Figure 2 - uploaded by Jiaxing Ye
Content may be subject to copyright.
The ultrasonic wave propagation imaging (ultrasonic wave propagation imaging) snapshots and corresponding condition indexes.
Source publication
Ultrasonic wave propagation imaging enables the detection of anomalies in various structures; hence, it has been applied as one of the promising techniques for damage identification in structural health monitoring (SHM). The interpretation of imaging data is vital to SHM; however, it relies significantly on expert subjective judgment, rendering the...
Context in source publication
Context 1
... aim of this study is to develop an efficient machine learning system to discern flaw-induced patterns in an ultrasonic wave propagation video. Figure 2 illustrates an example for ease of understanding. A sequence of ultrasonic wavefield snapshots was used as the input to the machine learning system. ...
Citations
... Reusing the neural network for further inversions by incorporating prior knowledge into the neural network through transfer learning [84] is also subject to further research. The transfer learning can be in the form of supervised learning, such as [85][86][87][88][89][90][91][92][93][94][95][96][97][98][99] or by learning from previous full waveform inversions. The concept is demonstrated in a preliminary study in [100]. ...
... Operator learning, such as conducted with DeepONets [17] or FourierNets [101] provide yet another possibility. These can either replace the forward operator and accelerate inversions through faster and differentiable forward simulations, or learn the inverse mapping directly, as in [85][86][87][88][89][90][91][92][93][94][95][96][97][98][99]. Table 3 Comparison of the investigated methods from Table 1. ...
Neural networks have recently gained attention in the context of solving inverse problems. Physics-Informed Neural
Networks (PINNs) are a prominent methodology for the task of solving both forward and inverse problems. In the paper
at hand, full waveform inversion is the inverse problem under consideration. The performance of PINNs is compared against
classical adjoint optimization. The comparison focuses on three key aspects: the forward-solver, the representation of the
material, and the sensitivity computation for the gradient-based minimization. Starting from PINNs, each of these key aspects
is investigated and adapted individually until the classical adjoint optimization emerges. It is shown that it is beneficial to use
the neural network only for the discretization of the unknown spatially varying material field. Here the neural network produces
reconstructions without oscillatory artifacts as typically encountered in classical full waveform inversion approaches. Due to
this finding, a hybrid method is proposed. It exploits both the efficient gradient computation with the continuous adjoint method
as well as the neural network ansatz for the unknown material field. This new hybrid method outperforms Physics-Informed
Neural Networks and the classical adjoint optimization in two-dimensional and three-dimensional settings.
... Applications of supervised learning in the context of ultrasonic nondestructive testing can e.g. be found in [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18]. The second methodology are physics-informed neural networks (PINNs) [19], [20], [21], where one network learns the forward field (i.e. ...
... This observation offers the chance to exploit the concept of transfer learning [27] for a better starting pointγ (0) obtained by better starting weights θ (0) . The neural network is pretrained in a supervised manner [18], [17] with data pairs (u M , γ M ) using a data-driven cost function defined in terms of the mean squared error ...
... Applications of supervised learning in the context of ultrasonic nondestructive testing can e.g. be found in [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21]. The second methodology are physics-informed neural networks (PINNs) [22], [23], [24], where one network learns the forward field (i.e. ...
... This observation offers the chance to exploit the concept of transfer learning [30] for a better starting pointγ (0) obtained by better starting weights θ (0) . The neural network is pretrained in a supervised manner [21], [20] with data pairs (u M , γ M ) using a data-driven cost function defined in terms of the mean squared error ...
We propose a way to favorably employ neural networks in the field of non-destructive testing using Full Waveform Inversion (FWI). The presented methodology discretizes the unknown material distribution in the domain with a neural network within an adjoint optimization. To further increase efficiency of the FWI, pretrained neural networks are used to provide a good starting point for the inversion. This reduces the number of iterations in the Full Waveform Inversion for specific, yet generalizable settings.
... Recently, machine learning algorithms have been applied to analyze ultrasonic signals [176,177]. In [178], ultrasonic test data were used to train six ML models predicting the degree of corrosion in reinforced concrete based on ultrasonic traits. Results showed that ML models could produce accurate and robust predictions of corrosion levels in the presence of outlier amplitudes and for training sets of varying sizes. ...
This paper reviews recent advances in sensor technologies for non-destructive testing (NDT) and structural health monitoring (SHM) of civil structures. The article is motivated by the rapid developments in sensor technologies and data analytics leading to ever-advancing systems for assessing and monitoring structures. Conventional and advanced sensor technologies are systematically reviewed and evaluated in the context of providing input parameters for NDT and SHM systems and for their suitability to determine the health state of structures. The presented sensing technologies and monitoring systems are selected based on their capabilities, reliability, maturity, affordability, popularity, ease of use, resilience, and innovation. A significant focus is placed on evaluating the selected technologies and associated data analytics, highlighting limitations, advantages, and disadvantages. The paper presents sensing techniques such as fiber optics, laser vibrometry, acoustic emission, ultrasonics, thermography, drones, microelectromechanical systems (MEMS), magnetostrictive sensors, and next-generation technologies.
... Wu Q et al [21] used an improved multilayer perceptron neural network to achieve leak localization for gas pipelines. Ye J et al [22] proposed automatic defect detection for ultrasonic wave propagation imaging method using spatiotemporal convolution neural networks. Wang X et al [23] proposed a rapid guided wave imaging method based on CNN to quantitatively evaluate the corrosion damage. ...
In this paper, a real-time leak source localization method based on convolutional neural network (CNN) of elastic wavefield images and spatio-temporal correlation analysis (STCA) is developed for the pressure vessel leakage. This method uses a single sensor array coupled to the wall to collect the elastic wave data excited by the leak source. Besides, the distance
R
and the direction θ between the leak source and the sensor array are calculated based on CNN and STCA respectively, to finally obtain the location (
R
, θ) of the leak source. In this paper, the digital twin model of the experimental platform is established, the training set is obtained by the finite element simulation, and the CNN model applied to the elastic wavefield images is studied and constructed. The experimental results show that the maximum locating error is 1.46 cm and the average locating error is about 0.56 cm within the range of a 1 m
<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>
experimental plate based on the method proposed in this paper.
The rapid growth of deep learning research, including within the field of computational mechanics, has resulted in an extensive and diverse body of literature. To help researchers identify key concepts and promising methodologies within this field, we provide an overview of deep learning in deterministic computational mechanics. Five main categories are identified and explored: simulation substitution, simulation enhancement, discretizations as neural networks, generative approaches, and deep reinforcement learning. This review focuses on deep learning methods rather than applications for computational mechanics, thereby enabling researchers to explore this field more effectively. As such, the review is not necessarily aimed at researchers with extensive knowledge of deep learning—instead, the primary audience is researchers on the verge of entering this field or those attempting to gain an overview of deep learning in computational mechanics. The discussed concepts are, therefore, explained as simple as possible.
This study presented an unsupervised anomaly detection-based framework for distributed damage detection in concrete using ultrasonic response signals. A deep fully connected auto-encoder was employed to reconstruct the ultrasonic response signals. This model was trained on the intact specimen’s responses. The auto-encoder demonstrated a relatively high prediction error encountering the damaged specimen’s responses. Two time-domain features (mean squared error and reconstructed-to-original signal ratio) and one frequency-domain feature (fundamental amplitude ratio) were defined to measure the reconstruction error of the auto-encoder (the damage-sensitive features). Finally, the Isolation Forest algorithm was implemented for anomaly (damage) detection. The beauty of this framework is that it requires a few numbers of data only from the intact specimen for training the auto-encoder and collecting the binary decision trees of the Isolation Forest. The framework was successfully implemented for damage detection in five geopolymer concrete specimens with different mix proportions. Using all three introduced damage-sensitive features, the framework demonstrated an average prediction accuracy of 95.0% and 93.0% for damaged and intact stages, respectively.
