Overview of ANN Algorithm Used in Structural Health Monitoring.pdf

Useful for civil engineering research students

1. Department of Civil Engineering
SHARAD INSTITUTE OF TECHNOLOGY COLLEGE OF ENGINEERING
YADRAV, KOLHAPUR, MAHARASHTRA -416121.
Overview of ANN Algorithm Used in Structural Health
Monitoring
Dr. Maloth Naresh

2. Flow of presentation
1 Introduction
2
Components of the ANN3
4 Advantages of ANN
References6
5 Disadvantages of ANN
Artificial neural network (ANN)

3. Introduction
•By identifying and diagnosing structural degradation before it results in catastrophic
failures, structural health monitoring(SHM), is essential to maintaining the lifetime and
safety of infrastructure. Conventional SHM methods use physics-based models and
hand inspections, which can be costly, time-consuming, and less effective for large-scale
structures. Machine learning (ML) methods have become a potent tool for automating
damage identification, localization, and prognosis in SHM in recent years. ML models
are perfect for real-time monitoring of buildings, bridges, wind turbines, and other vital
infrastructure because they can handle enormous volumes of sensor data, identify
intricate patterns, and improve prediction skills. With an emphasis on important
algorithms like artificial neural networks (ANN), support vector machines (SVM),
random forests, deep learning architectures, and hybrid models, as well as their uses,
difficulties, and potential future research areas, this literature review examines the
developments in ML-based SHM [1].

4. Artificial Neural Network (ANN) algorithm
•The human brain served as inspiration for ANN, which
are computer models made up of linked layers of
neurons that analyze and learn from their input (Fig. 1).
•SHM often uses ANN to evaluate sensor data for
damage diagnosis, pattern identification, and predictive
maintenance. It works well for detecting intricate
structural abnormalities because of its proficiency with
nonlinear interactions.
•Researchers have used various ANN topologies,
including feedforward, convolutional, and recurrent
neural networks, to improve SHM accuracy.
Notwithstanding their benefits, ANN models need a lot
of computing power and big datasets for training and
optimization [2,3].
Fig. 1: Workflow of ANN.

5. Components of the ANN algorithm
•Various essential parts constituteANN, which processesinputdata and providesresults.
The primary elements consist of [4,5]:
1. Node, or neurons
•An ANN's fundamental processing unit is modeled after biological neurons.
After processing input using an activation function, each neuron sends its output to the
layer below.
2. Layers of ANN
•Three primary layer types constitute the framework of ANNs:
(a) Input layer, (b) Hidden layer, (c) Output layer

6. Cont…
(a)Input layer:
•The first layer receives raw input data (features).
•Each neuron in this layer represents a feature (such as displacement, strain, or
acceleration in SHM).
(b) Hidden layer:
• In between the input and output layers are intermediate layers.
•To discover patterns, do calculations with weighted connections and activation
functions.
•There are several hidden layers in deep learning models.
(c) Output layer:
•Generates the final categorization or prediction outcome.
•The task determines how many neurons are needed.

7. Cont…
3. Weights:
•There is a weight attached to each neuronal connection.
•Weights determine the significance of each input in forecasting the result.
•To reduce prediction mistakes, weights are changed throughout training.
4. Bias:
• An extra parameter that helps change the activation function to improve learning.
•Altering the decision boundary, it enables the model to match the data more
accurately.
5. Activation function:
•Gives the model non-linearity, which allows it to recognize intricate patterns.
•Commonly utilized activation functions are ReLU, sigmoid, tanh, and softmax.

8. Advantages of ANN
•Capacity to Learn Complex Patterns: ANNs are very useful for Structural Health
Monitoring (SHM) because they can model complex and nonlinear correlations
in data.
•High Prediction Accuracy: ANNs can detect structural irregularities and
anticipate failures with greater accuracy if they have access to enough training
data.
•Adaptive and Self-Learning: By modifying weights during training, ANN may
gradually enhance its performance, adapting to changing structural
circumstances.
•Efficient at Managing Big Datasets: ANNs are capable of effectively processing
large volumes of sensor data and extracting valuable characteristics for damage
diagnosis.
•Robustness for Noisy Data: ANN models have the innate capacity to learn, which
allows them to generalize well even when trained on incomplete or noisy datasets
[6,7].

9. Disadvantages of ANN
•High Computational Cost: ANN training uses a lot of computer power.
•Data dependency: For ANNs to train well, big, labeled datasets are necessary.
•Overfitting risk: An ANN may overfit training data if it is not appropriately
regularised, which would limit its capacity to generalize to novel situations.
•Choice of Hyperparameters: The hyperparameters (like the number of layers,
neurons, and learning rate) has a big effect on how well an ANN works and
needs to be fine-tuned a lot [2,8,9].

10. References
1.Avci, O., Abdeljaber, O., Kiranyaz, S., Hussein, M., Gabbouj, M., & Inman, D. J. (2021). A review of vibration-based
damage detection in civil structures: From traditional methods to Machine Learning and Deep Learning
applications.Mechanical systems and signal processing,147, 107077.
2.Dadras Eslamlou, A., & Huang, S. (2022). Artificial-neural-network-based surrogate models for structural health
monitoring of civil structures: A literature review.Buildings,12(12), 2067.
3.Gharehbaghi, V. R., Noroozinejad Farsangi, E., Noori, M., Yang, T. Y., Li, S., Nguyen, A., ... & Mirjalili, S. (2022). A
critical review on structural health monitoring: Definitions, methods, and perspectives.Archives of computational
methods in engineering,29(4), 2209-2235.
4.Lopes Jr, V., Park, G., Cudney, H. H., & Inman, D. J. (2000). Impedance-based structural health monitoring with
artificial neural networks.Journal of Intelligent Material Systems and Structures,11(3), 206-214.
5.HekmatiAthar, S., Taheri, M., Secrist, J., & Taheri, H. (2020). Neural network for structural health monitoring with
combined direct and indirect methods.Journal of Applied Remote Sensing,14(1), 014511-014511.
6.Etim, B., Al-Ghosoun, A., Renno, J., Seaid, M., & Mohamed, M. S. (2024). Machine learning-based modeling for
structural engineering: A comprehensive survey and applications overview.Buildings,14(11), 3515.
7.Altabey, W. A., & Noori, M. (2022). Artificial-intelligence-based methods for structural health monitoring.Applied
Sciences,12(24), 12726.
8.Gomez-Cabrera, A., & Escamilla-Ambrosio, P. J. (2022). Review of machine-learning techniques applied to structural
health monitoring systems for building and bridge structures.Applied Sciences,12(21), 10754.
9.Hassani, S., & Dackermann, U. (2023). A systematic review of advanced sensor technologies for non-destructive
testing and structural health monitoring.Sensors,23(4), 2204.

11. Thank you