Predictive Fault Indicators for Physical System Diagnosis

The patent creates virtual copies of sensor measurements through predictive models. Instead of installing multiple physical sensors (hardware redundancy), the system uses machine learning models to generate predicted measurements that serve as virtual replicas. These predicted values are then compared with actual sensor readings to detect faults, achieving redundancy detection capability without the physical overhead of multiple sensors.

The patent replaces the mechanical approach of hardware redundancy (multiple physical sensors) with an information-processing approach using predictive models. The system substitutes physical sensor multiplication with computational modeling, where machine learning algorithms generate expected sensor readings based on system state predictions, and fault detection is achieved through comparison rather than through additional hardware.

The patent transitions from requiring precise mathematical equations (rigid parameter specification) to using data-driven predictive models with learned parameters. Instead of needing known physical equations, the system trains machine learning models on operational data to learn the relationships between system states and sensor measurements. This changes the approach from equation-based analysis to model-based prediction, where parameters are learned rather than predetermined.

The patent substitutes the traditional approach of using known mathematical equations for analytical redundancy with machine learning-based predictive models. The system replaces equation-driven measurement analysis with data-driven prediction, where neural networks or other ML models learn the complex relationships in the system without requiring explicit mathematical formulations of the underlying physics.