An unsupervised learning approach by novel damage indices in structural health monitoring for damage localization and quantification

Abstract

The aim of this article is to propose novel damage indices for damage localization and quantification based on time series modeling. In order to extract damage-sensitive features from time series models, it is essential to choose adequate and robust orders in such a way that the models are able to extract uncorrelated residuals. On this basis, a new iterative order determination method is proposed to select robust orders of time series models under residual analysis by Ljung–Box Q-test. The damage-sensitive features are the parameters and residuals of an AutoRegressive (AR) model obtained from current feature extraction approaches. In this study, the AR model is identified as the most compatible time series model with measured vibration time-domain responses using Box–Jenkins methodology and Leybourne–McCabe hypothesis test. The proposed damage indices are the parametric assurance criterion and the residual reliability criterion that exploit the parameters and residuals of AR models, respectively. The main idea behind locating a damage is to define threshold limits for both damage indices using the features of undamaged conditions based on an unsupervised learning way. The major contributions of this article are to propose an iterative order determination method for time series models and two novel damage indices for locating and quantifying damage. The accuracy and performance of the proposed methods are experimentally demonstrated on a three-story laboratory frame and a model-scale steel structure. Results show that the proposed iterative approach leads to uncorrelated residuals, and the proposed parametric assurance criterion and the residual reliability criterion methods are promising and efficient tools in damage detection problems under varying operational and environmental conditions.

Vibration-based damage detection using online learning algorithm for output-only structural health monitoring

“…Since the natural frequencies are temporal characteristic of the system, the proposed method with the natural frequencies cannot provide the information about damage location. Time-series models (such as AR model 40 and AR-ARX model 41 ) or some features from wavelet transform 42 can be used as DSF for the proposed method. These features contain the spatial information from each sensor, and the proposed method can be applied to establish the monitoring chart for each sensor.…”

Fully convolutional networks for structural health monitoring through multivariate time series classification

“…• supervised, when a label corresponding to one of the possible outputs of the classification task is associated to each structural response; • unsupervised [11], when no labelling is available;…”

An Unsupervised Learning Approach for Early Damage Detection by Time Series Analysis and Deep Neural Network to Deal with Output-Only (Big) Data

“…The central core of all these methods relies upon statistical pattern recognition, and comprises feature extraction and feature classification. The former step is a signal processing strategy, which aims at extracting meaningful information (here called damage-sensitive features) from raw measured data (e.g., acceleration time histories), while the latter is a machine learning algorithm for analyzing and classifying the extracted features for early damage detection, localization and quantification [4][5][6][7]. Time series modeling is one of the powerful feature extraction methods, which is intended to fit a parametric representation (model) to raw measured data [8,9].…”