TY - GEN
T1 - Coupling in situ microstructure observation with machine learning algorithms for damage diagnostics and prognostics
AU - Wisner, Brian J.
AU - Kontsos, Antonios
PY - 2017
Y1 - 2017
N2 - Acoustic Emission (AE) obtained from a novel experimental setup which involves recording of AE data from inside a Scanning Electron Microscope (SEM), is leveraged in this article in a signal processing approach that combines data reduction, classification and outlier analysis to identify data trends and relate them to detection of damage and predictions of remaining useful life at the specimen level. This approach offers the possibility to increase the reliability of NDE methods used in data-driven characterization of material behavior. To demonstrate the benefits of this approach, two types of Aluminum alloys, 2024-T3 and 7075-T651 were mechanically tested inside a SEM while simultaneously recording AE. Data obtained by this approach was then used to develop and validate a data reduction and classification algorithm with the goal to identify sensing information that appears to be the most sensitive to the activation of microstructural-level damage mechanisms. By combining this approach with an outlier analysis it is demonstrated that it is possible to eliminate noise and provide data trends that are found to follow the material degradation process in both monotonic and cyclic conditions. The framework presented can be extended to a variety of materials, while the produced data trends may serve as inputs to computational models.
AB - Acoustic Emission (AE) obtained from a novel experimental setup which involves recording of AE data from inside a Scanning Electron Microscope (SEM), is leveraged in this article in a signal processing approach that combines data reduction, classification and outlier analysis to identify data trends and relate them to detection of damage and predictions of remaining useful life at the specimen level. This approach offers the possibility to increase the reliability of NDE methods used in data-driven characterization of material behavior. To demonstrate the benefits of this approach, two types of Aluminum alloys, 2024-T3 and 7075-T651 were mechanically tested inside a SEM while simultaneously recording AE. Data obtained by this approach was then used to develop and validate a data reduction and classification algorithm with the goal to identify sensing information that appears to be the most sensitive to the activation of microstructural-level damage mechanisms. By combining this approach with an outlier analysis it is demonstrated that it is possible to eliminate noise and provide data trends that are found to follow the material degradation process in both monotonic and cyclic conditions. The framework presented can be extended to a variety of materials, while the produced data trends may serve as inputs to computational models.
UR - https://www.scopus.com/pages/publications/85032444966
UR - https://www.scopus.com/pages/publications/85032444966#tab=citedBy
U2 - 10.12783/shm2017/14007
DO - 10.12783/shm2017/14007
M3 - Conference contribution
T3 - Structural Health Monitoring 2017: Real-Time Material State Awareness and Data-Driven Safety Assurance - Proceedings of the 11th International Workshop on Structural Health Monitoring, IWSHM 2017
SP - 1357
EP - 1365
BT - Structural Health Monitoring 2017
A2 - Chang, Fu-Kuo
A2 - Kopsaftopoulos, Fotis
PB - DEStech Publications
T2 - 11th International Workshop on Structural Health Monitoring 2017: Real-Time Material State Awareness and Data-Driven Safety Assurance, IWSHM 2017
Y2 - 12 September 2017 through 14 September 2017
ER -