A Lightweight Deep Learning-Based Approach for Concrete Crack Characterization Using Acoustic Emission Signals

Abstract

This paper proposes an acoustic emission (AE) based automated crack characterization method for reinforced concrete (RC) beams using a memory efficient lightweight convolutional neural network named SqueezeNet. The proposed method also includes a signal-to-image technique, which is continuous wavelet transformation (CWT) that decomposes the AE signals over time-frequency scales and extracts the crack/fracture information in both the time and frequency domains. First, AE signals for two types of cracks (minor and severe), along with the normal condition (no crack), are collected from the experimental test bed. Second, the previously mentioned CWT based signal-to-image technique is applied to generate twodimensional time-frequency images that are then converted to gray scale images for faster computation. These images are supplied to the SqueezeNet for classification of the concrete crack types. We extensively modified the fire module of the SqueezeNet (SQN-MF) by introducing depth-wise convolutional kernels and channel shuffling operations. Not only does the proposed method utilize deep learning-based techniques for crack classification of concrete beams for the first time, but also the CWT-based imaging technique has not yet been explored in this field either. Additionally, this method does not follow the typical AE burst feature (features like AE counts, peak-amplitude, rise time, decay time, etc.) based methods, and as a result, we no longer require extensive human intervention and expertise to get deep understanding of the crack types. SQN-MF achieves AlexNet-level accuracy with fifty times fewer parameters and has an implementable memory size for the field programmable gate array boards. Overall, the method achieves 100% accuracy. It is 20.8% higher than the typical feature extraction and traditional machine learning based methods. We observed a 4% accuracy increase for the proposed SQN-MF compared to the typical SqueezeNet with bypass connections. INDEX TERMSConcrete crack characterization, continuous wavelet transformation, convolutional neural network, SqueezeNet. MD ARAFAT HABIB was born in Rangpur, Bangladesh. He received his B.Sc. degree from BRAC University, Bangladesh. After that he joined Chosun University in South Korea to pursue his master's degree in computer engineering. His program was funded in part by Chosun University and National Research Foundation of Korea. During his master's program he has received Chosun University excellent student award based on academic performance and Brain Korea Scholarship. He is currently working at the University of Ulsan, South Korea. His current interests include artificial intelligence, signal processing, fault diagnosis, and condition monitoring of industrial machinery. MD JUNAYED HASAN received a B.Sc. degree in Computer Engineering from

Attention-Based DenseNet for Lung Cancer Classification Using CT Scan and Histopathological Images

“…Machine learning algorithms have changed the research paradigm in the whole community of researchers starting from image processing [14] and network optimization [15,16] to healthcare applications. Lung cancer refers to the formation of malignant cells in the lungs, leading to an overall rise in mortality rates for both men and women due to the increasing incidence of cancer.…”

Reinforcement Learning Based Mobile Underwater Localization for Silent UUV in Underwater Acoustic Sensor Networks

“…Neural networks have been also applied in different applications including signal recognition and classification. For instance, an efficient convolutional neural network (CNN) is used to classify the acoustic signals of reinforced concrete (RC), which outperforms typical feature extraction and traditional machine learning based methods [29]. A deep belief network based modulation recognition scheme for wireless signals as proposed in [30] reaches 92.12% recognition rate under high signal-to-noise.…”

“…Considering all these advantages, we choose SqueezeNet to classify underwater LOS/NLOS signals whose model size is only 0.5 MB. [29] SqueezeNet is composed of several Fire module combined with convolution layers, downsampling layers, and fully connected layers; the developers of which mainly adopted three strategies to obtain fewer parameters:…”

“…Since n i and n si are independent normal distributions, their sum is n ti ~Nð0, σ 2 i + σ 2: si Þ. Based on (28), (29), and (35), the FIM for FðζÞ is derived as (31); thus, CRLB for the localization by considering the movement of the target and optimal anchor nodes is derived.…”

Real-time surface damage analysis of concrete structures using Python-based image analysis

“…A lightweight Neural network using a memory-efficient lightweight CNN named SqueezeNet [8] has been used previously for crack detection in reinforced concrete beams. This CNN is a Continuous Wavelet Transform (CWT) based imaging process that monitors the crack development in reinforced concrete beams.…”