Machine learning based crack mode classification from unlabeled acoustic emission waveform features (original) (raw)
Related papers
Proceedings of the 10th International Conference on Fracture Mechanics of Concrete and Concrete Structures, 2019
Four point bending tests on reinforced concrete (RC) beam specimens were carried out and simultaneously released Acoustic Emissions (AE) were recorded in laboratory. This study reports on the AE characteristics of RC beams under monotonically increased loading. By using load-displacement curves and the AE signal parameters data, the fracture process of RC beams was studied. As the occurrence of AE events is random, a probabilistic approach named Gaussian Mixture Model (GMM) is implemented for AE data clustering related to tensile cracking and shear cracking considering different AE parameters. A supervised learning model named Support Vector Machine (SVM) procedure has been used to separate the two AE clusters belonging to tensile and shear cracks by constructing hyperplane to overcome the uncertainty. The yielding of testing specimen is compared. Influence of shear reinforcement in RC beam and type of loading was also considered in this study. The yielding of test specimen is compared for different types of loading pattern. The combination of both GMM of AE and SVM procedure to identify the exact place for separation of AE clusters are useful procedures for crack classification in concrete structures.
Monitoring the crack modes in concrete is of importance because the performance of the entire structural system is revealed. The cause and location of cracks is crucial to determine which type of crack is predominant. Assessment of failure or structural monitoring by non-destructive methods is desirable. Acoustic Emission (AE) method shows promising outcomes for monitoring cracks in concrete at real-time using some AE parameters like Rise Angle (RA) and Average Frequency (AF). This paper introduces a probabilistic approach based on Gaussian Mixture Modeling (GMM) to classify the crack modes based on the AE signals. The crack classification is checked for accuracy using Support Vector Machine (SVM) method. The algorithms are validated by an experimental study on concrete cylinders subjected to uniaxial compression.
Advanced Damage Detection Technique by Integration of Unsupervised Clustering into Acoustic Emission
The use of acoustic emission (AE) technique for damage diagnostic is typically challenging due to difficulties associated with discrimination of events that occur during different stages of damage that take place in a material or structure. In this study, an unsupervised kernel fuzzy c-means pattern recognition analysis and the principal component method were utilized to categorize various damage stages in plain and steel fiber reinforced concrete specimens monitored by AE technique. Enhancement of the discrimination and characterization of damage mechanisms were achieved by processing time and frequency domain data. Both domains (time and frequency) were taken into account to propose new descriptors for crack classification purposes. A cluster of AE data in three classes of Kernel Fuzzy c-means (KFCM) was obtained. The clustered data was subsequently correlated with each particular damage stage for identifying the peak frequency range corresponding to the respective damage stages. Moreover, a novel quantitative technique called Spatial Intelligent b-value (SIb) Analysis was proposed to quantify damage for each stage.
A Study of Supervised Machine Learning Techniques for Structural Health Monitoring
2015
We report on work that is part of the development of an agentbased structural health monitoring system. The data used are acoustic emission signals, and we classify these signals according to source mechanisms. The agents are proxies for communicationand computation-intensive techniques and respond to the situation at hand by determining an appropriate constellation of techniques. It is critical that the system have a repertoire of classifiers with different characteristics so that a combination appropriate for the situation at hand can generally be found. We use unsupervised learning for identifying the existence and location of damage but supervised learning for identifying the type and severity of damage. This paper reports on results for supervised learning techniques: support vector machines (SVMs), naive Bayes classifiers (NBs), feedforward neural networks (FNNs), and two kinds of ensemble learning, random forests and AdaBoost. We found the SVMs to be the most precise and the ...
Damage and repair classification in reinforced concrete beams using frequency domain data
Materials and Structures, 2015
This research aims at developing a new vibration-based damage classification technique that can efficiently be applied to a real-time large data. Statistical pattern recognition paradigm is relevant to perform a reliable site-location damage diagnosis system. By adopting such paradigm, the finite element and other inverse models with their intensive computations, corrections and inherent inaccuracies can be avoided. In this research, a two-stage combination between principal component analysis and Karhunen-Loéve transformation (also known as canonical correlation analysis) was proposed as a statistical-based damage classification technique. Vibration measurements from frequency domain were tested as possible damage-sensitive features. The performance of the proposed system was tested and verified on real vibration measurements collected from five laboratory-scale reinforced concrete beams modelled with various ranges of defects. The results of the system helped in distinguishing between normal and damaged patterns in structural vibration data. Most importantly, the system further dissected reasonably each main damage group into subgroups according to their severity of damage. Its efficiency was conclusively proved on data from both frequency response functions and response-only functions. The outcomes of this two-stage system showed a realistic detection and classification and outperform results from the principal component analysis-only. The success of this classification model is substantially tenable because the observed clusters come from well-controlled and known state conditions.
Crack classification in concrete beams using AE parameters
IOP conference series, 2017
The acoustic emission (AE) technique is an effective tool for the evaluation of crack growth. The aim of this study is to evaluate crack classification in reinforced concrete beams using statistical analysis. AE has been applied for the early monitoring of reinforced concrete structures using AE parameters such as average frequency, rise time, amplitude counts and duration. This experimental study focuses on the utilisation of this method in evaluating reinforced concrete beams. Beam specimens measuring 150 mm x 250 mm x 1200 mm were tested using a three-point load flexural test using Universal Testing Machines (UTM) together with an AE monitoring system. The results indicated that RA value can be used to determine the relationship between tensile crack and shear movement in reinforced concrete beams.
Sensors and Smart Structures Technologies for Civil, Mechanical, and Aerospace Systems 2013, 2013
Reinforced Concrete (RC) has been widely used in construction of infrastructures for many decades. The cracking behavior in concrete is crucial due to the harmful effects on structural performance such as serviceability and durability requirements. In general, in loading such structures until failure, tensile cracks develop at the initial stages of loading, while shear cracks dominate later. Therefore, monitoring the cracking modes is of paramount importance as it can lead to the prediction of the structural performance. In the past two decades, significant efforts have been made toward the development of automated structural health monitoring (SHM) systems. Among them, a technique that shows promises for monitoring RC structures is the acoustic emission (AE). This paper introduces a novel probabilistic approach based on Gaussian Mixture Modeling (GMM) to classify AE signals related to each crack mode. The system provides an early warning by recognizing nucleation of numerous critical shear cracks. The algorithm is validated through an experimental study on a full-scale reinforced concrete shear wall subjected to a reversed cyclic loading. A modified conventional classification scheme and a new criterion for crack classification are also proposed.