Inspection of wind turbine blades using image deblurring and deep learning segmentationLu, Jiale; Gao, Qingbin; Zhou, Kai
doi: 10.1117/12.3009721pmid: N/A
Remote and complex work sites of wind turbines limit the accessibility of the condition assessment. Wind turbine blades are subject to sustained wind load and harsh natural environmental conditions, which are vulnerable to various faults. Robotic-enabled sensing technology appears to be a promising solution for an efficient wind turbine blade inspection. Together with the recent advances in image processing and deep learning segmentation, automated inspection of wind turbine blades becomes possible. Nevertheless, it remains a challenging task to quantify the damage accurately due to the complex condition of images concerning motion blurs. To address this issue, an integrated framework, i.e., the combination of a Deblur Generative Adversarial Network v2 (DeblurGAN-v2) and You Only Look Once v8 (YOLO-v8) was proposed in this study. Specifically, the mapping between the motion-blurred images and those in high quality was adopted from the open-access pretrained DeblurGAN-v2, based on which the deblurring performance for wind turbine images with various motion blur scales was discussed concerning the image quality. Subsequently, the transfer learning method was implemented to fine-tune YOLO-v8. The well-trained YOLO v8 was then utilized for target defect segmentation on the deblurred images. Finally, various metrics were calculated to evaluate the segmentation accuracy and efficiency. Results prove a good generalization of DeblurGAN-v2 on wind turbine images and clearly illustrate the enhanced performance of the proposed methodology especially when the motion blur scale is within 35. The integrated framework could serve as a reference for dealing with other fuzzy image-related issues.
Normal data-based motor fault diagnosis using stacked time-series imaging methodJung, W.; Lim, D. G.; Lim, B. H.; Park, Y. H.
doi: 10.1117/12.3025103pmid: N/A
In most engineering systems, the acquisition of faulty data is difficult or sometimes not feasible, while normal data are secured. To solve these problems, this paper proposes an fault diagnosis method for electric motor using only normal data with self-labeling based on stacked time-series imaging method. Since only normal data are used for fault diagnosis, a self-labeling method is used to generate a new labeled dataset based on pretext task. To emphasize faulty features from non-stationary faulty data, stacked time-series imaging method is developed. The overall procedure includes the following steps: (1) transformation of a one-dimensional current signal to a two-dimensional image in time-domain, (2) adding sparse features with sparse dictionary learning, (3) stacked images through every window size, and (4) fault classification based on Convolutional Neural Network (CNN) and Mahalanobis distance. Transformation of the time-series signal is based on Recurrence Plots (RP). The proposed RP method develops from sparse dictionary learning that provides the dominant fault feature representations in a robust way. To verify the proposed method, data from real-field manufacturing line is used.
Design and investigation of polymer-based terahertz nearfield imaging probes for the high-resolution nondestructive imaging applicationsMuthuramalingam, Karthickraj; Wang, Wei-Chih
doi: 10.1117/12.3014670pmid: N/A
This paper presents a comprehensive investigation of solid and hollow polymer-based near-field imaging probes, each coated with a metallic layer on the outer surface, designed to operate in the THz frequency range. These probes are tailored to exploit the near-field properties of THz radiation for achieving sub-wavelength resolution imaging. The proposed probes exhibit a versatile design that has been rigorously examined through advanced electromagnetic simulations. The solid probe focuses on exploiting the dielectric properties of the material to manipulate THz radiation. Conversely, the hollow probe leverages its cavity structure to create resonant modes within the THz frequency range. This resonance phenomenon enhances the probe's ability to guide THz radiation, resulting in superior imaging capabilities. The metallic coating further enhances performance by efficiently coupling with THz waves, leading to improved resolution and signal-to-noise ratios. Overall, this paper presents a thorough investigation of solid and hollow polymer-based THz near-field imaging probes and experimentally demonstrates their effectiveness for high-resolution sub-wavelength imaging applications. These 3D printable probes offer versatile, cost-effective, and disposable imaging solutions for non-destructive material evaluation and sub-cellular scale imaging in various domains.
Feature-based template approach for optimizing digital image correlation on complex deformationsPrasad, Sneha; Kumar, David; Kalra, Sumit; Khandelwal, Arpit
doi: 10.1117/12.3012491pmid: N/A
The initialization method in Digital Image Correlation (DIC) is essential for optimizing the correlation criteria and accurately computing the deformations of a material under load. At present, feature-based initialization techniques are widely explored for predicting the deformations of various complex circumstances, such as large deformations for soft materials, non-continuous deformations in heterogeneous materials, etc. However, due to the non-uniform distribution of the detected features, the initialization process goes through biased prediction. This bias occurs due to the sparsity of features in different regions of the sample, which can lead to inaccuracy in identifying the shape of deformation. This study addresses the issue of feature distribution and develops a feature-based template approach for providing initialization points for each subset on a finer scale. The features (interest points) are determined using KAZE feature detector and descriptor algorithm in nonlinear scale space due to its ability to determine consistent, repeatable, distinct features invariant to scale and rotation. The proposed algorithm uses bi-cubic b-spline interpolation to identify the strongest interest point at the subpixel level for each subset (of the input sample images), which works as an initial value for estimating the deformation. Further, a threshold-based incremental reference approach is developed for measuring large deformations and avoiding the cumulative errors associated with the commonly used incremental reference strategy, which is compute-intensive because of the comparison between every previous image and the subsequent images.
Distinctive trembling features following resonance peaks at zero group velocity frequencies in harmonic analysisLu, Runye; Shen, Yanfeng
doi: 10.1117/12.3010690pmid: N/A
The fascinating non-propagating lamb wave modes, Zero-Group-Velocity (ZGV) modes, have ignited profound research curiosity. ZGV modes possess the distinctive attribute of an elapsed group velocity with a finite nonzero wavenumber, indicating a spatially propagating wave package under a motionless envelope. This stationary mode engenders a localized resonance, confining the wave energy in the vicinity. These captivating phenomena have been scrutinized by researchers from the perspective of temporal and spatial domains. Nevertheless, it remains an uncharted frontier that how ZGV modes manifest their peculiarity for steady-state responses in harmonic analysis. Inspired by the unique trembling phenomenon following the appearance of ZGV resonance peaks, this paper aims at revealing the underlying mechanism and fundamental nature of the ZGV trembling phenomena in harmonic analysis, developing a deeper insight into lamb wave modes generation and propagation. The paper commences with the identification and extraction of ZGV modes under the frameworks of analytical analysis, serving as a reference for the subsequent analysis. This is followed by the construction of a finite element model for the implementation of harmonic analysis. Through the meticulous examination of displacement frequency spectra and dispersion curves, the trembling phenomenon following the ZGV resonances is visualized and evaluated. Ultimately, Electro-Mechanical Impedance Spectroscopy (EMIS) is employed to conduct the harmonic tests experimentally to validate the distinct trembling features. The distinct trembling features are attributed to the drastic fluctuations of participation factor for the emerging modes. This paper culminates with summary, concluding remarks, and suggestions for future work.
Exploratory investigation of early detection for high-C discharge-induced failure in 18650 lithium-ion batteriesAnthony, Goerge; Madden, Connor; Ogunniyi, Emmanuel; Downey, Austin R. J.; Limbaugh, Ryan; Peskar, Jarret; Bao, Jingjing; Huang, Xinyu
doi: 10.1117/12.3011097pmid: N/A
The surge in demand for high-energy-density lithium-ion batteries has led to the exploration of high-C (high current draw) discharges in various applications. However, these high-C discharges introduce significant challenges related to battery performance and safety. This exploratory study aims to investigate early current interrupt device failure detection mechanisms in 18650 lithium-ion batteries subjected to discharges up to 16C. Our controlled experimental setup induces a 40 amp discharge to a single lithium nickel cobalt aluminum oxide 18650 cell. Employing digital image correlation techniques, the structural changes in the battery are monitored during discharge, pinpointing subtle deformations and strain patterns as potential precursors to failure. This data, coupled with voltage and temperature measurements, offer a more comprehensive understanding of the battery performance under extreme conditions, allowing for future methods to further enhance safety protocols for high-C discharge.
From structure health monitoring to forensics: adapting computer vision to support victims of violenceAminfar, K.; Scafide, K.; Wojtusiak, J.; Lattanzi, D.
doi: 10.1117/12.2691482pmid: N/A
This paper explores the intersection between forensic science and Structural Health Monitoring (SHM), focusing on the pivotal role of visual indicators. These indicators are crucial in both contexts - from discerning injuries on humans to identifying structural defects. We present a novel approach utilizing computer vision-based diagnostics to aid victims of violence through advanced bruise detection, thus enhancing post-trauma care. Leveraging a specialized dataset, our study confronts the challenges inherent in data preparation and organization, as well as achieving expert consensus. We modify lightweight deep learning algorithms originally developed for engineered system diagnostics for application in the medical forensics domain. This adaptation aims to detect bruise areas under varying conditions, such as differences in skin color and lighting. A key question we address is the generalizability of these methods in diverse medical bruising scenarios, a fundamental challenge shared with SHM. Our research highlights the importance of domain knowledge transfer, drawing parallels between SHM and forensic science, and underscores the potential of this interdisciplinary approach.
An overview of geometric phases in elastic systems and their connection to topological invariants of elastic metamaterialsKumar, Mohit; Semperlotti, Fabio
doi: 10.1117/12.3010877pmid: N/A
The geometric phase is an additional phase factor acquired by oscillating dynamical systems. It has emerged as an insightful parameter to understand the dynamic behavior in a variety of systems, from molecular physics to elastic waveguides. In more recent years, the geometric phase has been widely exploited in connections with the analysis of topological materials. The present article reviews the concept of geometric phase in elastic systems and its connection to the design of elastic topological metamaterials. Examples are presented to explain the theoretical basis of the geometric phase by using arguments of differential geometry and topology. These concepts are then applied to the analysis of a one-dimensional elastic topological metamaterial that possesses localized vibration modes immune to geometric perturbations.
Modulation transfer technique for damage detection of structuresGorski, Jakub; Dziedziech, Kajetan; Klepka, Andrzej
doi: 10.1117/12.3015448pmid: N/A
The damage detection in structures using modulation transfer phenomena is a topic of increasing interest. However, the lack of comprehensive knowledge and established signal processing methods have hindered its widespread application. This paper explores the potential of the modulation transfer phenomenon for damage localization by conducting experiments on test stands with two structures: a damaged and an undamaged beam. A well-defined procedure for processing response signals and damage indicators was established. Before the experiments, modal analysis was conducted to select the appropriate excitation frequency. The presented results include spectra and trends of the damage indicators, demonstrating the viability of using the modulation transfer phenomenon for damage localization. Furthermore, the vibroacoustic modulation phenomenon was observed during the tests. These findings underscore the potential of modulation transfer techniques in structural health monitoring applications.
Comparison between a novel compressed sensing-based neural network and traditional neural network approaches for electrical impedance tomography reconstructionLi, Damond M.; Araiza, Marco G.; Wang, Long
doi: 10.1117/12.3010509pmid: N/A
Electrical Impedance Tomography (EIT) is a non-destructive and non-radioactive imaging technique used to detect anomalies in a material of interest. Applications of EIT range from medical imaging and early tumor detection to identifying structural damage. Within the past decade, deep learning (DL)-based EIT reconstruction has been an emerging field of study as it shows promise in addressing many of the challenges associated with the non-linear, ill-conditioned nature of EIT inverse problems. The DL-based approach allows for the conductivity of materials to be reconstructed directly through Neural Networks (NNs) as opposed to iteratively with conventional inverse reconstruction algorithms. So far, the reported DL-based NNs for EIT have mostly been trained by minimizing the Mean Squared Error (MSE) between the predicted and “true” outputs (i.e., conductivity distributions). The performance of these current NNs heavily relies on both the quality and quantity of training data. The NNs trained with simulated data may perform poorly with experimental data. On the other hand, generating sufficient experimental data NN training can be extremely expensive and time-consuming, if feasible at all. To advance the DL-based reconstruction for EIT, this study develops a novel NN architecture, trained with a custom loss function, that serves as a surrogate model for the compressed sensing-based EIT reconstruction algorithm. In other words, the NN is trained to mimic a compressed sensing algorithm that performs the EIT conductivity reconstruction. This approach enables the NN to accurately capture the electrical properties and characteristics of the sensing domain when trained with limited data of varying quality. The performance of the proposed NN was compared to other DL models trained with the traditional MSE loss function by evaluating their reconstruction resolution, accuracy, and other training metrics.