1

Dr. Bidyut Kr. Patra

Associate Professor

Department of Computer Science & Engineering

Indian Institute of Technology (BHU), Varanasi

2

• Introduction to Unsupervised Learning
• Data Clustering
• Anomaly (Outlier) Detection
• LOF: Traditional Unsupervised Techniques for Anomaly Detection
• Unsupervised Neural Model
• Background
• Autoencoder
• Variational Autoencoder
• Unsupervised Deep Learning for Structural Health Monitoring.

​

Topics to be discussed

3

What is Machine Learning?

• Tom M. Mitchell (1997): "A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P, if its performance at tasks in T, as measured by P, improves with experience E."

• Machine Learning is the systematic study of algorithms and systems that improves their knowledge or performance with experience.

What is Learning?

• An agent (Entity) is said to be learning if it improves its performance after making observations about the environment.

4

• Supervised Learning

​

• Unsupervised Learning

​

​

​

​

5

6

7

• Structural Health Monitoring

8

9

10

11

12

• Select k points as initial centroids.
• repeat

​

• Form k clusters by assigning each point to its closest centroid.

​

• Recompute the centroid of the each cluster.
• until Centroids do not change.

k-means Clustering

13

14

15

16

17

18

19

20

21

22

Unsupervised Outlier Detection Techniques

23

What is an outlier or anomaly?

Outlier or anomaly is an exceptional objects that deviate from the rest of the data set.

​

Definition: (Hawkins-Outlier) An outlier is an observation that deviates so much from other observations that it arouses the suspicion that it was generated by a different mechanism.

​

24

25

26

27

28

LOF: identifying density-based local outliers

• LOF is a density-based technique for finding outliers. It declares a point as inlier if it surrounds with a dense neighborhood. Object having a sparse neighborhood is treated as an outlier.

​

• LOF computes the neighborhood density of each object and the LOF score of each object acts as a deciding factor for outlier.

​

• For finding the LOF score, it finds k-distance of an object, reachability distance and local reachability density.

Breunig, M. M., Kriegel, H. P., Ng, R. T., Sander, J.: LOF: identifying density-based local outliers. In: ACM sigmod record, Vol. 29, pp. 93-104. ACM (2000).

29

• k-distance(o): k-distance of an object o can be defined as the k^th nearest neighbor distance of the object o, where k is a positive integer.

​

​

• Reachability distance of an object p w.r.t. object o:

reach-dist_k (p, o) = max{k-distance(o), d(p, o)} || d(.) distance function.

reach-dist₄(p1,o) and reach-dist₄(p2,o)

30

• local reachability density_k(p):

• The local outlier factor LOF (p), is defined as follows:

31

32

33

Unsupervised Neural Network Model

• Principal Component Analysis (PCA)
• Neural Model Autoencoder and Its Variants
• Application of Autoencoder based Outlier Detection for SHM.

34

Latent Variable Models

35

36

Principal Component Analysis (PCA)

37

38

residual or loss:

39

40

41

42

43

44

45

46

47

Autoencoders

48

49

50

51

52

53

54

Variational Autoencoders

55

56

57

Application of Autoencoder

58

Abhaya and Bidyut Kr. Patra. An Efficient Method for Autoencoder based Outlier Detection, Expert Systems With Applications, Volume 213, Part A, March, 118904, 2023.

59

60

61

62

63

Unsupervised structural health monitoring framework

​

64

Eduardo M. Coraça, Janito V. Ferreira, Eurípedes G.O. Nóbrega. An unsupervised structural health monitoring framework based on Variational Autoencoders and Hidden Markov Models, Volume 231, March 2023, 109025.

65

Eduardo M. Coraça, Janito V. Ferreira, Eurípedes G.O. Nóbrega. An unsupervised structural health monitoring framework based on Variational Autoencoders and Hidden Markov Models, Volume 231, March 2023, 109025.

66

67

Roberto Boccagna, Maurizio Bottini, Massimo Petracca, Alessia Amelio and Guido Camata. Unsupervised Deep Learning for Structural Health Monitoring, Big Data Cogn. Comput. 2023, 7, 99.

​

• The proposed approach is based on unsupervised deep learning algorithm, with the aim of establishing a reliable method of anomaly detection for data acquired from sensors positioned on buildings.

​

• The method tested on data generated from an OpenSees numerical model of a railway bridge and data acquired from physical sensors positioned on the Historical Tower of Ravenna (Italy).

​

68

Autoencoder used in this study.

69

Schematic representation of the variational autoencoder used in this study.

70

• Cha, Y.-J.; Wang, Z. Unsupervised novelty detection–based structural damage localization using a density peaks-based fast clustering algorithm. Struct. Health Monit. 2018, 17, 313–324.

71

Example of local density calculation.

72

73

74

75

76

77

Kareem Eltouny , Mohamed Gomaa and Xiao Liang. Unsupervised Learning Methods for Data-Driven Vibration-Based Structural Health Monitoring: A Review, Sensors 2023,

23, 3290.

​

78

• Fernandez-Navamuel, A.Magalhães, F. Zamora-Sánchez, D. Omella, A.J., Garcia-Sanchez, D., Pardo, D. Deep learning enhanced principal component analysis for structural health monitoring. Struct. Health Monit. 2022, 21, 1710–1722.
• Kim, M.; Song, J. Near-real-time identification of seismic damage using unsupervised deep neural network. J. Eng. Mech. 2022, 148, 04022006.
• Jiang, K.; Han, Q.; Du, X.; Ni, P. A decentralized unsupervised structural condition diagnosis approach using deep auto-encoders, Comput.-Aided Civ. Infrastruct. Eng. 2021, 36, 711–732
• Silva, M.F.; Santos, A.; Santos, R.; Figueiredo, E.; Costa, J.C. Damage-sensitive feature extraction with stacked autoencoders for unsupervised damage detection. Struct. Control Health Monit. 2021, 28, e2714

​

79

• Wang, Z.; Cha, Y.-J. Unsupervised deep learning approach using a deep auto-encoder with a one-class support vector machine to detect damage. Struct. Health Monit. 2021, 20, 406–425.
• Yuan, Z.; Zhu, S.; Chang, C.; Yuan, X.; Zhang, Q.; Zhai, W. An unsupervised method based on convolutional variational auto-encoder and anomaly detection algorithms for light rail squat localization. Constr. Build. Mater. 2021, 313, 125563.

​

80

Thank you very much for your attention and time.