Robust feature extraction for rapid classification of damage in composites

Clyde K. Coelho, Whitney Reynolds, Aditi Chattopadhyay

Research output: Chapter in Book/Report/Conference proceedingConference contribution

3 Citations (Scopus)

Abstract

The ability to detect anomalies in signals from sensors is imperative for structural health monitoring (SHM) applications. Many of the candidate algorithms for these applications either require a lot of training examples or are very computationally inefficient for large sample sizes. The damage detection framework presented in this paper uses a combination of Linear Discriminant Analysis (LDA) along with Support Vector Machines (SVM) to obtain a computationally efficient classification scheme for rapid damage state determination. LDA was used for feature extraction of damage signals from piezoelectric sensors on a composite plate and these features were used to train the SVM algorithm in parts, reducing the computational intensity associated with the quadratic optimization problem that needs to be solved during training. SVM classifiers were organized into a binary tree structure to speed up classification, which also reduces the total training time required. This framework was validated on composite plates that were impacted at various locations. The results show that the algorithm was able to correctly predict the different impact damage cases in composite laminates using less than 21 percent of the total available training data after data reduction.

Original languageEnglish (US)
Title of host publicationProceedings of SPIE - The International Society for Optical Engineering
Volume7286
DOIs
StatePublished - 2009
EventModeling, Signal Processing, and Control for Smart Structures 2009 - San Diego, CA, United States
Duration: Mar 11 2009Mar 12 2009

Other

OtherModeling, Signal Processing, and Control for Smart Structures 2009
CountryUnited States
CitySan Diego, CA
Period3/11/093/12/09

Fingerprint

pattern recognition
Feature Extraction
Support vector machines
Feature extraction
education
Damage
Composite
Discriminant analysis
damage
Support Vector Machine
composite materials
Composite Plates
Composite materials
Discriminant Analysis
Binary trees
Damage detection
Structural health monitoring
Sensors
impact damage
Piezoelectric Sensor

Keywords

  • Binary tree
  • Carbon fiber composite
  • Damage classification
  • Feature reduction
  • Impact
  • Linear discriminant analysis
  • Structural health monitoring
  • Support vector machines

ASJC Scopus subject areas

  • Applied Mathematics
  • Computer Science Applications
  • Electrical and Electronic Engineering
  • Electronic, Optical and Magnetic Materials
  • Condensed Matter Physics

Cite this

Coelho, C. K., Reynolds, W., & Chattopadhyay, A. (2009). Robust feature extraction for rapid classification of damage in composites. In Proceedings of SPIE - The International Society for Optical Engineering (Vol. 7286). [72860K] https://doi.org/10.1117/12.815903

Robust feature extraction for rapid classification of damage in composites. / Coelho, Clyde K.; Reynolds, Whitney; Chattopadhyay, Aditi.

Proceedings of SPIE - The International Society for Optical Engineering. Vol. 7286 2009. 72860K.

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Coelho, CK, Reynolds, W & Chattopadhyay, A 2009, Robust feature extraction for rapid classification of damage in composites. in Proceedings of SPIE - The International Society for Optical Engineering. vol. 7286, 72860K, Modeling, Signal Processing, and Control for Smart Structures 2009, San Diego, CA, United States, 3/11/09. https://doi.org/10.1117/12.815903
Coelho CK, Reynolds W, Chattopadhyay A. Robust feature extraction for rapid classification of damage in composites. In Proceedings of SPIE - The International Society for Optical Engineering. Vol. 7286. 2009. 72860K https://doi.org/10.1117/12.815903
Coelho, Clyde K. ; Reynolds, Whitney ; Chattopadhyay, Aditi. / Robust feature extraction for rapid classification of damage in composites. Proceedings of SPIE - The International Society for Optical Engineering. Vol. 7286 2009.
@inproceedings{1cb11c85355240af8f4e09aea91cfee3,
title = "Robust feature extraction for rapid classification of damage in composites",
abstract = "The ability to detect anomalies in signals from sensors is imperative for structural health monitoring (SHM) applications. Many of the candidate algorithms for these applications either require a lot of training examples or are very computationally inefficient for large sample sizes. The damage detection framework presented in this paper uses a combination of Linear Discriminant Analysis (LDA) along with Support Vector Machines (SVM) to obtain a computationally efficient classification scheme for rapid damage state determination. LDA was used for feature extraction of damage signals from piezoelectric sensors on a composite plate and these features were used to train the SVM algorithm in parts, reducing the computational intensity associated with the quadratic optimization problem that needs to be solved during training. SVM classifiers were organized into a binary tree structure to speed up classification, which also reduces the total training time required. This framework was validated on composite plates that were impacted at various locations. The results show that the algorithm was able to correctly predict the different impact damage cases in composite laminates using less than 21 percent of the total available training data after data reduction.",
keywords = "Binary tree, Carbon fiber composite, Damage classification, Feature reduction, Impact, Linear discriminant analysis, Structural health monitoring, Support vector machines",
author = "Coelho, {Clyde K.} and Whitney Reynolds and Aditi Chattopadhyay",
year = "2009",
doi = "10.1117/12.815903",
language = "English (US)",
isbn = "9780819475466",
volume = "7286",
booktitle = "Proceedings of SPIE - The International Society for Optical Engineering",

}

TY - GEN

T1 - Robust feature extraction for rapid classification of damage in composites

AU - Coelho, Clyde K.

AU - Reynolds, Whitney

AU - Chattopadhyay, Aditi

PY - 2009

Y1 - 2009

N2 - The ability to detect anomalies in signals from sensors is imperative for structural health monitoring (SHM) applications. Many of the candidate algorithms for these applications either require a lot of training examples or are very computationally inefficient for large sample sizes. The damage detection framework presented in this paper uses a combination of Linear Discriminant Analysis (LDA) along with Support Vector Machines (SVM) to obtain a computationally efficient classification scheme for rapid damage state determination. LDA was used for feature extraction of damage signals from piezoelectric sensors on a composite plate and these features were used to train the SVM algorithm in parts, reducing the computational intensity associated with the quadratic optimization problem that needs to be solved during training. SVM classifiers were organized into a binary tree structure to speed up classification, which also reduces the total training time required. This framework was validated on composite plates that were impacted at various locations. The results show that the algorithm was able to correctly predict the different impact damage cases in composite laminates using less than 21 percent of the total available training data after data reduction.

AB - The ability to detect anomalies in signals from sensors is imperative for structural health monitoring (SHM) applications. Many of the candidate algorithms for these applications either require a lot of training examples or are very computationally inefficient for large sample sizes. The damage detection framework presented in this paper uses a combination of Linear Discriminant Analysis (LDA) along with Support Vector Machines (SVM) to obtain a computationally efficient classification scheme for rapid damage state determination. LDA was used for feature extraction of damage signals from piezoelectric sensors on a composite plate and these features were used to train the SVM algorithm in parts, reducing the computational intensity associated with the quadratic optimization problem that needs to be solved during training. SVM classifiers were organized into a binary tree structure to speed up classification, which also reduces the total training time required. This framework was validated on composite plates that were impacted at various locations. The results show that the algorithm was able to correctly predict the different impact damage cases in composite laminates using less than 21 percent of the total available training data after data reduction.

KW - Binary tree

KW - Carbon fiber composite

KW - Damage classification

KW - Feature reduction

KW - Impact

KW - Linear discriminant analysis

KW - Structural health monitoring

KW - Support vector machines

UR - http://www.scopus.com/inward/record.url?scp=69649093944&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=69649093944&partnerID=8YFLogxK

U2 - 10.1117/12.815903

DO - 10.1117/12.815903

M3 - Conference contribution

AN - SCOPUS:69649093944

SN - 9780819475466

VL - 7286

BT - Proceedings of SPIE - The International Society for Optical Engineering

ER -