Biography
Leong Ying Mei is a Senior Lecturer in the Department of Computer Science, Faculty of Information and Communication Technology, Universiti Tunku Abdul Rahman (UTAR), Malaysia. She is currently pursuing her PhD in Computer Science at UTAR, with research focusing on artificial intelligence, machine learning, edge AI, anomaly detection and predictive maintenance for smart building systems. Her current research investigates multi-modal anomaly detection, adaptive and collaborative learning, and resource-efficient AI deployment for HVAC predictive maintenance. She has extensive experience in higher education, teaching, academic administration and student research supervision, with interests spanning intelligent systems, explainable AI, software engineering and applied machine learning.
Abstract
Title: Toward Edge-Intelligent Predictive Maintenance: A Multi-Modal Anomaly Detection Framework for Smart Building HVAC Systems
Abstract: Heating, Ventilation and Air Conditioning (HVAC) systems are critical components of modern smart buildings, yet their complex operating conditions, evolving usage patterns and heterogeneous sensor environments make reliable fault and anomaly detection challenging. Existing artificial intelligence approaches for HVAC predictive maintenance frequently rely on single-source operational data, centralized processing and static learning models, which may limit their ability to adapt to changing operating conditions and resource-constrained deployment environments.
This presentation introduces an edge-based multi-modal anomaly detection framework for intelligent predictive maintenance in smart building HVAC systems. The proposed framework integrates operational time-series data with complementary sensing modalities, including vibration and acoustic information, to provide a more comprehensive representation of equipment behaviour. It further incorporates adaptive learning mechanisms to address concept drift and changing operating conditions, collaborative learning across distributed environments, and hardware-aware optimisation to support efficient deployment on edge computing platforms.
The framework is intended to balance anomaly detection performance with practical deployment requirements such as false-alarm reduction, inference latency, computational efficiency and adaptability. Particular attention is given to the role of continual learning, federated learning and edge AI in enabling scalable and privacy-aware predictive maintenance.
The work highlights current research challenges and presents a systematic direction toward intelligent, adaptive and resource-efficient HVAC monitoring, with broader implications for AI-enabled smart buildings and sustainable infrastructure.
