Logo image
Back
A novel information entropy approach for crack monitoring leveraging nondestructive evaluation sensing
Journal article   Open access   Peer reviewed

A novel information entropy approach for crack monitoring leveraging nondestructive evaluation sensing

Sarah Malik and Antonios Kontsos
Mechanical systems and signal processing, v 214, 111207
15 May 2024
DOI: https://doi.org/10.1016/j.ymssp.2024.111207
url
https://doi.org/10.1016/j.ymssp.2024.111207View
Published, Version of Record (VoR) Open

Abstract

Engineering, Mechanical Science & Technology Engineering Technology
The accurate detection of crack activity is crucial for both material performance evaluation and structural damage assessment. Recent developments have found Shannon ' s Information Entropy method advantageous for characterizing damage in materials and applicable to real-time damage detection. This manuscript introduces a novel approach to crack detection that leverages Nondestructive Evaluation (NDE) datasets. Specifically, it introduces a discretization parameter to represent signal distributions and leverage information entropy to identify the onset of crack initiation. This approach enhances the reliability of material diagnostics by being applicable to multi -modal datasets thus improving the generalizability of the method. Moreover, by using raw sensing information and integrating information entropy with outlier detection, this approach reduces computational burden for real-time applications related to Mode I loading. To demonstrate this, lab experiments were conducted using compact -tension specimens of aluminum alloy and Acoustic Emission signals were collected to demonstrate the validity of the presented approach. Additionally, to investigate an extension of the approach, it was adopted to the case of image datasets from cyclic loading experiments, showcasing its potential for use with different NDE sensing methods and complex loading. The results were verified and compared with other signal processing methods such as machine learning. Furthermore, the approach ' s potential for real-time monitoring systems is also discussed.

Metrics

1 Record Views
22 citations in Web of Science

Details

Logo image