Author: Yang, Jianwei
Title: Resolution-lossless ultrasound tomography for health monitoring of composite structures : from nanocomposite sensor network development to machine learning-enabled imaging
Advisors: Su, Zhongqing (ME)
Degree: Ph.D.
Year: 2022
Subject: Structural health monitoring
Ultrasonic imaging
Hong Kong Polytechnic University -- Dissertations
Department: Department of Mechanical Engineering
Pages: xx, 105 pages : color illustrations
Language: English
Abstract: Ultrasound tomography (UT), by virtue of its high accuracy, intuitive presentation of results and standardized implementation, has gained prominence in structural health monitoring (SHM)-driven integrity assurance and health management of composite structures. With an appropriately selected type of sensor, UT for SHM canvasses subtle variation in ultrasound signals by benchmarking pristine counterparts, subsequently associates the variation to material deterioration or damage occurrence, and projects evaluation results in pixelized images through proper tomographic imaging algorithms.
However, a densely configured sensor network, either externally mounted on or internally implanted in composites, alters the microstructure of the fibre-reinforced matrix, influences the interlaminar stress distribution in the sensor vicinity, introduces artificial defect, and consequently lowers the structural load-carrying capacity. On the other hand, any attempt to minimize such intrusion by limiting the sensor number is usually at the cost of undermining the detection resolution and accuracy. Indeed, it is a challenging task to strike a balance between the sensing cost (i.e., the number of sensors) and sensing effectiveness (i.e., the accuracy of tomography) when implementing UT-based SHM.
In this PhD study, an implantable, nanocomposite-inspired, piezoresistive sensor network is developed for implementing UT-based SHM of carbon fibre-reinforced polymer (CFRP) laminates. The nanocomposite ink, formulated with graphene nanosheets (GNSs) and polyvinylpyrrolidone (PVP), is tailored to acquire the percolation threshold of conductive nanofillers. The above ink is then deposited on partially precured B-stage epoxy films using spray deposition process and circuited via highly conductive carbon nanotube fibres (CNT-fibres) as wires, to form a dense sensor network, which is then implanted into CFRP laminates during autoclaving procedure. With a morphologically optimized nano-architecture in nanocomposites, the quantum tunnelling effect can be triggered in percolated networks, which enables the sensors to faithfully response from quasi-static loads to high-frequency guided ultrasonic waves (GUWs). Quasi-static tensile test is performed to gauge possible degradation in tensile properties and change in failure modes of the CFRP laminates owing to the implantation of a sensor network.
Using the developed implantable sensor network, in conjunction with the use of only a handful of surface-mounted PZT wafers as excitation sources, a dense sensor network can be configured, to circumvent the limited-angle problem that conventional UT-based imaging algorithms may have. The implanted sensor network has been proved owing the capability in perceiving GUWs in a broad frequency regime with high precision up to 450 kHz experimentally. The enhanced reconstruction algorithm for the probabilistic inspection of damage (RAPID)-based imaging algorithm, which is revamped by continuously iterating and updating the scale parameter β, presents superior accuracy, compared with the conventional RAPID algorithm when used to evaluate both the location and shape of anomaly, endowing the UT-based SHM with higher imaging resolution while not at the cost of sacrificing the composites' original integrity.
To further achieve real in-situ UT-based SHM and solve the restricted sensing capability due to inadequate sensing paths in the implanted sensor network, a hierarchical, algebraic reconstruction technique (ART) based tomographic imaging approach, facilitated by convolutional neural network (CNN) based machine learning (ML), is developed, targeting resolution-lossless tomography for SHM of composites. The blurry ART images, as the inputs to train a CNN with an encoder-decoder-type architecture, are segmented using convolution and max-pooling to extract defect-modulated image features. The max-unpooling boosts the resolution of ART images with transposed convolution. Trained with the insufficient databases via a mixed numerical and experimental method, the CNN is used to detect and characterize artificial anomaly and delamination in the CFRP laminates. Results demonstrate that the developed approach accurately images artificial anomaly and delamination, in the meantime it minimizes the false alarm by eliminating image artifacts.
In conclusion, starting from mechanism study, through design to fabrication of sensors, new breeds of implantable, nanocomposite-inspired, piezoresistive sensor network is developed. Successful application paradigms in UT of the implanted sensor network, either using the enhanced RAPID or ML-enabled imaging, have accentuated the alluring potentials of in-situ UT-based SHM.
Rights: All rights reserved
Access: open access

Files in This Item:
File Description SizeFormat 
6477.pdfFor All Users4.9 MBAdobe PDFView/Open

Copyright Undertaking

As a bona fide Library user, I declare that:

  1. I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
  2. I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
  3. I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.

By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.

Show full item record

Please use this identifier to cite or link to this item: