Turbine Diagnostic Feature Selection for Faster Failure Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for selecting turbine engine parameters associated with failure events are time-consuming and inefficient, leading to prolonged downtime and increased operational costs due to delayed maintenance.
Innovation Solution
A turbine diagnostic feature selection system utilizing sensors, a raw parameter analyzer, model builder, clustering, interdependency, ranking, and sorting components to automate the process of selecting relevant parameters based on Akaike Information Criterion (AIC), Area under Curve (AUC), and p-values, enabling rapid identification of failure causes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If current manual methods are used for selecting turbine engine parameters, then parameter selection can be performed with simple tools, but the time required for diagnosis is excessively long
Solution Approach 1:
The patent replaces manual mechanical analysis methods with an automated computer-based system that uses sensors, processors, and software algorithms to select diagnostic parameters. The system automatically performs data collection, preprocessing, clustering, and parameter selection without manual intervention, thereby dramatically reducing diagnosis time while accepting increased system complexity.
Solution Approach 2:
The system enables self-service diagnostic capabilities by automatically selecting relevant parameters and generating diagnostic recommendations without requiring expert manual analysis. The automated feature selection process and ranking algorithms allow the system to independently identify critical parameters and provide maintenance guidance, reducing reliance on human expertise for routine diagnostics.
2Measurement precision
If a comprehensive set of parameters is monitored to ensure accurate failure diagnosis, then diagnostic accuracy is improved, but the complexity of data processing increases
Solution Approach 1:
The system extracts and selects only the most relevant diagnostic parameters from a comprehensive set of available sensor data. Through automated feature selection and ranking processes, the system identifies and extracts critical parameters that are most strongly associated with failure events, discarding redundant information. This maintains high diagnostic accuracy while reducing processing complexity by focusing on essential parameters only.
Solution Approach 2:
The patent segments the comprehensive parameter set into distinct groups using clustering algorithms that identify relationships among parameters. By segmenting parameters into meaningful clusters based on their interdependencies and relevance to failure modes, the system simplifies the analysis of large datasets while preserving diagnostic accuracy through structured organization of information.
3Productivity
If manual parameter selection processes are used, then system implementation is simpler, but maintenance costs and operational delays increase
Solution Approach 1:
The patent replaces manual parameter selection and diagnostic processes with automated computer-based systems that use sensors, processors, and algorithms to rapidly identify failure causes. This substitution dramatically improves maintenance productivity by reducing diagnosis time and enabling faster decision-making, though it requires implementation of complex automated systems with multiple components including data acquisition, preprocessing, clustering, and ranking modules.
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
A turbine diagnostic machine learning system builds one or more turbine engine performance models using one or more parameter or parameter characteristics. A model of turbine engine performance includes ranked parameters or parameter characteristics, the ranking of which is calculated by a model builder based upon a function of AIC, AUC and p-value, resulting in a corresponding importance rank. These raw parameters and raw parameter characteristics are then sorted according to their importance rank, and selected by a selection component to form one or more completed models. The one or more models are operatively coupled to one or more other models to facilitate further machine learning capabilities by the system.