Reduction in the Sensor Effect on Acoustic Emission Data to Create a Generalizable Library by Data Merging.

Area of Science:

• Materials Science
• Non-destructive Testing
• Signal Processing

Background:

• Acoustic emission (AE) is a powerful non-destructive testing technique.
• Sensor variability significantly affects AE signal interpretation and data reproducibility.
• Standardizing AE data acquisition is crucial for reliable analysis and database development.

Purpose of the Study:

• To analyze the influence of different sensors on acoustic emission signatures.
• To develop a methodology for mitigating sensor-induced effects in AE measurements.
• To enable the creation of generalized AE signature libraries for machine learning applications.

Main Methods:

• Controlled AE experiments using pencil lead breaks on PMMA plates.
• Comparison of various AE transducers for plate wave reproduction.
• Application of Principal Component Analysis (PCA) and Z-score normalization for data processing.
• Utilizing Kruskal-Wallis test for statistical analysis and outlier identification.

Main Results:

• Different AE sensors exhibit distinct responses, leading to variations in AE descriptors and test results.
• The proposed methodology effectively reduces sensor effects, yielding a common descriptor set across all sensors.
• Z-score normalization and outlier identification are key to achieving consistent AE data distributions.

Conclusions:

• Sensor selection and data processing significantly influence AE signature analysis.
• The developed procedure standardizes AE data, facilitating the merging of descriptors into a unified library.
• This work paves the way for generalized AE signature libraries and machine learning-based AE source classification.