Fuzzy Multi-SVR Learning Model for Reliability-Based Design Optimization of Turbine Blades

Area of Science:

• Mechanical Engineering
• Computational Science
• Artificial Intelligence

Background:

• Reliability-based design optimization (RBDO) is crucial for multi-failure turbine blades in power systems.
• Existing models often struggle with the complexity and multi-failure modes of turbine blades.
• Integrating machine learning offers potential for improved RBDO accuracy and efficiency.

Purpose of the Study:

• To propose a novel machine learning-based RBDO approach for multi-failure turbine blades.
• To enhance the precision and efficiency of RBDO by incorporating fuzzy theory and support vector regression (SVR).
• To validate the effectiveness of the proposed method in optimizing turbine blade design.

Main Methods:

• Developed a fuzzy multi-SVR learning method combining fuzzy theory, SVR, and multi-response surface methods.
• Optimized SVR model parameters using the artificial bee colony algorithm.
• Employed a multi-objective genetic algorithm to solve the RBDO model and procedure, considering fuzzy constraints.

Main Results:

• Achieved a significant reduction in turbine blade stress (92.38 MPa) and deformation (0.09838 mm).
• Improved the comprehensive reliability degree by 3.45%, from 95.4% to 98.85%.
• Demonstrated the method's workability for complex structures, offering high modeling precision, optimization efficiency, and accuracy.

Conclusions:

• The fuzzy multi-SVR learning method provides a robust and accurate approach for the RBDO of multi-failure turbine blades.
• This study establishes a new research direction for machine learning-based RBDO in complex structural designs.
• The findings expand the application of machine learning in mechanical reliability design, enriching existing theories and methods.