TY - GEN
T1 - AN INFORMATION THEORY APPROACH FOR INTERNET OF THINGS ENABLED DAMAGE MONITORING
AU - Malik, Sarah
AU - Kontsos, Antonios
N1 - Publisher Copyright:
Copyright © 2022 by ASME.
PY - 2022
Y1 - 2022
N2 - Nondestructive Evaluation (NDE) techniques such as acoustic emission, digital image correlation, infrared thermography, ultrasonic testing among others, have been used to monitor evolutionary failure processes, related to Structural Health Monitoring (SHM). Although such NDE sensing assists in understanding evolving material and structural states, machine learning methods capable of using real-time NDE data enable additional insights related to diagnostics and prognostics. In this context, this investigation presents a novel approach to enable the real time use of NDE data for damage detection. To achieve this goal, an Internet of Things (IoT) framework developed by the authors is used in conjunction with NDE datasets for near real-time diagnostics of crack initiation at the laboratory scale. Compact-tension specimens of an aerospace-grade aluminum alloy were used in accordance with ASTM standards. Acoustic emission NDE datasets were acquired and were subsequently used in an in-house built, scalable IoT system capable of edge and cloud computing with the purpose to reliably identify crack initiation. Once the signals are transmitted from the edge to the fog node, the trained information-entropy based model determines if the raw data is indicative of crack. The model used during live testing is trained offline on the Cloud. The model characterizes the disorder in signals using Shannon's Entropy to ultimately determine the amount of information per signal. Then a statistical model is used to characterize such information. During the testing process, the signals are segmented based on file chunks to allow for real-time transmission. The main innovation of this approach is the fact that a combination of hardware, computing and machine learning analysis proves to be advantageous in implementing a data structure that can be used at the edge and which can successfully flag the incubation and subsequent initiation of fracture. The IoT system described can be applied to a variety of test setups and at various length scales. Extensions of this work to include forecasting are also discussed.
AB - Nondestructive Evaluation (NDE) techniques such as acoustic emission, digital image correlation, infrared thermography, ultrasonic testing among others, have been used to monitor evolutionary failure processes, related to Structural Health Monitoring (SHM). Although such NDE sensing assists in understanding evolving material and structural states, machine learning methods capable of using real-time NDE data enable additional insights related to diagnostics and prognostics. In this context, this investigation presents a novel approach to enable the real time use of NDE data for damage detection. To achieve this goal, an Internet of Things (IoT) framework developed by the authors is used in conjunction with NDE datasets for near real-time diagnostics of crack initiation at the laboratory scale. Compact-tension specimens of an aerospace-grade aluminum alloy were used in accordance with ASTM standards. Acoustic emission NDE datasets were acquired and were subsequently used in an in-house built, scalable IoT system capable of edge and cloud computing with the purpose to reliably identify crack initiation. Once the signals are transmitted from the edge to the fog node, the trained information-entropy based model determines if the raw data is indicative of crack. The model used during live testing is trained offline on the Cloud. The model characterizes the disorder in signals using Shannon's Entropy to ultimately determine the amount of information per signal. Then a statistical model is used to characterize such information. During the testing process, the signals are segmented based on file chunks to allow for real-time transmission. The main innovation of this approach is the fact that a combination of hardware, computing and machine learning analysis proves to be advantageous in implementing a data structure that can be used at the edge and which can successfully flag the incubation and subsequent initiation of fracture. The IoT system described can be applied to a variety of test setups and at various length scales. Extensions of this work to include forecasting are also discussed.
UR - http://www.scopus.com/inward/record.url?scp=85143132061&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85143132061&partnerID=8YFLogxK
U2 - 10.1115/SMASIS2022-91119
DO - 10.1115/SMASIS2022-91119
M3 - Conference contribution
AN - SCOPUS:85143132061
T3 - Proceedings of ASME 2022 Conference on Smart Materials, Adaptive Structures and Intelligent Systems, SMASIS 2022
BT - Proceedings of ASME 2022 Conference on Smart Materials, Adaptive Structures and Intelligent Systems, SMASIS 2022
PB - American Society of Mechanical Engineers
T2 - ASME 2022 Conference on Smart Materials, Adaptive Structures and Intelligent Systems, SMASIS 2022
Y2 - 12 September 2022 through 14 September 2022
ER -