Voting-Based Fault Detection in Building Management Systems

The patent introduces an intermediary processing layer between raw sensor data and fault diagnosis. This layer includes data preprocessing modules that normalize, filter, and transform temporal data into standardized formats, and feature extraction modules that identify key diagnostic characteristics. This intermediary structure simplifies the analysis complexity while maintaining detection accuracy by breaking down the complex processing task into manageable stages.

The patent replaces traditional mechanical/threshold-based fault detection methods with data-driven analytical approaches. Instead of using fixed thresholds and simple rule-based systems, the invention employs statistical analysis, pattern recognition algorithms, and machine learning techniques to process temporal data. This substitution enables more accurate fault detection by leveraging the computational power to handle data complexity rather than relying on simplified mechanical decision rules.

The patent segments the fault detection system into modular functional components: data acquisition modules from multiple sensors, preprocessing modules for individual data streams, feature extraction modules, diagnosis engines, and output modules. Each module handles specific tasks independently, allowing the system to process real-time data from multiple sources without creating a monolithic complex structure. This segmentation enables scalable implementation where modules can be added or removed based on specific equipment monitoring needs.

The patent implements preliminary data preprocessing and feature extraction before the actual fault diagnosis occurs. Historical data is pre-processed and stored in standardized formats, and baseline operational characteristics are established in advance. When real-time monitoring is needed, the system compares current data against pre-established patterns, significantly reducing the computational burden during critical real-time analysis while maintaining high reliability for early degradation detection.

The patent merges data from multiple sensors and monitoring variables into a unified analysis framework. Instead of analyzing each variable separately, the invention combines temporal data streams into integrated diagnostic models that consider interrelationships between variables. This merging approach maintains complete fault information by preserving correlations between multiple variables while reducing analysis difficulty through unified processing methods and consolidated diagnostic rules.

The patent transforms multi-variable temporal data into a different dimensional representation that simplifies analysis. By applying dimensionality reduction techniques and transforming time-series data into feature spaces, the system maintains the complete information content of multiple variables while presenting it in a form that is easier to analyze. This dimensional transformation allows the system to handle comprehensive monitoring data without proportionally increasing analysis complexity.