Gas Turbine Performance Matching for Unmeasured Parameter Prediction

The patent creates a virtual copy (digital twin) of the gas turbine that replicates its behavior and performance characteristics. This virtual model serves as a surrogate for physical sensors, allowing prediction of unmeasured parameters without adding physical monitoring hardware. The digital twin is trained on historical sensor data and can simulate turbine behavior under various conditions, providing measurement capabilities without the complexity of additional physical sensor systems.

The patent introduces a machine learning model as an intermediary that infers unmeasured parameters from measured ones. Instead of installing sensors for every parameter of interest, the system uses the ML model to act as a mediator that calculates missing information based on correlations learned from training data. This approach provides comprehensive parameter coverage while avoiding the complexity of installing numerous sensors.

The patent transforms the complex physics-based model into a machine learning model that captures essential relationships in a computationally efficient form. By changing the representation from explicit physics equations to learned parameter relationships, the system maintains prediction accuracy for unmeasured parameters while reducing computational complexity. The ML model learns effective parameters and their relationships from training data, providing accurate predictions without the burden of solving complex differential equations in real-time.