PCA-Based Technical System Diagnosis for Multisensor Fault Detection

The patent transforms the monitoring approach by changing from simple univariate metrics to multivariate statistical analysis. It uses principal component analysis (PCA) to transform multiple sensor parameters into a reduced set of independent principal components, enabling comprehensive monitoring while maintaining computational simplicity. The method monitors multiple variables simultaneously (temperature, pressure, flow rates, etc.) and their interrelationships, rather than analyzing each parameter in isolation.

The patent creates a virtual model (digital twin) of the technical system's normal operating state through statistical learning during an adjustment phase. This reference model captures the complex relationships between multiple sensors and driving behaviors, allowing the system to automatically detect deviations without requiring explicit knowledge of all operational dependencies.

The patent applies principal component analysis to transform the original high-dimensional sensor data space into a lower-dimensional space of independent principal components. This dimensionality reduction maintains the essential information for fault detection while significantly reducing computational complexity. The method automatically identifies the most significant patterns in the data without requiring manual selection of which sensors to monitor.

The patent develops a universal monitoring framework that can handle multiple types of sensors (temperature, pressure, flow, vibration) and various operating conditions through a single statistical model. The PCA-based approach provides a unified method for analyzing diverse sensor data and detecting different types of faults, eliminating the need for separate monitoring systems for different parameter types.