Multivariable Fault Diagnosis Using Principal Component Baselines

The patent segments the complex monitoring task into distinct phases: a learning phase where normal behavior patterns are captured and stored as reference data, and an operating phase where deviations from these patterns are detected. This segmentation allows the system to handle complexity by separating data collection from analysis, enabling comprehensive monitoring through structured temporal phases rather than attempting simultaneous processing of all sensor data.

The patent performs preliminary action by conducting a learning phase before the actual monitoring operation. During this phase, the system collects and stores reference data representing normal component behavior under various conditions. This preliminary data preparation enables the subsequent operating phase to focus solely on detecting deviations from the established baseline, simplifying the real-time monitoring complexity while maintaining comprehensive fault detection capability.

The patent introduces an intermediary reference data structure that mediates between the simple comparison operation and the complex multivariate sensor data. The reference data, established during the learning phase, serves as an intermediate representation of normal behavior that encapsulates complex relationships between multiple measured quantities. This intermediary allows simple deviation detection operations to achieve high measurement precision by comparing against a comprehensive reference model rather than requiring complex real-time analysis.

The patent changes the parameter representation by transforming raw sensor measurements into a reference-based deviation metric. Instead of directly comparing multiple measured quantities with complex relationships, the system transforms the data into deviations from predefined reference values established during the learning phase. This parameter transformation maintains ease of operation through simple deviation calculations while achieving high measurement precision by incorporating comprehensive normal behavior patterns into the reference data.

The patent implements self-service by enabling the system to automatically generate its own reference data during the learning phase without requiring manual engineer intervention. Instead of relying on engineers to create standard curves, the system autonomously collects data, identifies normal behavior patterns, and stores reference values. This self-service approach maintains high measurement precision through accurate reference data while achieving full automation by eliminating the need for manual curve creation and maintenance measure generation.

The patent substitutes the mechanical/engineering process of creating and maintaining standard curves with an automated data-driven approach. Instead of engineers manually establishing reference relationships between variables, the system uses automated learning phase data collection and pattern recognition to generate reference data. This substitution replaces the manual mechanical process with automated computational methods, maintaining measurement precision while enabling automatic maintenance measure generation through rule-based systems that operate on the automated reference data.