Damage mechanism identification has scientific and practical ramifications for the structural health monitoring,design,and application of composite systems.Recent advances in machine learning uncover pathways to ident...Damage mechanism identification has scientific and practical ramifications for the structural health monitoring,design,and application of composite systems.Recent advances in machine learning uncover pathways to identify the waveform-damage mechanism relationship in higher-dimensional spaces for a comprehensive understanding of damage evolution.This review evaluates the state of the field,beginning with a physics-based understanding of acoustic emission waveform feature extraction,followed by a detailed overview of waveform clustering,labeling,and error analysis strategies.Fundamental requirements for damage mechanism identification in any machine learning framework,including those currently in use,under development,and yet to be explored,are discussed.展开更多
基金CM.and B.S.gratefully acknowledge financial support from the NASA Spce Tochnology Gaduate Research Opportunites Felowship(Grants:8ONSSC19K1164 and 8ONSSC17K0084,SD.and T.MP.gratefully acknowiedge fnanchl support from the Natonal Sclonce Found ation Uward 1984641)patt of the HDR IDEAS Insatute.The authors additonally thank Aaron Engel for the suggeston for this project and Dr Neal Brodnik for an Introduction to tSNE。
文摘Damage mechanism identification has scientific and practical ramifications for the structural health monitoring,design,and application of composite systems.Recent advances in machine learning uncover pathways to identify the waveform-damage mechanism relationship in higher-dimensional spaces for a comprehensive understanding of damage evolution.This review evaluates the state of the field,beginning with a physics-based understanding of acoustic emission waveform feature extraction,followed by a detailed overview of waveform clustering,labeling,and error analysis strategies.Fundamental requirements for damage mechanism identification in any machine learning framework,including those currently in use,under development,and yet to be explored,are discussed.
