An Online Data-Driven Fault Diagnosis Method for Air Handling Units by Rule and Convolutional Neural Networks

Maintaining consistent performance in air handling units remains a significant challenge for modern building management. Operators often struggle to identify subtle malfunctions before they impact overall energy efficiency. Prior research has shown that traditional monitoring techniques frequently fail to distinguish between sensor errors and complex mechanical failures. That uncertainty drove the development of more sophisticated diagnostic frameworks. No prior work had resolved the trade-off between high diagnostic precision and the need for rapid, real-time feedback. Existing approaches often rely on static thresholds that cannot adapt to changing operational environments. This gap motivated the exploration of hybrid models that integrate expert knowledge with automated pattern recognition. The current study addresses these limitations by proposing a dual-layered diagnostic architecture.

The study aims to develop an online, data-driven method for diagnosing faults in air handling units within HVAC systems. Researchers sought to resolve the limitations of existing monitoring tools that often struggle with complex mechanical failures. They focused on creating a framework that balances high diagnostic accuracy with the necessity for rapid, real-time operational feedback. The motivation stems from the need to extend the service life of building climate control equipment through proactive maintenance. By integrating rule-based logic with advanced machine learning, the authors intended to create a more versatile diagnostic tool. This project addresses the specific challenge of identifying both simple sensor errors and intricate system malfunctions simultaneously. The team aimed to validate this combined approach using historical data from building management systems. Ultimately, they sought to demonstrate that this hybrid architecture outperforms traditional, single-method diagnostic strategies in practical applications.

The review approach involves a hybrid design that integrates rule-based logic with deep learning architectures. Researchers first established threshold values derived from operational experience to perform initial sensor condition monitoring. They then implemented 1D convolutional neural networks to automate the extraction of diagnostic features from historical system logs. This design utilizes data gathered directly from existing building management infrastructure to train the classification model. The team validated the framework by comparing its performance against standard diagnostic benchmarks in a laboratory setting. They conducted offline testing to measure classification precision and performed online simulations to evaluate real-time processing capabilities. This methodology emphasizes the synthesis of expert-defined rules and automated pattern recognition to address diverse fault types. The study concludes by confirming the efficacy of this combined approach through empirical validation on a physical unit.

The hybrid RACNN framework achieved an identification accuracy of 99.15% during offline testing phases. This high level of precision indicates that the model effectively distinguishes between various fault categories. Online detection capabilities were confirmed to operate within a two-minute timeframe, meeting requirements for rapid system response. The experimental results show that the proposed method significantly improves diagnostic performance compared to traditional techniques. By combining rule-based filtering with convolutional neural networks, the system successfully identified both simple sensor errors and complex mechanical issues. These findings demonstrate that the integration of these two distinct methodologies provides a robust solution for AHU monitoring. The validation on a real-world system confirms the practical utility of the proposed diagnostic architecture. The data suggests that this approach offers a reliable pathway for enhancing maintenance efficiency in HVAC systems.

The authors demonstrate that integrating rule-based logic with deep learning significantly enhances diagnostic reliability. This hybrid framework successfully identifies both simple sensor errors and intricate system faults. Their findings suggest that combining these methodologies provides a more robust solution than using either approach in isolation. The researchers propose that this dual-layered strategy is suitable for real-time deployment in commercial building environments. Experimental validation confirms that the model achieves high precision while maintaining rapid response times. These results imply that data-driven techniques can effectively reduce maintenance overhead in large-scale infrastructure. The study highlights the potential for automated systems to improve the longevity of climate control equipment. Future applications may focus on scaling this architecture across diverse building management platforms.