Crack Classification of a Pressure Vessel Using Feature Selection and Deep Learning Methods.

Area of Science:

• Engineering
• Materials Science
• Non-destructive Testing

Background:

• Pressure vessels are critical in many industries but pose significant risks if cracks go undetected.
• Early crack detection in pressure vessels is essential for preventing catastrophic failures.
• Acoustic emission (AE) examination is a key non-destructive testing method for monitoring pressure vessel integrity.

Purpose of the Study:

• To develop a robust crack identification technique for pressure vessels.
• To enhance the accuracy of crack detection using acoustic emission data.
• To improve the efficiency of feature selection for deep neural network classification.

Main Methods:

• Hybrid feature extraction from multiple AE sensors across time, frequency, and time-frequency domains.
• Genetic algorithm (GA) with a novel objective function for discriminant feature selection.
• Deep neural network (DNN) classifier trained on selected features for crack identification.

Main Results:

• The proposed GA-based feature selection effectively identified discriminant features from high-dimensional AE data.
• The combined GA + DNN approach achieved a classification accuracy of 94.67% for pressure vessel crack types.
• Demonstrated high effectiveness in selecting features crucial for accurate crack identification.

Conclusions:

• The GA + DNN method provides a robust and accurate approach for crack identification in pressure vessels.
• Effective feature selection is critical for optimizing DNN performance in AE-based structural health monitoring.
• The proposed technique holds significant promise for enhancing the safety and reliability of pressure vessels.