industrial sensor data management.
• Establish semantic standardization and dynamic orchestration for data cleaning and normalization.

Main Methods:

• Introduced an Intelligent Catalog for semantic governance and metadata standardization.
• Developed an AI Agent-driven mechanism for dynamic orchestration of cleaning and normalization strategies.
• Modular integration of classical algorithms (PCA, KPCA, LSTM) without model dependency.

Main Results:

• Framework significantly outperforms baseline methods in normalization consistency.
• Demonstrated enhanced noise robustness and stability across heterogeneous IIoT datasets.
• Achieved semantic alignment before numerical processing.

Conclusions:

• The governance-centered paradigm offers a scalable and adaptive solution for complex industrial sensor data.
• Shift from algorithm-centric to governance-centric approach enhances data management.
• Framework provides unified governance for multi-source industrial sensor data.