Gas Turbine Compressor Operability Prediction Using Damage-Aware ML
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current computational fluid dynamic (CFD) techniques for predicting the impact of damage on gas turbine engine compressor operability are time-consuming, labor-intensive, and not sufficiently accurate, leading to inaccurate predictions and unnecessary scrapping of components.
Innovation Solution
A computer-implemented method using machine learning algorithms, specifically artificial neural networks, to analyze compressor damage data and predict operability by training with quantified damage parameters, optimizing the algorithms through cross-validation and importance determination of damage parameters.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If computational fluid dynamic (CFD) techniques are used to model the effect of damage on compressor operability, then detailed analysis can be performed, but the process becomes time consuming and labor intensive
Solution Approach 1:
The patent creates a simplified copy or surrogate model of the complex CFD simulation process using machine learning algorithms. This surrogate model replicates the predictive capabilities of detailed CFD analysis but executes much faster, trading off some computational complexity for speed while maintaining acceptable accuracy for operational decisions
Solution Approach 2:
The patent replaces the mechanical CFD simulation process with a computational machine learning model. Instead of solving complex fluid dynamics equations through iterative numerical methods, the system uses trained neural networks or other ML algorithms to directly predict compressor operability from damage inputs, eliminating the time-consuming simulation steps
2Productivity
If accurate predictions of blade damage impact are made quickly, then components can be sentenced efficiently, but insufficient accuracy leads to unnecessary scrapping
Solution Approach 1:
The patent implements a dynamic prediction system that adapts its analysis depth and methodology based on the specific damage characteristics and operational context. The machine learning model can adjust its confidence levels and request additional analysis only when necessary, providing fast predictions for clear cases while maintaining accuracy for borderline situations
Solution Approach 2:
The patent transforms the prediction approach by changing from fixed-threshold decision making to probabilistic predictions with confidence intervals. The machine learning model outputs not just binary pass/fail decisions but also uncertainty measures, allowing operators to make informed decisions about when fast predictions are sufficient and when additional verification is needed
3Measurement precision
If CFD techniques are used to model large separations caused by blunt blades, then detailed flow analysis is achieved, but the current industry CFD methods are beyond sufficient accuracy
Solution Approach 1:
The patent extracts the critical predictive capability from the complex CFD methodology and isolates it into a standalone machine learning model. Instead of attempting to improve or simplify the full CFD process, the system captures only the essential relationship between damage characteristics and stall point prediction, eliminating unnecessary computational complexity while retaining the core predictive function
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A computer-implemented method comprising: controlling input of at least a portion of a first training data set into a first machine learning algorithm, the first training data set including: data quantifying damage to one or more components of a first gas turbine engine; and data quantifying a first operating parameter of the first gas turbine engine; receiving data quantifying the first operating parameter as an output of the first machine learning algorithm; and training the first machine learning algorithm using: the received data output from the first machine learning algorithm; and data quantifying the first operating parameter of the first gas turbine engine, the trained first machine learning algorithm being configured to enable determination of operability of a second gas turbine engine.