STR-903: UNSUPERVISED NOVELTY DETECTION BASED STRUCTURAL DAMAGE DETECTION METHOD

Many structural damage detection methods using machine learning algorithms and clustering methods have been proposed and developed in recent years. Novelty detection is a common method that is based on an unsupervised learning technique to detect structural damage. The detection process involves applying the novelty detection algorithm to recognize abnormal data from the testing data sets. In order to make these algorithms capable of identifying abnormal data, sufficient normal data must first be obtained and used as training data. It is the fact that sufficient normal data is relatively convenient to measure compared to abnormal data for large-scale civil structures. Abnormal data from the testing data sets can be identified by using the well-trained normal model established by the algorithms. In this paper, a machine learning based novelty detection method called the Density Peaks based Fast Clustering Algorithm (DPFCA) is introduced and some improvements to this algorithm are made to increase the precision of detecting and localizing the damage in an experimental structure. Feature extraction is also an important factor in the process of damage detection. Thus, two damage-sensitive features such as crest factor, and transmissibility are extracted from the measured responses in the experiments. Experimental results showed good performance of the innovative method in detecting and locating the structural damage positions in various scenarios.

[1]  R. J. Barthorpe,et al.  On model- and data-based approaches to structural health monitoring , 2010 .

[2]  Charles R. Farrar,et al.  Structural Health Monitoring: A Machine Learning Perspective , 2012 .

[3]  Yu Ding,et al.  A Computable Plug-In Estimator of Minimum Volume Sets for Novelty Detection , 2010, Oper. Res..

[4]  Hoon Sohn,et al.  A Review of Structural Health Review of Structural Health Monitoring Literature 1996-2001. , 2002 .

[5]  Oral Büyüköztürk,et al.  Structural Damage Detection Using Modal Strain Energy and Hybrid Multiobjective Optimization , 2015, Comput. Aided Civ. Infrastructure Eng..

[6]  B. Schwarz,et al.  Measurements Required for Displaying Operating Deflection Shapes , 2004 .

[7]  Sean Hughes,et al.  Clustering by Fast Search and Find of Density Peaks , 2016 .

[8]  Oral Büyüköztürk,et al.  Automated Structural Damage Detection Using One-Class Machine Learning , 2014 .

[9]  Ammar Belatreche,et al.  An experimental evaluation of novelty detection methods , 2014, Neurocomputing.

[10]  Samir Mustapha,et al.  Robust dimensionality reduction and damage detection approaches in structural health monitoring , 2014 .

[11]  W. T. Yeung,et al.  Damage detection in bridges using neural networks for pattern recognition of vibration signatures , 2005 .

[12]  Hoon Sohn,et al.  Damage diagnosis under environmental and operational variations using unsupervised support vector machine , 2009 .

[13]  Christian Cremona,et al.  Static-based early-damage detection using symbolic data analysis and unsupervised learning methods , 2015 .

[14]  Michel Verleysen,et al.  Width optimization of the Gaussian kernels in Radial Basis Function Networks , 2002, ESANN.

[15]  Vishnuvardhan Janapati,et al.  A novel machine-learning approach for structural state identification using ultrasonic guided waves , 2014 .

[16]  Feiping Nie,et al.  Learning a Mahalanobis distance metric for data clustering and classification , 2008, Pattern Recognit..

[17]  Jaakko Hollmén,et al.  Feature Extraction and Selection from Vibration Measurements for Structural Health Monitoring , 2009, IDA.