{"id":{"repo_id":"wichita-thes","oai_identifier":"oai:soar.wichita.edu:10057/30355"},"canonical_url":"https://search.dev.ndltd.org/etd/wichita-thes/oai:soar.wichita.edu:10057/30355","repository":{"repo_id":"wichita-thes","name":"Wichita State University","base_url":"https://soar.wichita.edu/oai/request"},"display":{"title":"Detection of acoustically matched defects in fiber-reinforced composites using through-transmission ultrasound","abstract":"Detecting acoustically matched defects (AMD) in fiber-reinforced composites (FRC) remains a challenge for traditional through-transmission ultrasound (TTU) techniques, which rely on manual thresholding and lack time-of-flight data. This limitation creates a noticeable research gap that this dissertation directly addresses through advanced classification methods. First, a novel root-mean-square (RMS) classifier leverages the frequency-domain transformations and confidence intervals to establish robust thresholds. Compared with amplitude, the RMS classifier improved the detection accuracy by 21.34% and increased the signal-to-noise ratio (SNR) to 7.01. Second, we trained neural networks (NN) on domain-specific features extracted from the TTU waveform signals. Globally trained models achieved Macro-F1 scores of 0.96 (time-domain), 0.97 (frequency-domain), and 0.62 (wavelet-domain). A comparative analysis in Chapter 4, using regional heat maps and probability of detection (POD) curves, further demonstrated the superiority of both the RMS and NN classifiers over amplitude classifiers. The RMS and regionally trained NN classifiers consistently achieved regional recall rates exceeding 94% across all ply counts and defect depths, with regions achieving a 100% recall. Notably, the wavelet-domain model significantly improved when trained regionally (Macro-F1 score of 0.94). The regional Macro-F1 scores achieved 28–30% relative to the baseline amplitude classifier, highlighting their robustness in detecting AMDs in laminate regions, where traditional methods often fail. Together, these contributions provide a validated framework for detecting AMDs in FRC laminates using TTU, supporting more reliable automated defect detection and advancing nondestructive evaluation (NDE) capabilities in high-performance composite manufacturing.","abstract_html":"Detecting acoustically matched defects (AMD) in fiber-reinforced composites (FRC) remains a challenge for traditional through-transmission ultrasound (TTU) techniques, which rely on manual thresholding and lack time-of-flight data. This limitation creates a noticeable research gap that this dissertation directly addresses through advanced classification methods. First, a novel root-mean-square (RMS) classifier leverages the frequency-domain transformations and confidence intervals to establish robust thresholds. Compared with amplitude, the RMS classifier improved the detection accuracy by 21.34% and increased the signal-to-noise ratio (SNR) to 7.01. Second, we trained neural networks (NN) on domain-specific features extracted from the TTU waveform signals. Globally trained models achieved Macro-F1 scores of 0.96 (time-domain), 0.97 (frequency-domain), and 0.62 (wavelet-domain). A comparative analysis in Chapter 4, using regional heat maps and probability of detection (POD) curves, further demonstrated the superiority of both the RMS and NN classifiers over amplitude classifiers. The RMS and regionally trained NN classifiers consistently achieved regional recall rates exceeding 94% across all ply counts and defect depths, with regions achieving a 100% recall. Notably, the wavelet-domain model significantly improved when trained regionally (Macro-F1 score of 0.94). The regional Macro-F1 scores achieved 28–30% relative to the baseline amplitude classifier, highlighting their robustness in detecting AMDs in laminate regions, where traditional methods often fail. Together, these contributions provide a validated framework for detecting AMDs in FRC laminates using TTU, supporting more reliable automated defect detection and advancing nondestructive evaluation (NDE) capabilities in high-performance composite manufacturing.","abstract_has_math":false,"creators":["LeMay, Gary S."],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025-07","date_published":"2025-07","updated_at":"2026-07-24T06:05:36Z","subjects":[],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["hdl:10057/30355"],"render_values":[{"text":"hdl:10057/30355","href":null,"code":true}]}]},"links":{"outbound_url":null,"outbound_label":null,"outbound_source":null},"metadata_groups":[{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.issued","label":"Date","values":["2025-07"]},{"key":"dc:type","label":"Dc Type","values":["Dissertation"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["hdl:10057/30355"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.other","label":"Dc Description Other","values":["Detecting acoustically matched defects (AMD) in fiber-reinforced composites (FRC) remains a challenge for traditional through-transmission ultrasound (TTU) techniques, which rely on manual thresholding and lack time-of-flight data. This limitation creates a noticeable research gap that this dissertation directly addresses through advanced classification methods. First, a novel root-mean-square (RMS) classifier leverages the frequency-domain transformations and confidence intervals to establish robust thresholds. Compared with amplitude, the RMS classifier improved the detection accuracy by 21.34% and increased the signal-to-noise ratio (SNR) to 7.01. Second, we trained neural networks (NN) on domain-specific features extracted from the TTU waveform signals. Globally trained models achieved Macro-F1 scores of 0.96 (time-domain), 0.97 (frequency-domain), and 0.62 (wavelet-domain). A comparative analysis in Chapter 4, using regional heat maps and probability of detection (POD) curves, further demonstrated the superiority of both the RMS and NN classifiers over amplitude classifiers. The RMS and regionally trained NN classifiers consistently achieved regional recall rates exceeding 94% across all ply counts and defect depths, with regions achieving a 100% recall. Notably, the wavelet-domain model significantly improved when trained regionally (Macro-F1 score of 0.94). The regional Macro-F1 scores achieved 28–30% relative to the baseline amplitude classifier, highlighting their robustness in detecting AMDs in laminate regions, where traditional methods often fail. Together, these contributions provide a validated framework for detecting AMDs in FRC laminates using TTU, supporting more reliable automated defect detection and advancing nondestructive evaluation (NDE) capabilities in high-performance composite manufacturing."]},{"key":"dc:title","label":"Title","values":["Detection of acoustically matched defects in fiber-reinforced composites using through-transmission ultrasound"]}]}],"canonical_facts":{"dc:date.issued":["2025-07"],"dc:description.other":["Detecting acoustically matched defects (AMD) in fiber-reinforced composites (FRC) remains a challenge for traditional through-transmission ultrasound (TTU) techniques, which rely on manual thresholding and lack time-of-flight data. This limitation creates a noticeable research gap that this dissertation directly addresses through advanced classification methods. First, a novel root-mean-square (RMS) classifier leverages the frequency-domain transformations and confidence intervals to establish robust thresholds. Compared with amplitude, the RMS classifier improved the detection accuracy by 21.34% and increased the signal-to-noise ratio (SNR) to 7.01. Second, we trained neural networks (NN) on domain-specific features extracted from the TTU waveform signals. Globally trained models achieved Macro-F1 scores of 0.96 (time-domain), 0.97 (frequency-domain), and 0.62 (wavelet-domain). A comparative analysis in Chapter 4, using regional heat maps and probability of detection (POD) curves, further demonstrated the superiority of both the RMS and NN classifiers over amplitude classifiers. The RMS and regionally trained NN classifiers consistently achieved regional recall rates exceeding 94% across all ply counts and defect depths, with regions achieving a 100% recall. Notably, the wavelet-domain model significantly improved when trained regionally (Macro-F1 score of 0.94). The regional Macro-F1 scores achieved 28–30% relative to the baseline amplitude classifier, highlighting their robustness in detecting AMDs in laminate regions, where traditional methods often fail. Together, these contributions provide a validated framework for detecting AMDs in FRC laminates using TTU, supporting more reliable automated defect detection and advancing nondestructive evaluation (NDE) capabilities in high-performance composite manufacturing."],"dc:identifier":["hdl:10057/30355"],"dc:title":["Detection of acoustically matched defects in fiber-reinforced composites using through-transmission ultrasound"],"dc:type":["Dissertation"]},"updated_at":"2026-07-24T06:05:36Z"}