A Method for Detecting the Randomness of Barkhausen Noise in a Material Fatigue Test Using Sensitivity and...

Area of Science:

• Materials Science
• Non-Destructive Evaluation
• Signal Processing

Background:

• Magnetic Barkhausen noise (MBN) signals are crucial for assessing ferromagnetic material microstructure and degradation.
• Current methods lack quantitative definitions for MBN signal stochasticity, hindering reproducible feature evaluation.
• Existing approaches struggle with feature reproducibility and establishing uniform performance standards.

Purpose of the Study:

• To quantitatively define the stochastic characteristics of MBN signals.
• To transform MBN signal analysis into quantifying signal uncertainty and sensitivity for improved fatigue state prediction.
• To address both feature (observation) and model uncertainties in prediction models.

Main Methods:

• Developed a confidence interval sensitivity analysis to define and quantify feature uncertainty from probability distributions.
• Proposed a prior approximation method to handle parameter uncertainty in prediction models.
• Incorporated informed priors by optimizing Kullback-Leibler divergence to improve prediction accuracy with insufficient data.

Main Results:

• The confidence interval sensitivity analysis quantitatively determines MBN signal instability and reduces feature dispersion.
• Feature uncertainty quantification leads to positive additive effects, reducing the false prediction rate to near zero.
• Informed priors effectively measure model parameter uncertainties and perform comparably to maximum likelihood estimation (MLE).

Conclusions:

• The proposed methods provide a quantitative framework for analyzing MBN signal stochasticity and uncertainty.
• Accurate fatigue state prediction is achievable even with limited data through informed priors.
• While parameter uncertainty is addressed, its direct propagation to improve prediction uncertainty requires further investigation.