SmartDATA Lab Publications
2026
203.
Kang Yang, Zekun Yang, Zhihui Tian, Harsha Tetali, Joel B Harley
In: Lecture Notes in Computer Science, pp. 31–39, Springer Nature Switzerland, Cham, 2026.
@incollection{Yang2026-es,
title = {Weight decay optimized unsupervised autoencoder based anomaly
detection in uncontrolled dynamic structural health monitoring},
author = {Kang Yang and Zekun Yang and Zhihui Tian and Harsha Tetali and Joel B Harley},
url = {http://dx.doi.org/10.1007/978-3-031-94895-4_3},
doi = {10.1007/978-3-031-94895-4_3},
year = {2026},
date = {2026-01-01},
booktitle = {Lecture Notes in Computer Science},
pages = {31–39},
publisher = {Springer Nature Switzerland},
address = {Cham},
series = {Lecture notes in computer science},
abstract = {Current autoencoder-based unsupervised anomaly detection in
Dynamic Data Driven Applications Systems (DDDAS) typically
depends on a comprehensive collection of historical normal
signals as training data. However, such unsupervised models
struggle to perform effectively under dynamic conditions when
there is a significant divergence between the environmental
conditions of the evaluation data and those of the training data.
To address this, our previous study introduces an unsupervised
anomaly detection method trained solely on the current evaluation
data, eliminating the dependency on historical data and thus more
applicable in DDDAS. This method utilizes the inherent bias
learning property of neural networks, which typically prioritizes
learning from larger classes (normal signals) at the expense of
smaller ones. However, a crucial limitation of this new
autoencoder-based damage detection method is its performance
degradation when the evaluation data includes an increasing
number of abnormal signals. To enhance anomaly detection under
these conditions, an optimal weight decay regularization strategy
is provided in this study to limit the autoencoder’s ability to
learn abnormal signals. The efficacy of the novel
autoencoder-based anomaly detection method in DDDAS is validated
using guided waves gathered from a structural health monitoring
system under uncontrolled and dynamically varying environmental
conditions.},
keywords = {},
pubstate = {published},
tppubtype = {incollection}
}
Current autoencoder-based unsupervised anomaly detection in
Dynamic Data Driven Applications Systems (DDDAS) typically
depends on a comprehensive collection of historical normal
signals as training data. However, such unsupervised models
struggle to perform effectively under dynamic conditions when
there is a significant divergence between the environmental
conditions of the evaluation data and those of the training data.
To address this, our previous study introduces an unsupervised
anomaly detection method trained solely on the current evaluation
data, eliminating the dependency on historical data and thus more
applicable in DDDAS. This method utilizes the inherent bias
learning property of neural networks, which typically prioritizes
learning from larger classes (normal signals) at the expense of
smaller ones. However, a crucial limitation of this new
autoencoder-based damage detection method is its performance
degradation when the evaluation data includes an increasing
number of abnormal signals. To enhance anomaly detection under
these conditions, an optimal weight decay regularization strategy
is provided in this study to limit the autoencoder’s ability to
learn abnormal signals. The efficacy of the novel
autoencoder-based anomaly detection method in DDDAS is validated
using guided waves gathered from a structural health monitoring
system under uncontrolled and dynamically varying environmental
conditions.
Dynamic Data Driven Applications Systems (DDDAS) typically
depends on a comprehensive collection of historical normal
signals as training data. However, such unsupervised models
struggle to perform effectively under dynamic conditions when
there is a significant divergence between the environmental
conditions of the evaluation data and those of the training data.
To address this, our previous study introduces an unsupervised
anomaly detection method trained solely on the current evaluation
data, eliminating the dependency on historical data and thus more
applicable in DDDAS. This method utilizes the inherent bias
learning property of neural networks, which typically prioritizes
learning from larger classes (normal signals) at the expense of
smaller ones. However, a crucial limitation of this new
autoencoder-based damage detection method is its performance
degradation when the evaluation data includes an increasing
number of abnormal signals. To enhance anomaly detection under
these conditions, an optimal weight decay regularization strategy
is provided in this study to limit the autoencoder’s ability to
learn abnormal signals. The efficacy of the novel
autoencoder-based anomaly detection method in DDDAS is validated
using guided waves gathered from a structural health monitoring
system under uncontrolled and dynamically varying environmental
conditions.
2025
202.
Ghatu Subhash, Michael MacIsaac, Charlie Tran, Amanda Beck, Woohyun Eum, Joel B Harley
Interpreting material anisotropy through the fractional wave equation Journal Article
In: Ultrasonics, no. 107866, pp. 107866, 2025.
@article{Subhash2025-kb,
title = {Interpreting material anisotropy through the fractional wave
equation},
author = {Ghatu Subhash and Michael MacIsaac and Charlie Tran and Amanda Beck and Woohyun Eum and Joel B Harley},
url = {http://dx.doi.org/10.1016/j.ultras.2025.107866},
doi = {10.1016/j.ultras.2025.107866},
year = {2025},
date = {2025-10-01},
journal = {Ultrasonics},
number = {107866},
pages = {107866},
publisher = {Elsevier BV},
abstract = {In the fields of structural health monitoring and non-destructive
evaluation (NDE), guided waves are often used for detection of
material and structur…},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
In the fields of structural health monitoring and non-destructive
evaluation (NDE), guided waves are often used for detection of
material and structur…
evaluation (NDE), guided waves are often used for detection of
material and structur…
201.
J Harley, Amanda Beck, Woohyun Eum, Michael Macissac, Charlie Tran, G Subhash
Physics-informed filtering for ultrasonic guided wave structural health monitoring Journal Article
In: Proceedings of the 15th International Workshop on Structural Health Monitoring, 2025.
@article{Harley2025-st,
title = {Physics-informed filtering for ultrasonic guided wave structural
health monitoring},
author = {J Harley and Amanda Beck and Woohyun Eum and Michael Macissac and Charlie Tran and G Subhash},
url = {http://dx.doi.org/10.12783/shm2025/37487},
doi = {10.12783/shm2025/37487},
year = {2025},
date = {2025-09-01},
journal = {Proceedings of the 15th International Workshop on Structural
Health Monitoring},
abstract = {This paper presents a methodology for physics-informed filtering,
applied to ultrasonic guided waves. The physics-informed filters
are derived by integrating physicsbased constraints into an
optimization framework, where the cost function comprises of two
components: a reconstruction error term and a regularization term.
The reconstruction error term ensures the filtered data closely
aligns with the original measurements while the regularization
term enforces physical laws represented by a linear operator. We
derive a general closed-form solution for these filters and then
apply them with the two-dimensional Helmholtz equation. We apply
these filters, using various parameters, to ultrasonic guided wave
data collected by a laser Doppler vibrometer. We demonstrate the
ability to characterize damage regions with thickness losses from
the filtered data.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
This paper presents a methodology for physics-informed filtering,
applied to ultrasonic guided waves. The physics-informed filters
are derived by integrating physicsbased constraints into an
optimization framework, where the cost function comprises of two
components: a reconstruction error term and a regularization term.
The reconstruction error term ensures the filtered data closely
aligns with the original measurements while the regularization
term enforces physical laws represented by a linear operator. We
derive a general closed-form solution for these filters and then
apply them with the two-dimensional Helmholtz equation. We apply
these filters, using various parameters, to ultrasonic guided wave
data collected by a laser Doppler vibrometer. We demonstrate the
ability to characterize damage regions with thickness losses from
the filtered data.
applied to ultrasonic guided waves. The physics-informed filters
are derived by integrating physicsbased constraints into an
optimization framework, where the cost function comprises of two
components: a reconstruction error term and a regularization term.
The reconstruction error term ensures the filtered data closely
aligns with the original measurements while the regularization
term enforces physical laws represented by a linear operator. We
derive a general closed-form solution for these filters and then
apply them with the two-dimensional Helmholtz equation. We apply
these filters, using various parameters, to ultrasonic guided wave
data collected by a laser Doppler vibrometer. We demonstrate the
ability to characterize damage regions with thickness losses from
the filtered data.
200.
Kang Yang, Zhenhan Lin, Zekun Yang, Zhihui Tian, Jie Ma, José C Príncipe, Joel B Harley
Improved PCA Reconstruction-Based Unsupervised Anomaly Detection in Uncontrolled Structural Health Monitoring With Correntropy Journal Article
In: IEEE Transactions on Industrial Informatics, 2025.
@article{Yang2025-mm,
title = {Improved PCA Reconstruction-Based Unsupervised Anomaly
Detection in Uncontrolled Structural Health Monitoring With
Correntropy},
author = {Kang Yang and Zhenhan Lin and Zekun Yang and Zhihui Tian and Jie Ma and José C Príncipe and Joel B Harley},
url = {https://ieeexplore.ieee.org/abstract/document/11078390/},
year = {2025},
date = {2025-07-01},
journal = {IEEE Transactions on Industrial Informatics},
publisher = {IEEE},
abstract = {Guided wave-based structural health monitoring is extensively
utilized in various industrial applications to ensure the
integrity of components within industrial systems. Among these
monitoring techniques, principal component analysis (PCA)
reconstruction methods are widely used for anomaly detection due
to their computational efficiency and interoperability. However,
existing PCA reconstruction methods are semisupervised anomaly
detection approaches that require training on historical normal
data and fail to detect anomalous signals within the training
set. To address this limitation, this work proposes a correntropy
PCA (C-PCA), enabling fully unsupervised anomaly detection on raw
training data without requiring label information, when the
dataset contains a high proportion of abnormal signals. This
method allows anomaly detection on real-time measurements without
the need for precleaned historical …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Guided wave-based structural health monitoring is extensively
utilized in various industrial applications to ensure the
integrity of components within industrial systems. Among these
monitoring techniques, principal component analysis (PCA)
reconstruction methods are widely used for anomaly detection due
to their computational efficiency and interoperability. However,
existing PCA reconstruction methods are semisupervised anomaly
detection approaches that require training on historical normal
data and fail to detect anomalous signals within the training
set. To address this limitation, this work proposes a correntropy
PCA (C-PCA), enabling fully unsupervised anomaly detection on raw
training data without requiring label information, when the
dataset contains a high proportion of abnormal signals. This
method allows anomaly detection on real-time measurements without
the need for precleaned historical …
utilized in various industrial applications to ensure the
integrity of components within industrial systems. Among these
monitoring techniques, principal component analysis (PCA)
reconstruction methods are widely used for anomaly detection due
to their computational efficiency and interoperability. However,
existing PCA reconstruction methods are semisupervised anomaly
detection approaches that require training on historical normal
data and fail to detect anomalous signals within the training
set. To address this limitation, this work proposes a correntropy
PCA (C-PCA), enabling fully unsupervised anomaly detection on raw
training data without requiring label information, when the
dataset contains a high proportion of abnormal signals. This
method allows anomaly detection on real-time measurements without
the need for precleaned historical …
199.
Yi Wang, Zipeng Xu, Vivekanand Muralikrishnan, Joel B Harley, Michael R Tonks, Gregory S Rohrer, Amanda R Krause
4D Observations of the initiation of abnormal grain growth in commercially pure Ni Journal Article
In: Scripta materialia, vol. 264, no. 116715, pp. 116715, 2025.
@article{Wang2025-qz,
title = {4D Observations of the initiation of abnormal grain growth in
commercially pure Ni},
author = {Yi Wang and Zipeng Xu and Vivekanand Muralikrishnan and Joel B Harley and Michael R Tonks and Gregory S Rohrer and Amanda R Krause},
url = {https://www.sciencedirect.com/science/article/pii/S1359646225001782},
doi = {10.1016/j.scriptamat.2025.116715},
year = {2025},
date = {2025-07-01},
urldate = {2025-07-01},
journal = {Scripta materialia},
volume = {264},
number = {116715},
pages = {116715},
publisher = {Elsevier BV},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
198.
Kang Yang, Zekun Yang, Hanbo Yang, Junkai Zhou, Zhongzheng Ren Zhang, Linyuan Wang, Zhihui Tian, Sungwon Kim, J Harley
Dataset on guided waves from long-term structural health monitoring under uncontrolled and dynamic conditions Miscellaneous
2025.
@misc{Yang2025-oq,
title = {Dataset on guided waves from long-term structural health
monitoring under uncontrolled and dynamic conditions},
author = {Kang Yang and Zekun Yang and Hanbo Yang and Junkai Zhou and Zhongzheng Ren Zhang and Linyuan Wang and Zhihui Tian and Sungwon Kim and J Harley},
url = {http://dx.doi.org/10.1038/s41597-025-05300-5},
doi = {10.1038/s41597-025-05300-5},
year = {2025},
date = {2025-06-01},
urldate = {2025-06-01},
publisher = {Nature Publishing Group},
abstract = {Few studies address guided wave structural health monitoring
under controlled and dynamic environments, largely due to the
lack of a public benchmark dataset. To address this gap, this
paper presents a public dataset from a long-term outdoor
structural monitoring experiment conducted at the University of
Utah, Salt Lake City. The monitoring, spanning over 4.5 years,
collected approximately 6.4 million guided waves under both
regular environmental variations (e.g., daily temperature changes
ranging from 260.95 K (−12.2 °C) to 325.65 K (52.5 °C)) and
irregular variations (e.g., rain and snow). The measured guided
waves in the public dataset are also affected by sensor drift and
installation shifts consistently over time. Additionally,
thirteen types of damage were introduced to the monitored
structure to support damage detection and severity evaluation
under these conditions. The dataset includes measurement times,
temperature, humidity, air pressure, brightness, and weather
information to aid in damage detection. The provided public
dataset aims to assist researchers in developing more practical
methods for structural health monitoring in uncontrolled and
dynamic environments.},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Few studies address guided wave structural health monitoring
under controlled and dynamic environments, largely due to the
lack of a public benchmark dataset. To address this gap, this
paper presents a public dataset from a long-term outdoor
structural monitoring experiment conducted at the University of
Utah, Salt Lake City. The monitoring, spanning over 4.5 years,
collected approximately 6.4 million guided waves under both
regular environmental variations (e.g., daily temperature changes
ranging from 260.95 K (−12.2 °C) to 325.65 K (52.5 °C)) and
irregular variations (e.g., rain and snow). The measured guided
waves in the public dataset are also affected by sensor drift and
installation shifts consistently over time. Additionally,
thirteen types of damage were introduced to the monitored
structure to support damage detection and severity evaluation
under these conditions. The dataset includes measurement times,
temperature, humidity, air pressure, brightness, and weather
information to aid in damage detection. The provided public
dataset aims to assist researchers in developing more practical
methods for structural health monitoring in uncontrolled and
dynamic environments.
under controlled and dynamic environments, largely due to the
lack of a public benchmark dataset. To address this gap, this
paper presents a public dataset from a long-term outdoor
structural monitoring experiment conducted at the University of
Utah, Salt Lake City. The monitoring, spanning over 4.5 years,
collected approximately 6.4 million guided waves under both
regular environmental variations (e.g., daily temperature changes
ranging from 260.95 K (−12.2 °C) to 325.65 K (52.5 °C)) and
irregular variations (e.g., rain and snow). The measured guided
waves in the public dataset are also affected by sensor drift and
installation shifts consistently over time. Additionally,
thirteen types of damage were introduced to the monitored
structure to support damage detection and severity evaluation
under these conditions. The dataset includes measurement times,
temperature, humidity, air pressure, brightness, and weather
information to aid in damage detection. The provided public
dataset aims to assist researchers in developing more practical
methods for structural health monitoring in uncontrolled and
dynamic environments.
197.
Lin Yang, Bryan Conry, Hailey E Hall, Joel B Harley, Michael S Kesler, Michael R Tonks, Amanda R Krause
Elucidating grain boundary energy minimization mechanisms in textured Ca-doped alumina with inclination-dependent Monte Carlo Potts simulations Journal Article
In: Acta materialia, vol. 288, no. 120876, pp. 120876, 2025.
@article{Yang2025-hi,
title = {Elucidating grain boundary energy minimization mechanisms in
textured Ca-doped alumina with inclination-dependent Monte Carlo
Potts simulations},
author = {Lin Yang and Bryan Conry and Hailey E Hall and Joel B Harley and Michael S Kesler and Michael R Tonks and Amanda R Krause},
url = {http://dx.doi.org/10.1016/j.actamat.2025.120876},
doi = {10.1016/j.actamat.2025.120876},
year = {2025},
date = {2025-04-01},
journal = {Acta materialia},
volume = {288},
number = {120876},
pages = {120876},
publisher = {Elsevier BV},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
196.
Kang Yang, Tianqi Liu, Zekun Yang, Yang Zhou, Zhihui Tian, Nam H Kim, Joel B Harley
Baseline optimized autoencoder-based unsupervised anomaly detection in uncontrolled dynamic structural health monitoring Journal Article
In: Structural health monitoring, pp. 14759217251324107, 2025.
@article{Yang2025-jp,
title = {Baseline optimized autoencoder-based unsupervised anomaly
detection in uncontrolled dynamic structural health monitoring},
author = {Kang Yang and Tianqi Liu and Zekun Yang and Yang Zhou and Zhihui Tian and Nam H Kim and Joel B Harley},
url = {https://scholar.google.com/citations?view_op=view_citation&hl=en&citation_for_view=Isf8yn0AAAAJ:qwy9JoKyICEC},
doi = {10.1177/14759217251324107},
year = {2025},
date = {2025-03-01},
urldate = {2025-03-01},
journal = {Structural health monitoring},
pages = {14759217251324107},
publisher = {SAGE Publications},
abstract = {Autoencoder reconstruction-based unsupervised anomaly detection
is widely used in structural health monitoring. However, these
methods typically require training on historical data from
healthy structures, collected under environmental conditions
similar to the test data. This limits their practical use, as it
demands a comprehensive dataset of historical guided waves
gathered across various environmental and operational conditions.
Additionally, these methods fail when the training data contain a
significant portion of damage-induced guided waves, as the
autoencoder may reconstruct damaged waves just as effectively as
normal ones. To overcome these challenges, our anomaly detection
model is trained directly on current measurements, eliminating
the risk of environmental discrepancies between training and test
data. Furthermore, our baseline optimization strategy biases the
autoencoder toward reconstructing normal guided waves, enabling
reliable anomaly detection even when a large proportion of the
training data are damage-induced waves. Additionally, we present
a strategy to enhance the model’s practical performance by
optimizing the weight factor for baseline loss and the baseline
set size, based on guided wave reconstruction performance,
without relying on damage labels. The effectiveness of this
baseline-optimized autoencoder model, even when the training data
contain significant damage-induced guided waves, is validated
through measurements from 10 regions, each spanning 80 days of
guided wave data collected under uncontrolled and dynamic
environmental conditions.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Autoencoder reconstruction-based unsupervised anomaly detection
is widely used in structural health monitoring. However, these
methods typically require training on historical data from
healthy structures, collected under environmental conditions
similar to the test data. This limits their practical use, as it
demands a comprehensive dataset of historical guided waves
gathered across various environmental and operational conditions.
Additionally, these methods fail when the training data contain a
significant portion of damage-induced guided waves, as the
autoencoder may reconstruct damaged waves just as effectively as
normal ones. To overcome these challenges, our anomaly detection
model is trained directly on current measurements, eliminating
the risk of environmental discrepancies between training and test
data. Furthermore, our baseline optimization strategy biases the
autoencoder toward reconstructing normal guided waves, enabling
reliable anomaly detection even when a large proportion of the
training data are damage-induced waves. Additionally, we present
a strategy to enhance the model’s practical performance by
optimizing the weight factor for baseline loss and the baseline
set size, based on guided wave reconstruction performance,
without relying on damage labels. The effectiveness of this
baseline-optimized autoencoder model, even when the training data
contain significant damage-induced guided waves, is validated
through measurements from 10 regions, each spanning 80 days of
guided wave data collected under uncontrolled and dynamic
environmental conditions.
is widely used in structural health monitoring. However, these
methods typically require training on historical data from
healthy structures, collected under environmental conditions
similar to the test data. This limits their practical use, as it
demands a comprehensive dataset of historical guided waves
gathered across various environmental and operational conditions.
Additionally, these methods fail when the training data contain a
significant portion of damage-induced guided waves, as the
autoencoder may reconstruct damaged waves just as effectively as
normal ones. To overcome these challenges, our anomaly detection
model is trained directly on current measurements, eliminating
the risk of environmental discrepancies between training and test
data. Furthermore, our baseline optimization strategy biases the
autoencoder toward reconstructing normal guided waves, enabling
reliable anomaly detection even when a large proportion of the
training data are damage-induced waves. Additionally, we present
a strategy to enhance the model’s practical performance by
optimizing the weight factor for baseline loss and the baseline
set size, based on guided wave reconstruction performance,
without relying on damage labels. The effectiveness of this
baseline-optimized autoencoder model, even when the training data
contain significant damage-induced guided waves, is validated
through measurements from 10 regions, each spanning 80 days of
guided wave data collected under uncontrolled and dynamic
environmental conditions.
195.
Lee Shi Yn, Zairil Zaludin, Joel B Harley, Jung-Ryul Lee, Mohammad Yazdi Harmin, Chia Chen Ciang
Statistical Thresholding of Ultrasonic Amplitude Maps for Automated Damage Segmentation Journal Article
In: Journal of Aeronautics, Astronautics and Aviation, vol. 57, no. 3S, pp. 319–327, 2025.
@article{Yn2025-bx,
title = {Statistical Thresholding of Ultrasonic Amplitude Maps for
Automated Damage Segmentation},
author = {Lee Shi Yn and Zairil Zaludin and Joel B Harley and Jung-Ryul Lee and Mohammad Yazdi Harmin and Chia Chen Ciang},
url = {https://www.airitilibrary.com/Article/Detail/P20140627004-N202504100011-00005},
doi = {10.6125/JoAAA.202503_57(3S).04},
year = {2025},
date = {2025-03-01},
urldate = {2025-03-01},
journal = {Journal of Aeronautics, Astronautics and Aviation},
volume = {57},
number = {3S},
pages = {319–327},
publisher = {Aeronautical and Astronautical Society of the Republic of China},
abstract = {Amplitude maps generated by ultrasound imaging are frequently
utilized to visualize invisible damages in thin-walled
aero-mechanical structures. Accurate evaluation of damage size
from these maps is crucial; however, a reliable automated method
for this purpose under the influence of imaging noise is not
available. To address this issue, four threshold calculation
methods based on statistical analysis of noise content in an
amplitude map were developed. These candidates were numerically
optimized using amplitude maps containing damages of various
sizes and Gaussian noise of differing intensities. The candidate
demonstrating the greatest immunity to parameter variations and
the highest potential for accurate damage size evaluation was
identified. This candidate was then parametrically optimized and
benchmarked against k-means clustering. The results demonstrate
that the newly proposed statistical …},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Amplitude maps generated by ultrasound imaging are frequently
utilized to visualize invisible damages in thin-walled
aero-mechanical structures. Accurate evaluation of damage size
from these maps is crucial; however, a reliable automated method
for this purpose under the influence of imaging noise is not
available. To address this issue, four threshold calculation
methods based on statistical analysis of noise content in an
amplitude map were developed. These candidates were numerically
optimized using amplitude maps containing damages of various
sizes and Gaussian noise of differing intensities. The candidate
demonstrating the greatest immunity to parameter variations and
the highest potential for accurate damage size evaluation was
identified. This candidate was then parametrically optimized and
benchmarked against k-means clustering. The results demonstrate
that the newly proposed statistical …
utilized to visualize invisible damages in thin-walled
aero-mechanical structures. Accurate evaluation of damage size
from these maps is crucial; however, a reliable automated method
for this purpose under the influence of imaging noise is not
available. To address this issue, four threshold calculation
methods based on statistical analysis of noise content in an
amplitude map were developed. These candidates were numerically
optimized using amplitude maps containing damages of various
sizes and Gaussian noise of differing intensities. The candidate
demonstrating the greatest immunity to parameter variations and
the highest potential for accurate damage size evaluation was
identified. This candidate was then parametrically optimized and
benchmarked against k-means clustering. The results demonstrate
that the newly proposed statistical …
194.
Maximillian T Diaz, Lavanya Durai, Kalyn M Kearney, Erica M Lindbeck, Isaly Tappan, Joel B Harley, Jennifer A Nichols
Evaluating recruitment methods for selection bias: A large, experimental study of hand biomechanics Journal Article
In: Journal of biomechanics, vol. 182, no. 112558, pp. 112558, 2025.
@article{Diaz2025-cm,
title = {Evaluating recruitment methods for selection bias: A large,
experimental study of hand biomechanics},
author = {Maximillian T Diaz and Lavanya Durai and Kalyn M Kearney and Erica M Lindbeck and Isaly Tappan and Joel B Harley and Jennifer A Nichols},
url = {https://scholar.google.com/citations?view_op=view_citation&hl=en&citation_for_view=Isf8yn0AAAAJ:CYCckWUYoCcC},
doi = {10.1016/j.jbiomech.2025.112558},
year = {2025},
date = {2025-03-01},
urldate = {2025-03-01},
journal = {Journal of biomechanics},
volume = {182},
number = {112558},
pages = {112558},
publisher = {Elsevier BV},
abstract = {Biomechanics studies rely on non-random recruitment methods to
obtain study participants. However, the use of common recruitment
methods and small sample sizes may influence a given study's
generalizability due to selection bias. To improve
generalizability, ecological validity, and participant
convenience, recent biomechanics studies have moved beyond lab
conditions. However, it is unknown if simply leaving the lab
space and increasing sample sizes reduces the risks associated
with selection bias. Previous studies relied on chart and
literature reviews to identify selection bias. Herein, we build
upon this work by exploring the potential for and influence of
selection bias in three common recruiting methods by performing
an experimental, population-level study on hand biomechanics.
Hand biomechanics was assessed in the community using a portable
lab setup to measure hand function, grip strength, and pinch
strength. A total of 642 apparently healthy participants were
recruited across 18 locations, with 426 participants selected
based on complete data responses and being between the ages of 18
to 35. Sex stratified analysis was performed to see how
recruiting only biomechanists, undergraduate students, or
university affiliates changed population estimates of hand
strength. The presence of selection bias was observed in all
three test cases with both male and female biomechanists,
graduate students, and non-university affiliates having
significant increases in pinch and grip strengths ranging from
6.2% to 19.4% above overall population values. This study
quantitively shows how simply leaving the lab and increasing
subject recruitment does not eliminate the potential for
selection bias in studies of hand biomechanics.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Biomechanics studies rely on non-random recruitment methods to
obtain study participants. However, the use of common recruitment
methods and small sample sizes may influence a given study's
generalizability due to selection bias. To improve
generalizability, ecological validity, and participant
convenience, recent biomechanics studies have moved beyond lab
conditions. However, it is unknown if simply leaving the lab
space and increasing sample sizes reduces the risks associated
with selection bias. Previous studies relied on chart and
literature reviews to identify selection bias. Herein, we build
upon this work by exploring the potential for and influence of
selection bias in three common recruiting methods by performing
an experimental, population-level study on hand biomechanics.
Hand biomechanics was assessed in the community using a portable
lab setup to measure hand function, grip strength, and pinch
strength. A total of 642 apparently healthy participants were
recruited across 18 locations, with 426 participants selected
based on complete data responses and being between the ages of 18
to 35. Sex stratified analysis was performed to see how
recruiting only biomechanists, undergraduate students, or
university affiliates changed population estimates of hand
strength. The presence of selection bias was observed in all
three test cases with both male and female biomechanists,
graduate students, and non-university affiliates having
significant increases in pinch and grip strengths ranging from
6.2% to 19.4% above overall population values. This study
quantitively shows how simply leaving the lab and increasing
subject recruitment does not eliminate the potential for
selection bias in studies of hand biomechanics.
obtain study participants. However, the use of common recruitment
methods and small sample sizes may influence a given study's
generalizability due to selection bias. To improve
generalizability, ecological validity, and participant
convenience, recent biomechanics studies have moved beyond lab
conditions. However, it is unknown if simply leaving the lab
space and increasing sample sizes reduces the risks associated
with selection bias. Previous studies relied on chart and
literature reviews to identify selection bias. Herein, we build
upon this work by exploring the potential for and influence of
selection bias in three common recruiting methods by performing
an experimental, population-level study on hand biomechanics.
Hand biomechanics was assessed in the community using a portable
lab setup to measure hand function, grip strength, and pinch
strength. A total of 642 apparently healthy participants were
recruited across 18 locations, with 426 participants selected
based on complete data responses and being between the ages of 18
to 35. Sex stratified analysis was performed to see how
recruiting only biomechanists, undergraduate students, or
university affiliates changed population estimates of hand
strength. The presence of selection bias was observed in all
three test cases with both male and female biomechanists,
graduate students, and non-university affiliates having
significant increases in pinch and grip strengths ranging from
6.2% to 19.4% above overall population values. This study
quantitively shows how simply leaving the lab and increasing
subject recruitment does not eliminate the potential for
selection bias in studies of hand biomechanics.
193.
Kang Yang, Kang Gao, Junkai Zhou, Chao Gao, Tingsong Xiao, Harsha Vardhan Tetali, Joel B Harley
In: Ultrasonics, no. 107632, pp. 107632, 2025.
@article{Yang2025-aj,
title = {Optimal principal component and measurement interval selection
for PCA reconstruction-based anomaly detection in uncontrolled
structural health monitoring},
author = {Kang Yang and Kang Gao and Junkai Zhou and Chao Gao and Tingsong Xiao and Harsha Vardhan Tetali and Joel B Harley},
url = {https://scholar.google.com/citations?view_op=view_citation&hl=en&citation_for_view=Isf8yn0AAAAJ:hSRAE-fF4OAC},
doi = {10.1016/j.ultras.2025.107632},
year = {2025},
date = {2025-03-01},
urldate = {2025-03-01},
journal = {Ultrasonics},
number = {107632},
pages = {107632},
publisher = {Elsevier BV},
abstract = {PCA reconstruction-based techniques are widely used in guided
wave structural health monitoring to facilitate unsupervised
damage detection. The measurement interval of collecting
evaluation data significantly influences the correlation among
the data points, impacting principal component values and,
consequently, the accuracy of damage detection. Despite its
importance, there has been limited research on the selection of
suitable components and measurement intervals to reduce false
alarms. This paper seeks to develop strategies for identifying
the optimal number of principal components and measurement
intervals for PCA reconstruction-based damage detection methods.
Our results indicate that the patterns of change in
reconstruction coefficients, based on the number of components
used in PCA reconstruction and the measurement interval for
collecting evaluation data, are effective indicators for
determining the optimal principal components and measurement
intervals for damage detection, without using any damage
information. The effectiveness of the indicators for determining
optimal components and measurement intervals is validated using
evaluation sets collected under uncontrolled and dynamic
monitoring conditions, with measurement intervals ranging from 86
to 8600 s per measurement.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
PCA reconstruction-based techniques are widely used in guided
wave structural health monitoring to facilitate unsupervised
damage detection. The measurement interval of collecting
evaluation data significantly influences the correlation among
the data points, impacting principal component values and,
consequently, the accuracy of damage detection. Despite its
importance, there has been limited research on the selection of
suitable components and measurement intervals to reduce false
alarms. This paper seeks to develop strategies for identifying
the optimal number of principal components and measurement
intervals for PCA reconstruction-based damage detection methods.
Our results indicate that the patterns of change in
reconstruction coefficients, based on the number of components
used in PCA reconstruction and the measurement interval for
collecting evaluation data, are effective indicators for
determining the optimal principal components and measurement
intervals for damage detection, without using any damage
information. The effectiveness of the indicators for determining
optimal components and measurement intervals is validated using
evaluation sets collected under uncontrolled and dynamic
monitoring conditions, with measurement intervals ranging from 86
to 8600 s per measurement.
wave structural health monitoring to facilitate unsupervised
damage detection. The measurement interval of collecting
evaluation data significantly influences the correlation among
the data points, impacting principal component values and,
consequently, the accuracy of damage detection. Despite its
importance, there has been limited research on the selection of
suitable components and measurement intervals to reduce false
alarms. This paper seeks to develop strategies for identifying
the optimal number of principal components and measurement
intervals for PCA reconstruction-based damage detection methods.
Our results indicate that the patterns of change in
reconstruction coefficients, based on the number of components
used in PCA reconstruction and the measurement interval for
collecting evaluation data, are effective indicators for
determining the optimal principal components and measurement
intervals for damage detection, without using any damage
information. The effectiveness of the indicators for determining
optimal components and measurement intervals is validated using
evaluation sets collected under uncontrolled and dynamic
monitoring conditions, with measurement intervals ranging from 86
to 8600 s per measurement.
192.
Kang Yang, Chao Zhang, Hanbo Yang, Linyuan Wang, Nam H Kim, Joel B Harley
Improving unsupervised long-term damage detection in an uncontrolled environment through noise-augmentation strategy Journal Article
In: Mechanical systems and signal processing, vol. 224, no. 112076, pp. 112076, 2025.
@article{Yang2025-ur,
title = {Improving unsupervised long-term damage detection in an
uncontrolled environment through noise-augmentation strategy},
author = {Kang Yang and Chao Zhang and Hanbo Yang and Linyuan Wang and Nam H Kim and Joel B Harley},
url = {https://scholar.google.com/citations?view_op=view_citation&hl=en&citation_for_view=Isf8yn0AAAAJ:TlpoogIpr_IC},
doi = {10.1016/j.ymssp.2024.112076},
year = {2025},
date = {2025-02-01},
urldate = {2025-02-01},
journal = {Mechanical systems and signal processing},
volume = {224},
number = {112076},
pages = {112076},
publisher = {Elsevier BV},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
191.
Ziqin Ding, Si Chen, Hanbo Yang, Kang Yang, Vladimir A Rakov, Joel B Harley, Yanan Zhu
Identification of lightning strikes to towers using machine-learning approach Miscellaneous
105th Annual Meeting of the American Meteorological Society, 2025.
@misc{Ding2025-zl,
title = {Identification of lightning strikes to towers using
machine-learning approach},
author = {Ziqin Ding and Si Chen and Hanbo Yang and Kang Yang and Vladimir A Rakov and Joel B Harley and Yanan Zhu},
url = {https://ams.confex.com/ams/105ANNUAL/meetingapp.cgi/Paper/447159},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
publisher = {AMS},
abstract = {Lightning often strikes tall objects and, hence, poses
significant threat to infrastructure. Modern lightning
locating systems can provide crucial information about
lightning, such as strike coordinates, peak current, etc., but
cannot tell whether the stroke terminated on a tall object
such as tower. Lightning electromagnetic field pulses (LEMP)
have been used to infer lightning currents. Using the unique
electric field pulse signatures of the tower terminated
lightning collected at the Lightning Observatory in
Gainesville (LOG), Florida, we present a machine-learning
approach to classify lightning strikes to towers. The
classification model used in this study is based on supervised
Multi-Layer Perceptron model (MLP), aiming to capture complex
pulse patterns with the neural network architecture. We
refined the model by tuning the configuration of the training
dataset, and validated its performance with Local
Interpretable Model-agnostic Explanations. The result shows
that an overall 99.97% classification accuracy can be
achieved, with 90% and 99.97% classification accuracy for
tower terminated lightning and non-tower terminated lightning,
respectively.},
howpublished = {105th Annual Meeting of the American Meteorological Society},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Lightning often strikes tall objects and, hence, poses
significant threat to infrastructure. Modern lightning
locating systems can provide crucial information about
lightning, such as strike coordinates, peak current, etc., but
cannot tell whether the stroke terminated on a tall object
such as tower. Lightning electromagnetic field pulses (LEMP)
have been used to infer lightning currents. Using the unique
electric field pulse signatures of the tower terminated
lightning collected at the Lightning Observatory in
Gainesville (LOG), Florida, we present a machine-learning
approach to classify lightning strikes to towers. The
classification model used in this study is based on supervised
Multi-Layer Perceptron model (MLP), aiming to capture complex
pulse patterns with the neural network architecture. We
refined the model by tuning the configuration of the training
dataset, and validated its performance with Local
Interpretable Model-agnostic Explanations. The result shows
that an overall 99.97% classification accuracy can be
achieved, with 90% and 99.97% classification accuracy for
tower terminated lightning and non-tower terminated lightning,
respectively.
significant threat to infrastructure. Modern lightning
locating systems can provide crucial information about
lightning, such as strike coordinates, peak current, etc., but
cannot tell whether the stroke terminated on a tall object
such as tower. Lightning electromagnetic field pulses (LEMP)
have been used to infer lightning currents. Using the unique
electric field pulse signatures of the tower terminated
lightning collected at the Lightning Observatory in
Gainesville (LOG), Florida, we present a machine-learning
approach to classify lightning strikes to towers. The
classification model used in this study is based on supervised
Multi-Layer Perceptron model (MLP), aiming to capture complex
pulse patterns with the neural network architecture. We
refined the model by tuning the configuration of the training
dataset, and validated its performance with Local
Interpretable Model-agnostic Explanations. The result shows
that an overall 99.97% classification accuracy can be
achieved, with 90% and 99.97% classification accuracy for
tower terminated lightning and non-tower terminated lightning,
respectively.
190.
Kang Yang, Zhenhan Lin, Zekun Yang, Zhihui Tian, Jie Ma, José C Príncipe, Joel B Harley
Improved PCA reconstruction-based unsupervised anomaly detection in uncontrolled structural health monitoring with correntropy Journal Article
In: IEEE transactions on industrial informatics, vol. PP, no. 99, pp. 1–11, 2025.
@article{Yang2025-ac,
title = {Improved PCA reconstruction-based unsupervised anomaly
detection in uncontrolled structural health monitoring with
correntropy},
author = {Kang Yang and Zhenhan Lin and Zekun Yang and Zhihui Tian and Jie Ma and José C Príncipe and Joel B Harley},
url = {http://dx.doi.org/10.1109/TII.2025.3584458},
doi = {10.1109/tii.2025.3584458},
year = {2025},
date = {2025-01-01},
urldate = {2025-01-01},
journal = {IEEE transactions on industrial informatics},
volume = {PP},
number = {99},
pages = {1–11},
publisher = {Institute of Electrical and Electronics Engineers (IEEE)},
abstract = {Guided wave-based structural health monitoring is extensively
utilized in various industrial applications to ensure the
integrity of components within industrial systems. Among these
monitoring techniques, principal component analysis (PCA)
reconstruction methods are widely used for anomaly detection due
to their computational efficiency and interoperability. However,
existing PCA reconstruction methods are semisupervised anomaly
detection approaches that require training on historical normal
data and fail to detect anomalous signals within the training
set. To address this limitation, this work proposes a correntropy
PCA (C-PCA), enabling fully unsupervised anomaly detection on raw
training data without requiring label information, when the
dataset contains a high proportion of abnormal signals. This
method allows anomaly detection on real-time measurements without
the need for precleaned historical normal data or can also be
used to generate clean data for existing semisupervised anomaly
detection methods. In correntropy PCA, principal components are
extracted from the correntropy matrix rather than the correlation
matrix. The correntropy, representing the statistical dependence
between samples of guided waves, is estimated utilizing a
Gaussian kernel with a specified kernel width. Through the
optimization of the kernel width, the correntropy PCA
reconstruction method demonstrates superior anomaly detection
performance compared with the standard PCA reconstruction method,
especially in scenarios where training data are contaminated by a
significant proportion of abnormal signals. Guidelines for the
optimization of the kernel width are provided. The effectiveness
of the correntropy PCA reconstruction-based anomaly detection
method is validated using data collected from ten regions over an
80-day period, encompassing guided waves induced by damage
occurring over durations ranging from 2 to 20 days.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Guided wave-based structural health monitoring is extensively
utilized in various industrial applications to ensure the
integrity of components within industrial systems. Among these
monitoring techniques, principal component analysis (PCA)
reconstruction methods are widely used for anomaly detection due
to their computational efficiency and interoperability. However,
existing PCA reconstruction methods are semisupervised anomaly
detection approaches that require training on historical normal
data and fail to detect anomalous signals within the training
set. To address this limitation, this work proposes a correntropy
PCA (C-PCA), enabling fully unsupervised anomaly detection on raw
training data without requiring label information, when the
dataset contains a high proportion of abnormal signals. This
method allows anomaly detection on real-time measurements without
the need for precleaned historical normal data or can also be
used to generate clean data for existing semisupervised anomaly
detection methods. In correntropy PCA, principal components are
extracted from the correntropy matrix rather than the correlation
matrix. The correntropy, representing the statistical dependence
between samples of guided waves, is estimated utilizing a
Gaussian kernel with a specified kernel width. Through the
optimization of the kernel width, the correntropy PCA
reconstruction method demonstrates superior anomaly detection
performance compared with the standard PCA reconstruction method,
especially in scenarios where training data are contaminated by a
significant proportion of abnormal signals. Guidelines for the
optimization of the kernel width are provided. The effectiveness
of the correntropy PCA reconstruction-based anomaly detection
method is validated using data collected from ten regions over an
80-day period, encompassing guided waves induced by damage
occurring over durations ranging from 2 to 20 days.
utilized in various industrial applications to ensure the
integrity of components within industrial systems. Among these
monitoring techniques, principal component analysis (PCA)
reconstruction methods are widely used for anomaly detection due
to their computational efficiency and interoperability. However,
existing PCA reconstruction methods are semisupervised anomaly
detection approaches that require training on historical normal
data and fail to detect anomalous signals within the training
set. To address this limitation, this work proposes a correntropy
PCA (C-PCA), enabling fully unsupervised anomaly detection on raw
training data without requiring label information, when the
dataset contains a high proportion of abnormal signals. This
method allows anomaly detection on real-time measurements without
the need for precleaned historical normal data or can also be
used to generate clean data for existing semisupervised anomaly
detection methods. In correntropy PCA, principal components are
extracted from the correntropy matrix rather than the correlation
matrix. The correntropy, representing the statistical dependence
between samples of guided waves, is estimated utilizing a
Gaussian kernel with a specified kernel width. Through the
optimization of the kernel width, the correntropy PCA
reconstruction method demonstrates superior anomaly detection
performance compared with the standard PCA reconstruction method,
especially in scenarios where training data are contaminated by a
significant proportion of abnormal signals. Guidelines for the
optimization of the kernel width are provided. The effectiveness
of the correntropy PCA reconstruction-based anomaly detection
method is validated using data collected from ten regions over an
80-day period, encompassing guided waves induced by damage
occurring over durations ranging from 2 to 20 days.
189.
Ayobami S Edun, Cody LaFlamme, Evan J Benoit, Cynthia Furse, Joel B Harley
A statistical analysis of fault detection in photovoltaics using reflectometry Journal Article
In: IEEE journal of photovoltaics, vol. PP, no. 99, pp. 1–10, 2025.
@article{Edun2025-qn,
title = {A statistical analysis of fault detection in photovoltaics using
reflectometry},
author = {Ayobami S Edun and Cody LaFlamme and Evan J Benoit and Cynthia Furse and Joel B Harley},
url = {https://ieeexplore.ieee.org/document/11184635},
doi = {10.1109/jphotov.2025.3608439},
year = {2025},
date = {2025-01-01},
journal = {IEEE journal of photovoltaics},
volume = {PP},
number = {99},
pages = {1–10},
publisher = {Institute of Electrical and Electronics Engineers (IEEE)},
abstract = {Spread spectrum time domain reflectometry (SSTDR) has been used
to detect different kinds of faults in cables, aircraft wiring,
and photovoltaic (PV) setups. One significant problem is that,
since most methods that use reflectometry are based on a
comparison with a known baseline, variations in the baseline
caused by system or environmental variations can overshadow the
reflections produced by faults and reduce the probability of
detection. The effects of these variations are exacerbated for
faults far from the test device. The objective of this work is to
statistically estimate the probability that a reflection from a
PV array is or is not a fault amid environmental variations,
system variations, and at different locations along the line. Our
results show the probability of detecting fault presence at
different distances from the test device when there are partial
and full disconnects over a range of signal-to-noise ratio (SNR).
At a distance of 97.54 m (320 ft) from the tester, the
probability of detection of a full disconnect is as high as 0.75
with an SNR of about 9 dB, while the probability of detection is
nearly zero for partial faults at the same location. We also
present scenarios where faults can no longer be detected and how
averaging could help improve the probability of detection.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Spread spectrum time domain reflectometry (SSTDR) has been used
to detect different kinds of faults in cables, aircraft wiring,
and photovoltaic (PV) setups. One significant problem is that,
since most methods that use reflectometry are based on a
comparison with a known baseline, variations in the baseline
caused by system or environmental variations can overshadow the
reflections produced by faults and reduce the probability of
detection. The effects of these variations are exacerbated for
faults far from the test device. The objective of this work is to
statistically estimate the probability that a reflection from a
PV array is or is not a fault amid environmental variations,
system variations, and at different locations along the line. Our
results show the probability of detecting fault presence at
different distances from the test device when there are partial
and full disconnects over a range of signal-to-noise ratio (SNR).
At a distance of 97.54 m (320 ft) from the tester, the
probability of detection of a full disconnect is as high as 0.75
with an SNR of about 9 dB, while the probability of detection is
nearly zero for partial faults at the same location. We also
present scenarios where faults can no longer be detected and how
averaging could help improve the probability of detection.
to detect different kinds of faults in cables, aircraft wiring,
and photovoltaic (PV) setups. One significant problem is that,
since most methods that use reflectometry are based on a
comparison with a known baseline, variations in the baseline
caused by system or environmental variations can overshadow the
reflections produced by faults and reduce the probability of
detection. The effects of these variations are exacerbated for
faults far from the test device. The objective of this work is to
statistically estimate the probability that a reflection from a
PV array is or is not a fault amid environmental variations,
system variations, and at different locations along the line. Our
results show the probability of detecting fault presence at
different distances from the test device when there are partial
and full disconnects over a range of signal-to-noise ratio (SNR).
At a distance of 97.54 m (320 ft) from the tester, the
probability of detection of a full disconnect is as high as 0.75
with an SNR of about 9 dB, while the probability of detection is
nearly zero for partial faults at the same location. We also
present scenarios where faults can no longer be detected and how
averaging could help improve the probability of detection.
2024
188.
Zhou Tang, Jiayi Song, Igor Bretas, Liza Garcia, Luana Queiroz, Yifei Suo, Cristian Erazo, Joel Harley, Alina Zare, Ebrahim Babaeian, Nikolaos Tziolas, Sabine Grunwald, Jose Dubeux, Chang Zhao
American Geophysical Union Annual Meeting, 2024.
@misc{Tang2024-mq,
title = {Spatial prediction of soil organic carbon stocks across
grazing lands in Florida using earth observation and deep
neural networks},
author = {Zhou Tang and Jiayi Song and Igor Bretas and Liza Garcia and Luana Queiroz and Yifei Suo and Cristian Erazo and Joel Harley and Alina Zare and Ebrahim Babaeian and Nikolaos Tziolas and Sabine Grunwald and Jose Dubeux and Chang Zhao},
url = {https://ui.adsabs.harvard.edu/abs/2024AGUFMGC21J0008T/abstract?},
year = {2024},
date = {2024-12-01},
volume = {2024},
number = {8},
pages = {GC21J–0008},
abstract = {Grazing lands are crucial carbon sinks and play a significant
role in mitigating global climate change. In Florida, nearly
half of the agricultural land is used for grazing. Accurate
predictive maps of soil organic carbon (SOC) stocks are
essential for assessing the capacity of these extensive lands
as carbon reservoirs. However, predicting SOC in grazing lands
is challenging due to complex soil-landscape relationships,
high spatial heterogeneity, and limited historical data. This
study aimed to develop an accurate end-to-end deep learning
model for predicting topsoil SOC stocks (0-15 cm) using earth
observation data across Florida's grazing lands. To enhance
the spatial representation of soil carbon datasets, 113
topsoil observations were collected from 2022 to 2024 in
different land uses at 36 farm locations, selected with the
conditioned Latin hypercube sampling method. This new soil
database was integrated with over 1000 legacy soil samples
from various land use types, collected during the 2008-2009
Florida Carbon Project, to expand the sample pool for model
calibration. The model employs convolutional neural networks
(CNNs) to extract spectral and spatial features from satellite
images, combined with ancillary environmental variables, and
processed through a multi-layer perceptron (MLP) for
regression. Our model utilized PlanetScope multispectral
imagery at 3-meter resolution and a holistic set of geospatial
covariates, including climate, vegetation, soil properties,
topography, and hydrology factors. Our custom model (CNN-MLP)
achieved high accuracy with an R² of 0.71 and a root mean
squared error (RMSE) of 20.15 Mg C ha-1 in Florida's grazing
lands, which is comparable to other models (MLP and random
forest) under 5-fold cross-validation. Image features derived
from the CNN were significant in SOC prediction. Additionally,
new features like neighboring wildfire areas and vegetation
dynamics, measured by time-series vegetation index statistics,
were important predictors. This study provides baseline
information and modeling tools for estimating SOC stocks in
grazing lands. As new soil carbon data becomes available, the
model can be readily fine-tuned to enhance accuracy for
large-scale grazing land SOC mapping over time.},
howpublished = {American Geophysical Union Annual Meeting},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Grazing lands are crucial carbon sinks and play a significant
role in mitigating global climate change. In Florida, nearly
half of the agricultural land is used for grazing. Accurate
predictive maps of soil organic carbon (SOC) stocks are
essential for assessing the capacity of these extensive lands
as carbon reservoirs. However, predicting SOC in grazing lands
is challenging due to complex soil-landscape relationships,
high spatial heterogeneity, and limited historical data. This
study aimed to develop an accurate end-to-end deep learning
model for predicting topsoil SOC stocks (0-15 cm) using earth
observation data across Florida's grazing lands. To enhance
the spatial representation of soil carbon datasets, 113
topsoil observations were collected from 2022 to 2024 in
different land uses at 36 farm locations, selected with the
conditioned Latin hypercube sampling method. This new soil
database was integrated with over 1000 legacy soil samples
from various land use types, collected during the 2008-2009
Florida Carbon Project, to expand the sample pool for model
calibration. The model employs convolutional neural networks
(CNNs) to extract spectral and spatial features from satellite
images, combined with ancillary environmental variables, and
processed through a multi-layer perceptron (MLP) for
regression. Our model utilized PlanetScope multispectral
imagery at 3-meter resolution and a holistic set of geospatial
covariates, including climate, vegetation, soil properties,
topography, and hydrology factors. Our custom model (CNN-MLP)
achieved high accuracy with an R² of 0.71 and a root mean
squared error (RMSE) of 20.15 Mg C ha-1 in Florida's grazing
lands, which is comparable to other models (MLP and random
forest) under 5-fold cross-validation. Image features derived
from the CNN were significant in SOC prediction. Additionally,
new features like neighboring wildfire areas and vegetation
dynamics, measured by time-series vegetation index statistics,
were important predictors. This study provides baseline
information and modeling tools for estimating SOC stocks in
grazing lands. As new soil carbon data becomes available, the
model can be readily fine-tuned to enhance accuracy for
large-scale grazing land SOC mapping over time.
role in mitigating global climate change. In Florida, nearly
half of the agricultural land is used for grazing. Accurate
predictive maps of soil organic carbon (SOC) stocks are
essential for assessing the capacity of these extensive lands
as carbon reservoirs. However, predicting SOC in grazing lands
is challenging due to complex soil-landscape relationships,
high spatial heterogeneity, and limited historical data. This
study aimed to develop an accurate end-to-end deep learning
model for predicting topsoil SOC stocks (0-15 cm) using earth
observation data across Florida's grazing lands. To enhance
the spatial representation of soil carbon datasets, 113
topsoil observations were collected from 2022 to 2024 in
different land uses at 36 farm locations, selected with the
conditioned Latin hypercube sampling method. This new soil
database was integrated with over 1000 legacy soil samples
from various land use types, collected during the 2008-2009
Florida Carbon Project, to expand the sample pool for model
calibration. The model employs convolutional neural networks
(CNNs) to extract spectral and spatial features from satellite
images, combined with ancillary environmental variables, and
processed through a multi-layer perceptron (MLP) for
regression. Our model utilized PlanetScope multispectral
imagery at 3-meter resolution and a holistic set of geospatial
covariates, including climate, vegetation, soil properties,
topography, and hydrology factors. Our custom model (CNN-MLP)
achieved high accuracy with an R² of 0.71 and a root mean
squared error (RMSE) of 20.15 Mg C ha-1 in Florida's grazing
lands, which is comparable to other models (MLP and random
forest) under 5-fold cross-validation. Image features derived
from the CNN were significant in SOC prediction. Additionally,
new features like neighboring wildfire areas and vegetation
dynamics, measured by time-series vegetation index statistics,
were important predictors. This study provides baseline
information and modeling tools for estimating SOC stocks in
grazing lands. As new soil carbon data becomes available, the
model can be readily fine-tuned to enhance accuracy for
large-scale grazing land SOC mapping over time.
187.
Peter Toma, Md Ali Muntaha, Joel B Harley, Michael R Tonks
Modeling fission gas release at the mesoscale using multiscale DenseNet regression with attention mechanism and inception blocks Journal Article
In: Journal of nuclear materials, vol. 601, no. 155315, pp. 155315, 2024.
@article{Toma2024-jk,
title = {Modeling fission gas release at the mesoscale using
multiscale DenseNet regression with attention mechanism and
inception blocks},
author = {Peter Toma and Md Ali Muntaha and Joel B Harley and Michael R Tonks},
url = {https://scholar.google.com/citations?view_op=view_citation&hl=en&user=Isf8yn0AAAAJ&sortby=pubdate&citation_for_view=Isf8yn0AAAAJ:pAkWuXOU-OoC},
doi = {10.1016/j.jnucmat.2024.155315},
year = {2024},
date = {2024-12-01},
journal = {Journal of nuclear materials},
volume = {601},
number = {155315},
pages = {155315},
publisher = {Elsevier BV},
abstract = {Mesoscale simulations of fission gas release (FGR) in nuclear
fuel provide a powerful tool for understanding how
microstructure evolution impacts FGR, but they are
computationally intensive. In this study, we present an
alternate, data-driven approach, using deep learning to
predict instantaneous FGR flux from 2D nuclear fuel
microstructure images. Four convolutional neural network
(CNN) architectures with multiscale regression are trained
and evaluated on simulated FGR data generated using a hybrid
phase field/cluster dynamics model. All four networks show
high predictive power, with $R^2$ values above 98%. The
best performing network combine a Convolutional Block
Attention Module (CBAM) and InceptionNet mechanisms to
provide superior accuracy (mean absolute percentage error of
4.4%), training stability, and robustness on very low
instantaneous FGR flux values.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Mesoscale simulations of fission gas release (FGR) in nuclear
fuel provide a powerful tool for understanding how
microstructure evolution impacts FGR, but they are
computationally intensive. In this study, we present an
alternate, data-driven approach, using deep learning to
predict instantaneous FGR flux from 2D nuclear fuel
microstructure images. Four convolutional neural network
(CNN) architectures with multiscale regression are trained
and evaluated on simulated FGR data generated using a hybrid
phase field/cluster dynamics model. All four networks show
high predictive power, with $R^2$ values above 98%. The
best performing network combine a Convolutional Block
Attention Module (CBAM) and InceptionNet mechanisms to
provide superior accuracy (mean absolute percentage error of
4.4%), training stability, and robustness on very low
instantaneous FGR flux values.
fuel provide a powerful tool for understanding how
microstructure evolution impacts FGR, but they are
computationally intensive. In this study, we present an
alternate, data-driven approach, using deep learning to
predict instantaneous FGR flux from 2D nuclear fuel
microstructure images. Four convolutional neural network
(CNN) architectures with multiscale regression are trained
and evaluated on simulated FGR data generated using a hybrid
phase field/cluster dynamics model. All four networks show
high predictive power, with $R^2$ values above 98%. The
best performing network combine a Convolutional Block
Attention Module (CBAM) and InceptionNet mechanisms to
provide superior accuracy (mean absolute percentage error of
4.4%), training stability, and robustness on very low
instantaneous FGR flux values.
186.
Liza Garcia, Jose C B Dubeux, Igor Lima Bretas, Luana M Dantas Queiroz, Cristian T E Mendes, Chang Zao, Joel Harley, Alina Zare, Zhou Tang
Exploring Soil Carbon Sequestration Potential in Florida: A Comparative Analysis of Beef Cattle Management Practices Miscellaneous
ASA, CSSA, SSSA International Annual Meeting, 2024.
@misc{Garcia2024-hq,
title = {Exploring Soil Carbon Sequestration Potential in Florida: A
Comparative Analysis of Beef Cattle Management Practices},
author = {Liza Garcia and Jose C B Dubeux and Igor Lima Bretas and Luana M Dantas Queiroz and Cristian T E Mendes and Chang Zao and Joel Harley and Alina Zare and Zhou Tang},
url = {https://scisoc.confex.com/scisoc/2024am/meetingapp.cgi/Paper/156762},
year = {2024},
date = {2024-11-01},
booktitle = {ASA, CSSA, SSSA International Annual Meeting},
publisher = {ASA-CSSA-SSSA},
abstract = {Florida's beef cattle systems rely on grazing forages, with
grazing lands conta...},
howpublished = {ASA, CSSA, SSSA International Annual Meeting},
keywords = {},
pubstate = {published},
tppubtype = {misc}
}
Florida's beef cattle systems rely on grazing forages, with
grazing lands conta...
grazing lands conta...
185.
Rishabh Gupta, Satya K Pothapragada, Weihuang Xu, Prateek Kumar Goel, Miguel A Barrera, Mira S Saldanha, Joel B Harley, Kelly T Morgan, Alina Zare, Lincoln Zotarelli
Estimating soil mineral nitrogen from data-sparse field experiments using crop model-guided deep learning approach Journal Article
In: Computers and electronics in agriculture, vol. 225, no. 109355, pp. 109355, 2024.
@article{Gupta2024-zx,
title = {Estimating soil mineral nitrogen from data-sparse field
experiments using crop model-guided deep learning approach},
author = {Rishabh Gupta and Satya K Pothapragada and Weihuang Xu and Prateek Kumar Goel and Miguel A Barrera and Mira S Saldanha and Joel B Harley and Kelly T Morgan and Alina Zare and Lincoln Zotarelli},
url = {https://scholar.google.com/citations?view_op=view_citation&hl=en&citation_for_view=Isf8yn0AAAAJ:QUX0mv85b1cC},
doi = {10.1016/j.compag.2024.109355},
year = {2024},
date = {2024-10-01},
urldate = {2024-09-01},
journal = {Computers and electronics in agriculture},
volume = {225},
number = {109355},
pages = {109355},
publisher = {Elsevier BV},
abstract = {Sandy soils are susceptible to excessive nitrogen (N) leaching
under intensive crop production which is linked with the soil's
low nutrient holding capacity and high-water infiltration rate.
Estimating soil mineral nitrogen (SMN) at the daily time-step is
crucial in providing fertilizer recommendations balancing plant
nitrogen use efficiency (NUE) and N losses to the environment.
Crop models [e.g., Decision Support System for Agrotechnology
Transfer (DSSAT)] can simulate the trend of SMN in varied
fertilizer rates and timing of application but are unable to
replicate its magnitude due to the inability to capture
high-water table conditions in a sub-irrigated soil. As an
alternative to such physics-based model, time-series deep
learning (DL) models based on a long short-term memory (LSTM) are
promising in understanding nonlinearity among complex variables.
Yet, purely data-driven DL models for crops are difficult to
obtain due to the insufficient amount of data available and the
excessive costs with producing more data. To address this
challenge, a hybrid model (hybrid-LSTM) was developed by
leveraging both the DSSAT andLSTM models to estimate daily SMN
primarily using daily weather, applied fertilizer rates- timings,
and the SMN sparse observations. This study used the observations
from field trials conducted between 2010-2014 in Hastings, FL.
The first step was to calibrate the DSSAT-SUBSTOR-Potato model to
produce reliable SMN of the topsoil for treatments with varied N
applied fertilizer rates split among the pre-planting, emergence,
and tuber-initiation stages of the potato crop. Thereafter, the
hybrid-LSTM model was trained on the calibrated DSSAT simulated
SMN time-series and fine-tuned its predictions using the observed
SMN to improve DSSAT simulated SMN. The hybrid-LSTM model was
then tested on both calibrated and uncalibrated DSSAT SMN
simulations where it outperformed the DSSAT model (range of
improvement ranged ~18-30% on comparing the normalized root mean
squared error) in providing reliable estimates of SMN across most
of the farms and years. This novel hybrid modeling approach could
guide stakeholders and farmers to build sustainable N management
with improved crop NUE and yield and help in minimizing
environmental losses.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Sandy soils are susceptible to excessive nitrogen (N) leaching
under intensive crop production which is linked with the soil's
low nutrient holding capacity and high-water infiltration rate.
Estimating soil mineral nitrogen (SMN) at the daily time-step is
crucial in providing fertilizer recommendations balancing plant
nitrogen use efficiency (NUE) and N losses to the environment.
Crop models [e.g., Decision Support System for Agrotechnology
Transfer (DSSAT)] can simulate the trend of SMN in varied
fertilizer rates and timing of application but are unable to
replicate its magnitude due to the inability to capture
high-water table conditions in a sub-irrigated soil. As an
alternative to such physics-based model, time-series deep
learning (DL) models based on a long short-term memory (LSTM) are
promising in understanding nonlinearity among complex variables.
Yet, purely data-driven DL models for crops are difficult to
obtain due to the insufficient amount of data available and the
excessive costs with producing more data. To address this
challenge, a hybrid model (hybrid-LSTM) was developed by
leveraging both the DSSAT andLSTM models to estimate daily SMN
primarily using daily weather, applied fertilizer rates- timings,
and the SMN sparse observations. This study used the observations
from field trials conducted between 2010-2014 in Hastings, FL.
The first step was to calibrate the DSSAT-SUBSTOR-Potato model to
produce reliable SMN of the topsoil for treatments with varied N
applied fertilizer rates split among the pre-planting, emergence,
and tuber-initiation stages of the potato crop. Thereafter, the
hybrid-LSTM model was trained on the calibrated DSSAT simulated
SMN time-series and fine-tuned its predictions using the observed
SMN to improve DSSAT simulated SMN. The hybrid-LSTM model was
then tested on both calibrated and uncalibrated DSSAT SMN
simulations where it outperformed the DSSAT model (range of
improvement ranged ~18-30% on comparing the normalized root mean
squared error) in providing reliable estimates of SMN across most
of the farms and years. This novel hybrid modeling approach could
guide stakeholders and farmers to build sustainable N management
with improved crop NUE and yield and help in minimizing
environmental losses.
under intensive crop production which is linked with the soil's
low nutrient holding capacity and high-water infiltration rate.
Estimating soil mineral nitrogen (SMN) at the daily time-step is
crucial in providing fertilizer recommendations balancing plant
nitrogen use efficiency (NUE) and N losses to the environment.
Crop models [e.g., Decision Support System for Agrotechnology
Transfer (DSSAT)] can simulate the trend of SMN in varied
fertilizer rates and timing of application but are unable to
replicate its magnitude due to the inability to capture
high-water table conditions in a sub-irrigated soil. As an
alternative to such physics-based model, time-series deep
learning (DL) models based on a long short-term memory (LSTM) are
promising in understanding nonlinearity among complex variables.
Yet, purely data-driven DL models for crops are difficult to
obtain due to the insufficient amount of data available and the
excessive costs with producing more data. To address this
challenge, a hybrid model (hybrid-LSTM) was developed by
leveraging both the DSSAT andLSTM models to estimate daily SMN
primarily using daily weather, applied fertilizer rates- timings,
and the SMN sparse observations. This study used the observations
from field trials conducted between 2010-2014 in Hastings, FL.
The first step was to calibrate the DSSAT-SUBSTOR-Potato model to
produce reliable SMN of the topsoil for treatments with varied N
applied fertilizer rates split among the pre-planting, emergence,
and tuber-initiation stages of the potato crop. Thereafter, the
hybrid-LSTM model was trained on the calibrated DSSAT simulated
SMN time-series and fine-tuned its predictions using the observed
SMN to improve DSSAT simulated SMN. The hybrid-LSTM model was
then tested on both calibrated and uncalibrated DSSAT SMN
simulations where it outperformed the DSSAT model (range of
improvement ranged ~18-30% on comparing the normalized root mean
squared error) in providing reliable estimates of SMN across most
of the farms and years. This novel hybrid modeling approach could
guide stakeholders and farmers to build sustainable N management
with improved crop NUE and yield and help in minimizing
environmental losses.
184.
Kalyn M Kearney, Tamara Ordonez Diaz, Joel B Harley, Jennifer A Nichols
From simulation to reality: Predicting torque with fatigue onset via transfer learning Journal Article
In: IEEE transactions on neural systems and rehabilitation engineering: a publication of the IEEE Engineering in Medicine and Biology Society, vol. 32, pp. 3669–3676, 2024.
@article{Kearney2024-ys,
title = {From simulation to reality: Predicting torque with fatigue onset
via transfer learning},
author = {Kalyn M Kearney and Tamara Ordonez Diaz and Joel B Harley and Jennifer A Nichols},
url = {https://scholar.google.com/citations?view_op=view_citation&hl=en&citation_for_view=Isf8yn0AAAAJ:mWEH9CqjF64C},
doi = {10.1109/TNSRE.2024.3465016},
year = {2024},
date = {2024-10-01},
urldate = {2024-10-01},
journal = {IEEE transactions on neural systems and rehabilitation
engineering: a publication of the IEEE Engineering in Medicine
and Biology Society},
volume = {32},
pages = {3669–3676},
publisher = {Institute of Electrical and Electronics Engineers (IEEE)},
abstract = {Muscle fatigue impacts upper extremity function but is often
overlooked in biomechanical models. The present work leveraged a
transfer learning approach to improve torque predictions during
fatiguing upper extremity movements. We developed two artificial
neural networks to model sustained elbow flexion: one trained
solely on recorded data (i.e., direct learning) and one
pre-trained on simulated data and fine-tuned on recorded data
(i.e., transfer learning). We simulated muscle activations and
joint torques using a musculoskeletal model and a muscle fatigue model (n = 1,701 simulations). We also recorded static
subject-specific features (e.g., anthropometric measurements) and
dynamic muscle activations and torques during sustained elbow flexion in healthy young adults (n = 25 subjects). Using the
simulated dataset, we pre-trained a long short-term memory neural
network (LSTM) to regress fatiguing elbow flexion torque from
muscle activations. We concatenated this pre-trained LSTM with a
feedforward architecture, and fine-tuned the model on recorded
muscle activations and static features to predict elbow flexion
torques. We trained a similar architecture solely on the recorded
data and compared each neural network's predictions on 5
leave-out subjects' data. The transfer learning model
outperformed the direct learning model, as indicated by a
decrease of 24.9% in their root-mean-square-errors (6.22 Nm and
8.28 Nm, respectively). The transfer learning model and direct
learning model outperformed analogous musculoskeletal
simulations, which consistently underpredicted elbow flexion
torque. Our results suggest that transfer learning from simulated
to recorded datasets can decrease reliance on assumptions
inherent to biomechanical models and yield predictions robust to
real-world conditions.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Muscle fatigue impacts upper extremity function but is often
overlooked in biomechanical models. The present work leveraged a
transfer learning approach to improve torque predictions during
fatiguing upper extremity movements. We developed two artificial
neural networks to model sustained elbow flexion: one trained
solely on recorded data (i.e., direct learning) and one
pre-trained on simulated data and fine-tuned on recorded data
(i.e., transfer learning). We simulated muscle activations and
joint torques using a musculoskeletal model and a muscle fatigue model (n = 1,701 simulations). We also recorded static
subject-specific features (e.g., anthropometric measurements) and
dynamic muscle activations and torques during sustained elbow flexion in healthy young adults (n = 25 subjects). Using the
simulated dataset, we pre-trained a long short-term memory neural
network (LSTM) to regress fatiguing elbow flexion torque from
muscle activations. We concatenated this pre-trained LSTM with a
feedforward architecture, and fine-tuned the model on recorded
muscle activations and static features to predict elbow flexion
torques. We trained a similar architecture solely on the recorded
data and compared each neural network's predictions on 5
leave-out subjects' data. The transfer learning model
outperformed the direct learning model, as indicated by a
decrease of 24.9% in their root-mean-square-errors (6.22 Nm and
8.28 Nm, respectively). The transfer learning model and direct
learning model outperformed analogous musculoskeletal
simulations, which consistently underpredicted elbow flexion
torque. Our results suggest that transfer learning from simulated
to recorded datasets can decrease reliance on assumptions
inherent to biomechanical models and yield predictions robust to
real-world conditions.
overlooked in biomechanical models. The present work leveraged a
transfer learning approach to improve torque predictions during
fatiguing upper extremity movements. We developed two artificial
neural networks to model sustained elbow flexion: one trained
solely on recorded data (i.e., direct learning) and one
pre-trained on simulated data and fine-tuned on recorded data
(i.e., transfer learning). We simulated muscle activations and
joint torques using a musculoskeletal model and a muscle fatigue model (n = 1,701 simulations). We also recorded static
subject-specific features (e.g., anthropometric measurements) and
dynamic muscle activations and torques during sustained elbow flexion in healthy young adults (n = 25 subjects). Using the
simulated dataset, we pre-trained a long short-term memory neural
network (LSTM) to regress fatiguing elbow flexion torque from
muscle activations. We concatenated this pre-trained LSTM with a
feedforward architecture, and fine-tuned the model on recorded
muscle activations and static features to predict elbow flexion
torques. We trained a similar architecture solely on the recorded
data and compared each neural network's predictions on 5
leave-out subjects' data. The transfer learning model
outperformed the direct learning model, as indicated by a
decrease of 24.9% in their root-mean-square-errors (6.22 Nm and
8.28 Nm, respectively). The transfer learning model and direct
learning model outperformed analogous musculoskeletal
simulations, which consistently underpredicted elbow flexion
torque. Our results suggest that transfer learning from simulated
to recorded datasets can decrease reliance on assumptions
inherent to biomechanical models and yield predictions robust to
real-world conditions.