Chien-Pu Huang
Researcher, Department of Civil Engineering
National Taiwan University (NTU)
Education
- Ph.D. in Civil Engineering, National Taiwan University
Expertise: Smart Built Environment Operation & Maintenance Automation, Graph-Native Semantic Action Management, Engineering Informatics
Experience
- Postdoctoral Researcher, National Taiwan University: Dedicated to establishing an automated "Data-to-Action" decision-making framework for smart facility management.
NSTC Graduate Students Study Abroad Program (LEAP/千里馬) Visiting Scholar, Department of Computer Science, University of Houston: Specialized in information governance and security architectures for the integration of Building Information Modeling (BIM) and Facility Management (FM) systems.
Licensed Architect (R.O.C. / Taiwan): Possessing practical, hands-on experience in the design development, procurement, contracting, and operation & maintenance (O&M) workflow integration of high-tech manufacturing facilities.
Speech Title and Abstract
An Explainable and Auditable Data-Driven Predictive Maintenance Mechanism: Combining Semantic Modeling with Automated Execution
Predictive Maintenance (PdM) in smart facility management faces key challenges, including the difficulty of rapidly deploying AI models across heterogeneous sites, a lack of explainability in automated decision-making, and the inability to audit executed actions. This study proposes a formalized framework that systematically transforms heterogeneous sensor data into semantic graphs, reasoning queries, executable actions, and provenance logs.
The core of this framework consists of a three-layered mechanism:
Semantic Modeling Layer (Domain Pack): Utilizes graph databases to encapsulate equipment ontologies and maintenance Standard Operating Procedures (SOPs), enabling plug-and-play, rapid cross-factory deployment.
TIAA Automation Engine: Drives a second-level closed-loop decision process through hybrid reasoning that combines machine learning with semantic rules.
PROV-O Provenance Layer: Records the complete decision-making path in accordance with W3C standards, supporting "decision playback" and forming a self-correcting feedback loop.
Validated through HVAC systems and campus carbon management case studies, the framework achieved an 82.5% automated execution rate and saved over 80% in manual data consolidation costs. The empirical results verify that this framework effectively shortens deployment cycles and establishes an explainable, executable, and auditable data-driven management backbone for smart facilities, significantly strengthening operational resilience and sustainability metric attainment.